added all the files

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+fig/TPS_architecture.png filter=lfs diff=lfs merge=lfs -text
+results/results.xlsx filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,139 @@
+# Temporal Patch Shuffle (TPS): Leveraging Patch-Level Shuffling to Boost Generalization and Robustness in Time Series Forecasting [![arXiv](https://img.shields.io/badge/arXiv-Paper-B31B1B.svg)](https://arxiv.org/abs/2604.09067)
+
+
+## Abstract 
+Data augmentation is a crucial technique for improving model generalization and robustness, particularly in deep learning models where training data is limited. Although many augmentation methods have been developed for time series classification, most are not directly applicable to time series forecasting due to the need to preserve temporal coherence. In this work, we propose Temporal Patch Shuffle (TPS), a simple and model-agnostic data augmentation method for forecasting that extracts overlapping temporal patches, selectively shuffles a subset of patches using variance-based ordering as a conservative heuristic, and reconstructs the sequence by averaging overlapping regions. This design increases sample diversity while preserving forecast-consistent local temporal structure. We extensively evaluate TPS across nine long-term forecasting datasets using five recent model families (TSMixer, DLinear, PatchTST, TiDE, and LightTS), and across four short-term forecasting datasets using PatchTST, observing consistent performance improvements. Comprehensive ablation studies further demonstrate the effectiveness, robustness, and design rationale of the proposed method.
+
+## Key Contributions
+- **TPS (Temporal Patch Shuffle)**:
+  - Time Series Forecasting implementation: `time_series_forecasting/utils/augmentations.py`
+  - Univariate classification implementation (MiniRocket): `time_series_classification/minirocket/src/augmentation.py`
+  - Multivariate classification implementation (MultiRocket): `time_series_classification/MultiRocket/augmentation.py`
+
+![TPS Architecture](fig/TPS_architecture.png)
+
+## Repository Structure
+- `time_series_forecasting/`: forecasting models + augmentation pipeline (TPS and others)
+- `time_series_classification/`:
+  - `minirocket/`: univariate classification (MiniRocket) + augmentations (TPS and others)
+  - `MultiRocket/`: multivariate classification (MultiRocket) + augmentations (TPS and others)
+
+## Quick Start
+
+### 1) Time Series Forecasting
+
+#### Dataset
+Download all forecasting datasets from:
+- https://drive.google.com/drive/folders/1ZOYpTUa82_jCcxIdTmyr0LXQfvaM9vIy
+
+Create the dataset folder and put the downloaded files inside:
+```bash
+mkdir -p time_series_forecasting/dataset
+```
+
+#### Install & Run
+```bash
+git clone https://github.com/jafarbakhshaliyev/TPS.git
+cd TPS
+
+python3 -m venv .venv_forecasting
+source .venv_forecasting/bin/activate
+
+pip install -r time_series_forecasting/requirements.txt
+
+# Install PyTorch (choose the right command for your CUDA/CPU setup):
+# https://pytorch.org/get-started/locally/
+
+bash time_series_forecasting/scripts/example.sh
+```
+
+Notes:
+- `time_series_forecasting/scripts/example.sh` now runs relative to the repo
+- The training entrypoint is `time_series_forecasting/run_longExp.py`.
+
+### 2) Univariate Time Series Classification (MiniRocket)
+
+#### Dataset (UCR Archive)
+- https://www.cs.ucr.edu/~eamonn/time_series_data_2018/
+
+The MiniRocket code expects TSV files under a folder you set in `time_series_classification/minirocket/src/main.py` via `UCR_PATH`.
+
+#### Install & Run
+```bash
+cd TPS
+
+python3 -m venv .venv_minirocket
+source .venv_minirocket/bin/activate
+
+pip install numpy pandas scikit-learn tqdm
+
+# Run the provided script (SLURM `srun`):
+bash time_series_classification/minirocket/scripts/example.sh
+
+# If you're not on SLURM, run directly:
+cd time_series_classification/minirocket
+python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup
+```
+
+### 3) Multivariate Time Series Classification (MultiRocket)
+
+#### Dataset (UEA Multivariate Archive)
+- https://www.timeseriesclassification.com/index.php
+
+Place `.ts` files under (recommended):
+```bash
+mkdir -p time_series_classification/MultiRocket/data/Multivariate_ts
+```
+
+#### Install & Run
+MultiRocket has its own pinned dependencies in `time_series_classification/MultiRocket/requirements.txt`.
+
+```bash
+cd TPS
+
+python3 -m venv .venv_multirocket
+source .venv_multirocket/bin/activate
+
+pip install -r time_series_classification/MultiRocket/requirements.txt
+
+# The provided script uses SLURM `srun` and contains cluster-specific paths.
+# For a portable local run, prefer calling `main.py` directly:
+cd time_series_classification/MultiRocket
+python3 main.py --datapath ./data/Multivariate_ts --problem FaceDetection --iter 5 --verbose 1
+```
+
+## Augmentation Code (Direct Links)
+- Forecasting augmentations (TPS + others):
+  - [`time_series_forecasting/utils/augmentations.py`](time_series_forecasting/utils/augmentations.py)
+- MiniRocket augmentations (TPS + others):
+  - [`time_series_classification/minirocket/src/augmentation.py`](time_series_classification/minirocket/src/augmentation.py)
+- MultiRocket augmentations (TPS + others):
+  - [`time_series_classification/MultiRocket/augmentation.py`](time_series_classification/MultiRocket/augmentation.py)
+
+## Results
+
+Headline numbers from the tables:
+- Long-term forecasting (9 datasets × 4 horizons): TPS is rank-1 in most settings (e.g., **DLinear**: 35/36 wins for MSE; 34/36 wins for MAE).
+- Short-term traffic forecasting (PeMS03/04/07/08 with PatchTST): TPS wins on most metrics (e.g., PeMS03: MSE/MAE **0.104/0.216**).
+- Classification (mean ± std): TPS improves both univariate MiniRocket (**0.804 ± 0.0098**) and multivariate MultiRocket (**0.643 ± 0.0253**).
+
+## Citation
+If you find this repository useful, please cite our paper:
+
+```bibtex
+@misc{bakhshaliyev2026temporalpatchshuffletps,
+      title={Temporal Patch Shuffle (TPS): Leveraging Patch-Level Shuffling to Boost Generalization and Robustness in Time Series Forecasting}, 
+      author={Jafar Bakhshaliyev and Johannes Burchert and Niels Landwehr and Lars Schmidt-Thieme},
+      year={2026},
+      eprint={2604.09067},
+      archivePrefix={arXiv},
+      primaryClass={cs.LG},
+      url={https://arxiv.org/abs/2604.09067}, 
+}
+```
+
+## Acknowledgements
+- Forecasting code credits: see headers in `time_series_forecasting/` files and references in `time_series_forecasting/`.
+- MiniRocket is modified from the official implementation (see `time_series_classification/minirocket/LICENSE`).
+- MultiRocket is modified from the official implementation (see `time_series_classification/MultiRocket/LICENSE`).
+- Third-party attributions: `THIRD_PARTY_LICENSES.txt` inside each classification submodule.

results/results.xlsx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5a6ceaa68aaeb6e98fe43aae4018cb53f7349d297290ae93c7424f08de9c199
+size 102945

time_series_classification/MultiRocket/LICENSE

@@ -0,0 +1,692 @@
+ This project is licensed under the GNU General Public License v3.0 (GPL-3.0)
+
+Original Authors:
+  - Chang Wei Tan
+  - Angus Dempster
+  - Christoph Bergmeir
+  - Geoffrey I. Webb
+  Original repository: https://github.com/ChangWeiTan/MultiRocket
+
+Modifications by:
+  - Jafar Bakhshaliyev, 2025
+
+This project includes modifications to the original MultiRocket codebase.
+All changes are released under the same GPL-3.0 license.
+
+-----------------------------------------------------------------------
+ 
+ 
+                    GNU GENERAL PUBLIC LICENSE
+                       Version 3, 29 June 2007
+
+ Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The GNU General Public License is a free, copyleft license for
+software and other kinds of works.
+
+  The licenses for most software and other practical works are designed
+to take away your freedom to share and change the works.  By contrast,
+the GNU General Public License is intended to guarantee your freedom to
+share and change all versions of a program--to make sure it remains free
+software for all its users.  We, the Free Software Foundation, use the
+GNU General Public License for most of our software; it applies also to
+any other work released this way by its authors.  You can apply it to
+your programs, too.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+them if you wish), that you receive source code or can get it if you
+want it, that you can change the software or use pieces of it in new
+free programs, and that you know you can do these things.
+
+  To protect your rights, we need to prevent others from denying you
+these rights or asking you to surrender the rights.  Therefore, you have
+certain responsibilities if you distribute copies of the software, or if
+you modify it: responsibilities to respect the freedom of others.
+
+  For example, if you distribute copies of such a program, whether
+gratis or for a fee, you must pass on to the recipients the same
+freedoms that you received.  You must make sure that they, too, receive
+or can get the source code.  And you must show them these terms so they
+know their rights.
+
+  Developers that use the GNU GPL protect your rights with two steps:
+(1) assert copyright on the software, and (2) offer you this License
+giving you legal permission to copy, distribute and/or modify it.
+
+  For the developers' and authors' protection, the GPL clearly explains
+that there is no warranty for this free software.  For both users' and
+authors' sake, the GPL requires that modified versions be marked as
+changed, so that their problems will not be attributed erroneously to
+authors of previous versions.
+
+  Some devices are designed to deny users access to install or run
+modified versions of the software inside them, although the manufacturer
+can do so.  This is fundamentally incompatible with the aim of
+protecting users' freedom to change the software.  The systematic
+pattern of such abuse occurs in the area of products for individuals to
+use, which is precisely where it is most unacceptable.  Therefore, we
+have designed this version of the GPL to prohibit the practice for those
+products.  If such problems arise substantially in other domains, we
+stand ready to extend this provision to those domains in future versions
+of the GPL, as needed to protect the freedom of users.
+
+  Finally, every program is threatened constantly by software patents.
+States should not allow patents to restrict development and use of
+software on general-purpose computers, but in those that do, we wish to
+avoid the special danger that patents applied to a free program could
+make it effectively proprietary.  To prevent this, the GPL assures that
+patents cannot be used to render the program non-free.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                       TERMS AND CONDITIONS
+
+  0. Definitions.
+
+  "This License" refers to version 3 of the GNU General Public License.
+
+  "Copyright" also means copyright-like laws that apply to other kinds of
+works, such as semiconductor masks.
+
+  "The Program" refers to any copyrightable work licensed under this
+License.  Each licensee is addressed as "you".  "Licensees" and
+"recipients" may be individuals or organizations.
+
+  To "modify" a work means to copy from or adapt all or part of the work
+in a fashion requiring copyright permission, other than the making of an
+exact copy.  The resulting work is called a "modified version" of the
+earlier work or a work "based on" the earlier work.
+
+  A "covered work" means either the unmodified Program or a work based
+on the Program.
+
+  To "propagate" a work means to do anything with it that, without
+permission, would make you directly or secondarily liable for
+infringement under applicable copyright law, except executing it on a
+computer or modifying a private copy.  Propagation includes copying,
+distribution (with or without modification), making available to the
+public, and in some countries other activities as well.
+
+  To "convey" a work means any kind of propagation that enables other
+parties to make or receive copies.  Mere interaction with a user through
+a computer network, with no transfer of a copy, is not conveying.
+
+  An interactive user interface displays "Appropriate Legal Notices"
+to the extent that it includes a convenient and prominently visible
+feature that (1) displays an appropriate copyright notice, and (2)
+tells the user that there is no warranty for the work (except to the
+extent that warranties are provided), that licensees may convey the
+work under this License, and how to view a copy of this License.  If
+the interface presents a list of user commands or options, such as a
+menu, a prominent item in the list meets this criterion.
+
+  1. Source Code.
+
+  The "source code" for a work means the preferred form of the work
+for making modifications to it.  "Object code" means any non-source
+form of a work.
+
+  A "Standard Interface" means an interface that either is an official
+standard defined by a recognized standards body, or, in the case of
+interfaces specified for a particular programming language, one that
+is widely used among developers working in that language.
+
+  The "System Libraries" of an executable work include anything, other
+than the work as a whole, that (a) is included in the normal form of
+packaging a Major Component, but which is not part of that Major
+Component, and (b) serves only to enable use of the work with that
+Major Component, or to implement a Standard Interface for which an
+implementation is available to the public in source code form.  A
+"Major Component", in this context, means a major essential component
+(kernel, window system, and so on) of the specific operating system
+(if any) on which the executable work runs, or a compiler used to
+produce the work, or an object code interpreter used to run it.
+
+  The "Corresponding Source" for a work in object code form means all
+the source code needed to generate, install, and (for an executable
+work) run the object code and to modify the work, including scripts to
+control those activities.  However, it does not include the work's
+System Libraries, or general-purpose tools or generally available free
+programs which are used unmodified in performing those activities but
+which are not part of the work.  For example, Corresponding Source
+includes interface definition files associated with source files for
+the work, and the source code for shared libraries and dynamically
+linked subprograms that the work is specifically designed to require,
+such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+  The Corresponding Source need not include anything that users
+can regenerate automatically from other parts of the Corresponding
+Source.
+
+  The Corresponding Source for a work in source code form is that
+same work.
+
+  2. Basic Permissions.
+
+  All rights granted under this License are granted for the term of
+copyright on the Program, and are irrevocable provided the stated
+conditions are met.  This License explicitly affirms your unlimited
+permission to run the unmodified Program.  The output from running a
+covered work is covered by this License only if the output, given its
+content, constitutes a covered work.  This License acknowledges your
+rights of fair use or other equivalent, as provided by copyright law.
+
+  You may make, run and propagate covered works that you do not
+convey, without conditions so long as your license otherwise remains
+in force.  You may convey covered works to others for the sole purpose
+of having them make modifications exclusively for you, or provide you
+with facilities for running those works, provided that you comply with
+the terms of this License in conveying all material for which you do
+not control copyright.  Those thus making or running the covered works
+for you must do so exclusively on your behalf, under your direction
+and control, on terms that prohibit them from making any copies of
+your copyrighted material outside their relationship with you.
+
+  Conveying under any other circumstances is permitted solely under
+the conditions stated below.  Sublicensing is not allowed; section 10
+makes it unnecessary.
+
+  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+
+  No covered work shall be deemed part of an effective technological
+measure under any applicable law fulfilling obligations under article
+11 of the WIPO copyright treaty adopted on 20 December 1996, or
+similar laws prohibiting or restricting circumvention of such
+measures.
+
+  When you convey a covered work, you waive any legal power to forbid
+circumvention of technological measures to the extent such circumvention
+is effected by exercising rights under this License with respect to
+the covered work, and you disclaim any intention to limit operation or
+modification of the work as a means of enforcing, against the work's
+users, your or third parties' legal rights to forbid circumvention of
+technological measures.
+
+  4. Conveying Verbatim Copies.
+
+  You may convey verbatim copies of the Program's source code as you
+receive it, in any medium, provided that you conspicuously and
+appropriately publish on each copy an appropriate copyright notice;
+keep intact all notices stating that this License and any
+non-permissive terms added in accord with section 7 apply to the code;
+keep intact all notices of the absence of any warranty; and give all
+recipients a copy of this License along with the Program.
+
+  You may charge any price or no price for each copy that you convey,
+and you may offer support or warranty protection for a fee.
+
+  5. Conveying Modified Source Versions.
+
+  You may convey a work based on the Program, or the modifications to
+produce it from the Program, in the form of source code under the
+terms of section 4, provided that you also meet all of these conditions:
+
+    a) The work must carry prominent notices stating that you modified
+    it, and giving a relevant date.
+
+    b) The work must carry prominent notices stating that it is
+    released under this License and any conditions added under section
+    7.  This requirement modifies the requirement in section 4 to
+    "keep intact all notices".
+
+    c) You must license the entire work, as a whole, under this
+    License to anyone who comes into possession of a copy.  This
+    License will therefore apply, along with any applicable section 7
+    additional terms, to the whole of the work, and all its parts,
+    regardless of how they are packaged.  This License gives no
+    permission to license the work in any other way, but it does not
+    invalidate such permission if you have separately received it.
+
+    d) If the work has interactive user interfaces, each must display
+    Appropriate Legal Notices; however, if the Program has interactive
+    interfaces that do not display Appropriate Legal Notices, your
+    work need not make them do so.
+
+  A compilation of a covered work with other separate and independent
+works, which are not by their nature extensions of the covered work,
+and which are not combined with it such as to form a larger program,
+in or on a volume of a storage or distribution medium, is called an
+"aggregate" if the compilation and its resulting copyright are not
+used to limit the access or legal rights of the compilation's users
+beyond what the individual works permit.  Inclusion of a covered work
+in an aggregate does not cause this License to apply to the other
+parts of the aggregate.
+
+  6. Conveying Non-Source Forms.
+
+  You may convey a covered work in object code form under the terms
+of sections 4 and 5, provided that you also convey the
+machine-readable Corresponding Source under the terms of this License,
+in one of these ways:
+
+    a) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by the
+    Corresponding Source fixed on a durable physical medium
+    customarily used for software interchange.
+
+    b) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by a
+    written offer, valid for at least three years and valid for as
+    long as you offer spare parts or customer support for that product
+    model, to give anyone who possesses the object code either (1) a
+    copy of the Corresponding Source for all the software in the
+    product that is covered by this License, on a durable physical
+    medium customarily used for software interchange, for a price no
+    more than your reasonable cost of physically performing this
+    conveying of source, or (2) access to copy the
+    Corresponding Source from a network server at no charge.
+
+    c) Convey individual copies of the object code with a copy of the
+    written offer to provide the Corresponding Source.  This
+    alternative is allowed only occasionally and noncommercially, and
+    only if you received the object code with such an offer, in accord
+    with subsection 6b.
+
+    d) Convey the object code by offering access from a designated
+    place (gratis or for a charge), and offer equivalent access to the
+    Corresponding Source in the same way through the same place at no
+    further charge.  You need not require recipients to copy the
+    Corresponding Source along with the object code.  If the place to
+    copy the object code is a network server, the Corresponding Source
+    may be on a different server (operated by you or a third party)
+    that supports equivalent copying facilities, provided you maintain
+    clear directions next to the object code saying where to find the
+    Corresponding Source.  Regardless of what server hosts the
+    Corresponding Source, you remain obligated to ensure that it is
+    available for as long as needed to satisfy these requirements.
+
+    e) Convey the object code using peer-to-peer transmission, provided
+    you inform other peers where the object code and Corresponding
+    Source of the work are being offered to the general public at no
+    charge under subsection 6d.
+
+  A separable portion of the object code, whose source code is excluded
+from the Corresponding Source as a System Library, need not be
+included in conveying the object code work.
+
+  A "User Product" is either (1) a "consumer product", which means any
+tangible personal property which is normally used for personal, family,
+or household purposes, or (2) anything designed or sold for incorporation
+into a dwelling.  In determining whether a product is a consumer product,
+doubtful cases shall be resolved in favor of coverage.  For a particular
+product received by a particular user, "normally used" refers to a
+typical or common use of that class of product, regardless of the status
+of the particular user or of the way in which the particular user
+actually uses, or expects or is expected to use, the product.  A product
+is a consumer product regardless of whether the product has substantial
+commercial, industrial or non-consumer uses, unless such uses represent
+the only significant mode of use of the product.
+
+  "Installation Information" for a User Product means any methods,
+procedures, authorization keys, or other information required to install
+and execute modified versions of a covered work in that User Product from
+a modified version of its Corresponding Source.  The information must
+suffice to ensure that the continued functioning of the modified object
+code is in no case prevented or interfered with solely because
+modification has been made.
+
+  If you convey an object code work under this section in, or with, or
+specifically for use in, a User Product, and the conveying occurs as
+part of a transaction in which the right of possession and use of the
+User Product is transferred to the recipient in perpetuity or for a
+fixed term (regardless of how the transaction is characterized), the
+Corresponding Source conveyed under this section must be accompanied
+by the Installation Information.  But this requirement does not apply
+if neither you nor any third party retains the ability to install
+modified object code on the User Product (for example, the work has
+been installed in ROM).
+
+  The requirement to provide Installation Information does not include a
+requirement to continue to provide support service, warranty, or updates
+for a work that has been modified or installed by the recipient, or for
+the User Product in which it has been modified or installed.  Access to a
+network may be denied when the modification itself materially and
+adversely affects the operation of the network or violates the rules and
+protocols for communication across the network.
+
+  Corresponding Source conveyed, and Installation Information provided,
+in accord with this section must be in a format that is publicly
+documented (and with an implementation available to the public in
+source code form), and must require no special password or key for
+unpacking, reading or copying.
+
+  7. Additional Terms.
+
+  "Additional permissions" are terms that supplement the terms of this
+License by making exceptions from one or more of its conditions.
+Additional permissions that are applicable to the entire Program shall
+be treated as though they were included in this License, to the extent
+that they are valid under applicable law.  If additional permissions
+apply only to part of the Program, that part may be used separately
+under those permissions, but the entire Program remains governed by
+this License without regard to the additional permissions.
+
+  When you convey a copy of a covered work, you may at your option
+remove any additional permissions from that copy, or from any part of
+it.  (Additional permissions may be written to require their own
+removal in certain cases when you modify the work.)  You may place
+additional permissions on material, added by you to a covered work,
+for which you have or can give appropriate copyright permission.
+
+  Notwithstanding any other provision of this License, for material you
+add to a covered work, you may (if authorized by the copyright holders of
+that material) supplement the terms of this License with terms:
+
+    a) Disclaiming warranty or limiting liability differently from the
+    terms of sections 15 and 16 of this License; or
+
+    b) Requiring preservation of specified reasonable legal notices or
+    author attributions in that material or in the Appropriate Legal
+    Notices displayed by works containing it; or
+
+    c) Prohibiting misrepresentation of the origin of that material, or
+    requiring that modified versions of such material be marked in
+    reasonable ways as different from the original version; or
+
+    d) Limiting the use for publicity purposes of names of licensors or
+    authors of the material; or
+
+    e) Declining to grant rights under trademark law for use of some
+    trade names, trademarks, or service marks; or
+
+    f) Requiring indemnification of licensors and authors of that
+    material by anyone who conveys the material (or modified versions of
+    it) with contractual assumptions of liability to the recipient, for
+    any liability that these contractual assumptions directly impose on
+    those licensors and authors.
+
+  All other non-permissive additional terms are considered "further
+restrictions" within the meaning of section 10.  If the Program as you
+received it, or any part of it, contains a notice stating that it is
+governed by this License along with a term that is a further
+restriction, you may remove that term.  If a license document contains
+a further restriction but permits relicensing or conveying under this
+License, you may add to a covered work material governed by the terms
+of that license document, provided that the further restriction does
+not survive such relicensing or conveying.
+
+  If you add terms to a covered work in accord with this section, you
+must place, in the relevant source files, a statement of the
+additional terms that apply to those files, or a notice indicating
+where to find the applicable terms.
+
+  Additional terms, permissive or non-permissive, may be stated in the
+form of a separately written license, or stated as exceptions;
+the above requirements apply either way.
+
+  8. Termination.
+
+  You may not propagate or modify a covered work except as expressly
+provided under this License.  Any attempt otherwise to propagate or
+modify it is void, and will automatically terminate your rights under
+this License (including any patent licenses granted under the third
+paragraph of section 11).
+
+  However, if you cease all violation of this License, then your
+license from a particular copyright holder is reinstated (a)
+provisionally, unless and until the copyright holder explicitly and
+finally terminates your license, and (b) permanently, if the copyright
+holder fails to notify you of the violation by some reasonable means
+prior to 60 days after the cessation.
+
+  Moreover, your license from a particular copyright holder is
+reinstated permanently if the copyright holder notifies you of the
+violation by some reasonable means, this is the first time you have
+received notice of violation of this License (for any work) from that
+copyright holder, and you cure the violation prior to 30 days after
+your receipt of the notice.
+
+  Termination of your rights under this section does not terminate the
+licenses of parties who have received copies or rights from you under
+this License.  If your rights have been terminated and not permanently
+reinstated, you do not qualify to receive new licenses for the same
+material under section 10.
+
+  9. Acceptance Not Required for Having Copies.
+
+  You are not required to accept this License in order to receive or
+run a copy of the Program.  Ancillary propagation of a covered work
+occurring solely as a consequence of using peer-to-peer transmission
+to receive a copy likewise does not require acceptance.  However,
+nothing other than this License grants you permission to propagate or
+modify any covered work.  These actions infringe copyright if you do
+not accept this License.  Therefore, by modifying or propagating a
+covered work, you indicate your acceptance of this License to do so.
+
+  10. Automatic Licensing of Downstream Recipients.
+
+  Each time you convey a covered work, the recipient automatically
+receives a license from the original licensors, to run, modify and
+propagate that work, subject to this License.  You are not responsible
+for enforcing compliance by third parties with this License.
+
+  An "entity transaction" is a transaction transferring control of an
+organization, or substantially all assets of one, or subdividing an
+organization, or merging organizations.  If propagation of a covered
+work results from an entity transaction, each party to that
+transaction who receives a copy of the work also receives whatever
+licenses to the work the party's predecessor in interest had or could
+give under the previous paragraph, plus a right to possession of the
+Corresponding Source of the work from the predecessor in interest, if
+the predecessor has it or can get it with reasonable efforts.
+
+  You may not impose any further restrictions on the exercise of the
+rights granted or affirmed under this License.  For example, you may
+not impose a license fee, royalty, or other charge for exercise of
+rights granted under this License, and you may not initiate litigation
+(including a cross-claim or counterclaim in a lawsuit) alleging that
+any patent claim is infringed by making, using, selling, offering for
+sale, or importing the Program or any portion of it.
+
+  11. Patents.
+
+  A "contributor" is a copyright holder who authorizes use under this
+License of the Program or a work on which the Program is based.  The
+work thus licensed is called the contributor's "contributor version".
+
+  A contributor's "essential patent claims" are all patent claims
+owned or controlled by the contributor, whether already acquired or
+hereafter acquired, that would be infringed by some manner, permitted
+by this License, of making, using, or selling its contributor version,
+but do not include claims that would be infringed only as a
+consequence of further modification of the contributor version.  For
+purposes of this definition, "control" includes the right to grant
+patent sublicenses in a manner consistent with the requirements of
+this License.
+
+  Each contributor grants you a non-exclusive, worldwide, royalty-free
+patent license under the contributor's essential patent claims, to
+make, use, sell, offer for sale, import and otherwise run, modify and
+propagate the contents of its contributor version.
+
+  In the following three paragraphs, a "patent license" is any express
+agreement or commitment, however denominated, not to enforce a patent
+(such as an express permission to practice a patent or covenant not to
+sue for patent infringement).  To "grant" such a patent license to a
+party means to make such an agreement or commitment not to enforce a
+patent against the party.
+
+  If you convey a covered work, knowingly relying on a patent license,
+and the Corresponding Source of the work is not available for anyone
+to copy, free of charge and under the terms of this License, through a
+publicly available network server or other readily accessible means,
+then you must either (1) cause the Corresponding Source to be so
+available, or (2) arrange to deprive yourself of the benefit of the
+patent license for this particular work, or (3) arrange, in a manner
+consistent with the requirements of this License, to extend the patent
+license to downstream recipients.  "Knowingly relying" means you have
+actual knowledge that, but for the patent license, your conveying the
+covered work in a country, or your recipient's use of the covered work
+in a country, would infringe one or more identifiable patents in that
+country that you have reason to believe are valid.
+
+  If, pursuant to or in connection with a single transaction or
+arrangement, you convey, or propagate by procuring conveyance of, a
+covered work, and grant a patent license to some of the parties
+receiving the covered work authorizing them to use, propagate, modify
+or convey a specific copy of the covered work, then the patent license
+you grant is automatically extended to all recipients of the covered
+work and works based on it.
+
+  A patent license is "discriminatory" if it does not include within
+the scope of its coverage, prohibits the exercise of, or is
+conditioned on the non-exercise of one or more of the rights that are
+specifically granted under this License.  You may not convey a covered
+work if you are a party to an arrangement with a third party that is
+in the business of distributing software, under which you make payment
+to the third party based on the extent of your activity of conveying
+the work, and under which the third party grants, to any of the
+parties who would receive the covered work from you, a discriminatory
+patent license (a) in connection with copies of the covered work
+conveyed by you (or copies made from those copies), or (b) primarily
+for and in connection with specific products or compilations that
+contain the covered work, unless you entered into that arrangement,
+or that patent license was granted, prior to 28 March 2007.
+
+  Nothing in this License shall be construed as excluding or limiting
+any implied license or other defenses to infringement that may
+otherwise be available to you under applicable patent law.
+
+  12. No Surrender of Others' Freedom.
+
+  If conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot convey a
+covered work so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you may
+not convey it at all.  For example, if you agree to terms that obligate you
+to collect a royalty for further conveying from those to whom you convey
+the Program, the only way you could satisfy both those terms and this
+License would be to refrain entirely from conveying the Program.
+
+  13. Use with the GNU Affero General Public License.
+
+  Notwithstanding any other provision of this License, you have
+permission to link or combine any covered work with a work licensed
+under version 3 of the GNU Affero General Public License into a single
+combined work, and to convey the resulting work.  The terms of this
+License will continue to apply to the part which is the covered work,
+but the special requirements of the GNU Affero General Public License,
+section 13, concerning interaction through a network will apply to the
+combination as such.
+
+  14. Revised Versions of this License.
+
+  The Free Software Foundation may publish revised and/or new versions of
+the GNU General Public License from time to time.  Such new versions will
+be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+  Each version is given a distinguishing version number.  If the
+Program specifies that a certain numbered version of the GNU General
+Public License "or any later version" applies to it, you have the
+option of following the terms and conditions either of that numbered
+version or of any later version published by the Free Software
+Foundation.  If the Program does not specify a version number of the
+GNU General Public License, you may choose any version ever published
+by the Free Software Foundation.
+
+  If the Program specifies that a proxy can decide which future
+versions of the GNU General Public License can be used, that proxy's
+public statement of acceptance of a version permanently authorizes you
+to choose that version for the Program.
+
+  Later license versions may give you additional or different
+permissions.  However, no additional obligations are imposed on any
+author or copyright holder as a result of your choosing to follow a
+later version.
+
+  15. Disclaimer of Warranty.
+
+  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
+APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
+OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
+IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
+ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+  16. Limitation of Liability.
+
+  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
+THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
+GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
+USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
+DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
+PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
+EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
+SUCH DAMAGES.
+
+  17. Interpretation of Sections 15 and 16.
+
+  If the disclaimer of warranty and limitation of liability provided
+above cannot be given local legal effect according to their terms,
+reviewing courts shall apply local law that most closely approximates
+an absolute waiver of all civil liability in connection with the
+Program, unless a warranty or assumption of liability accompanies a
+copy of the Program in return for a fee.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+state the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software: you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation, either version 3 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+Also add information on how to contact you by electronic and paper mail.
+
+  If the program does terminal interaction, make it output a short
+notice like this when it starts in an interactive mode:
+
+    <program>  Copyright (C) <year>  <name of author>
+    This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
+    This is free software, and you are welcome to redistribute it
+    under certain conditions; type `show c' for details.
+
+The hypothetical commands `show w' and `show c' should show the appropriate
+parts of the General Public License.  Of course, your program's commands
+might be different; for a GUI interface, you would use an "about box".
+
+  You should also get your employer (if you work as a programmer) or school,
+if any, to sign a "copyright disclaimer" for the program, if necessary.
+For more information on this, and how to apply and follow the GNU GPL, see
+<https://www.gnu.org/licenses/>.
+
+  The GNU General Public License does not permit incorporating your program
+into proprietary programs.  If your program is a subroutine library, you
+may consider it more useful to permit linking proprietary applications with
+the library.  If this is what you want to do, use the GNU Lesser General
+Public License instead of this License.  But first, please read
+<https://www.gnu.org/licenses/why-not-lgpl.html>.

time_series_classification/MultiRocket/THIRD_PARTY_LICENSES.txt

@@ -0,0 +1,221 @@
+THIRD PARTY LICENSES
+
+This file contains the licenses for third-party components used in this project.
+
+================================================================================
+
+Time Series Augmentation Components
+
+Some augmentation methods in this repository are derived from or inspired by the 
+time_series_augmentation repository:
+
+Source: https://github.com/uchidalab/time_series_augmentation
+Authors: Brian Kenji Iwana and Seiichi Uchida
+License: Apache License 2.0
+Files affected: ./time_series_classification/MultiRocket/augmentation.py
+
+================================================================================
+
+
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

time_series_classification/MultiRocket/augmentation.py

@@ -0,0 +1,543 @@
+# Adapted from time_series_augmentation (https://github.com/uchidalab/time_series_augmentation)
+# Original: Apache License 2.0 by Brian Kenji Iwana and Seiichi Uchida
+# Modified by Jafar Bakhshaliyev (2025) - Licensed under GPL v3.0
+
+import numpy as np
+from tqdm import tqdm
+
+
+
+
+def tps(x, y, patch_len=0, stride=0, shuffle_rate=0.0):
+    """
+    Temporal Patch Shuffle (TPS) augmentation for time series classification.
+    
+    Parameters:
+    -----------
+    x : numpy.ndarray
+        Input time series data of shape (n_samples, timesteps, n_features)
+    patch_len : int
+        Length of each patch
+    stride : int
+        Stride between patches
+    shuffle_rate : float
+        Proportion of patches to shuffle (between 0 and 1)
+        
+    Returns:
+    --------
+    numpy.ndarray
+        Augmented time series data with same shape as input
+    """
+    n_samples, T, n_features = x.shape
+        
+    ret = np.zeros_like(x)
+    
+    # Calculate required padding to avoid zeros at the end
+    total_patches = (T - patch_len + stride - 1) // stride + 1
+    total_len = (total_patches - 1) * stride + patch_len
+    padding_needed = total_len - T
+    
+    # Process each sample
+    for i in range(n_samples):
+        # current sample
+        sample = x[i]  # shape: (timesteps, n_features)
+        
+        # Apply padding if needed
+        if padding_needed > 0:
+            padded_sample = np.pad(sample, ((0, padding_needed), (0, 0)), mode='edge')
+            T_padded = T + padding_needed
+        else:
+            padded_sample = sample
+            T_padded = T
+            
+        num_patches = ((T_padded - patch_len) // stride) + 1
+        
+        # Create patches for current sample
+        patches = np.zeros((num_patches, patch_len, n_features))
+        for j in range(num_patches):
+            start = j * stride
+            patches[j] = padded_sample[start:start + patch_len]
+            
+        # importance of each patch 
+        importance_scores = np.var(patches, axis=(1, 2))
+        
+        # number of patches to shuffle
+        num_to_shuffle = int(num_patches * shuffle_rate)
+        
+        if num_to_shuffle > 0:
+            # indices of least important patches
+            shuffle_indices = np.argsort(importance_scores)[:num_to_shuffle]
+            
+            # Shuffle these patches among themselves
+            patches_to_shuffle = patches[shuffle_indices].copy()
+            shuffled_order = np.random.permutation(num_to_shuffle)
+            
+            for idx, new_idx in enumerate(shuffled_order):
+                patch_idx = shuffle_indices[idx]
+                new_patch = patches_to_shuffle[new_idx]
+                patches[patch_idx] = new_patch
+            
+        # Reconstruct the time series
+        reconstructed = np.zeros((T_padded, n_features))
+        counts = np.zeros((T_padded, n_features))
+        
+        for j in range(num_patches):
+            start = j * stride
+            end = start + patch_len
+            reconstructed[start:end] += patches[j]
+            counts[start:end] += 1
+            
+        # Average overlapping patches and handle potential zeros
+        # Use a mask to identify zero counts
+        mask = counts == 0
+        if np.any(mask):
+            # Fill in zeros with nearest non-zero values
+            for feat in range(n_features):
+                feat_mask = mask[:, feat]
+                if np.any(feat_mask):
+                    # Get indices of zero and non-zero values
+                    zero_indices = np.where(feat_mask)[0]
+                    nonzero_indices = np.where(~feat_mask)[0]
+                    
+                    if len(nonzero_indices) > 0:
+                        # Find nearest non-zero index for each zero index
+                        for zero_idx in zero_indices:
+                            nearest_idx = nonzero_indices[np.argmin(np.abs(nonzero_indices - zero_idx))]
+                            reconstructed[zero_idx, feat] = reconstructed[nearest_idx, feat]
+                            counts[zero_idx, feat] = 1
+        
+        # Avoid division by zero
+        counts[counts == 0] = 1
+        reconstructed = reconstructed / counts
+        
+        # Remove padding 
+        ret[i] = reconstructed[:T]
+        
+    return ret
+
+def jitter(x, sigma=0.03):
+    # https://arxiv.org/pdf/1706.00527.pdf
+    return x + np.random.normal(loc=0., scale=sigma, size=x.shape)
+
+def scaling(x, sigma=0.1):
+    # https://arxiv.org/pdf/1706.00527.pdf
+    factor = np.random.normal(loc=1., scale=sigma, size=(x.shape[0],x.shape[2]))
+    return np.multiply(x, factor[:,np.newaxis,:])
+
+def rotation(x):
+    flip = np.random.choice([-1, 1], size=(x.shape[0],x.shape[2]))
+    rotate_axis = np.arange(x.shape[2])
+    np.random.shuffle(rotate_axis)    
+    return flip[:,np.newaxis,:] * x[:,:,rotate_axis]
+
+
+
+def magnitude_warp(x, sigma=0.2, knot=4):
+    from scipy.interpolate import CubicSpline
+    orig_steps = np.arange(x.shape[1])
+    
+    random_warps = np.random.normal(loc=1.0, scale=sigma, size=(x.shape[0], knot+2, x.shape[2]))
+    warp_steps = (np.ones((x.shape[2],1))*(np.linspace(0, x.shape[1]-1., num=knot+2))).T
+    ret = np.zeros_like(x)
+    for i, pat in enumerate(x):
+        warper = np.array([CubicSpline(warp_steps[:,dim], random_warps[i,:,dim])(orig_steps) for dim in range(x.shape[2])]).T
+        ret[i] = pat * warper
+
+    return ret
+
+def time_warp(x, sigma=0.2, knot=4):
+    from scipy.interpolate import CubicSpline
+    orig_steps = np.arange(x.shape[1])
+    
+    random_warps = np.random.normal(loc=1.0, scale=sigma, size=(x.shape[0], knot+2, x.shape[2]))
+    warp_steps = (np.ones((x.shape[2],1))*(np.linspace(0, x.shape[1]-1., num=knot+2))).T
+    
+    ret = np.zeros_like(x)
+    for i, pat in enumerate(x):
+        for dim in range(x.shape[2]):
+            time_warp = CubicSpline(warp_steps[:,dim], warp_steps[:,dim] * random_warps[i,:,dim])(orig_steps)
+            scale = (x.shape[1]-1)/time_warp[-1]
+            ret[i,:,dim] = np.interp(orig_steps, np.clip(scale*time_warp, 0, x.shape[1]-1), pat[:,dim]).T
+    return ret
+
+def window_slice(x, reduce_ratio=0.9):
+    # https://halshs.archives-ouvertes.fr/halshs-01357973/document
+    target_len = np.ceil(reduce_ratio*x.shape[1]).astype(int)
+    if target_len >= x.shape[1]:
+        return x
+    starts = np.random.randint(low=0, high=x.shape[1]-target_len, size=(x.shape[0])).astype(int)
+    ends = (target_len + starts).astype(int)
+    
+    ret = np.zeros_like(x)
+    for i, pat in enumerate(x):
+        for dim in range(x.shape[2]):
+            ret[i,:,dim] = np.interp(np.linspace(0, target_len, num=x.shape[1]), np.arange(target_len), pat[starts[i]:ends[i],dim]).T
+    return ret
+
+def permutation(x, max_segments=5, seg_mode="equal"):
+    orig_steps = np.arange(x.shape[1])
+    num_segs = np.random.randint(1, max_segments, size=(x.shape[0]))
+
+    ret = np.zeros_like(x)
+    for i, pat in enumerate(x):
+        if num_segs[i] > 1:
+            if seg_mode == "random":
+                # Fix: Check if we have enough points to sample from
+                available_points = x.shape[1] - 2
+                needed_points = num_segs[i] - 1
+                
+                # Ensure we have enough points to sample and adjust if necessary
+                if available_points <= 0:
+                    # Not enough points for random splitting, fallback to equal segments
+                    splits = np.array_split(orig_steps, num_segs[i])
+                elif needed_points > available_points:
+                    # Too many segments requested, adjust number of segments
+                    actual_segs = min(available_points + 1, num_segs[i])
+                    splits = np.array_split(orig_steps, actual_segs)
+                else:
+                    # Original logic can work
+                    split_points = np.random.choice(available_points, needed_points, replace=False)
+                    split_points.sort()
+                    splits = np.split(orig_steps, split_points)
+            else:
+                splits = np.array_split(orig_steps, num_segs[i])
+
+            # Only permute if we have more than one segment
+            if len(splits) > 1:
+                perm = np.random.permutation(len(splits))
+                warp = np.concatenate([splits[j] for j in perm]).ravel()
+                ret[i] = pat[warp]
+            else:
+                ret[i] = pat
+        else:
+            ret[i] = pat
+    return ret
+
+# Fixed window_warp function
+def window_warp(x, window_ratio=0.1, scales=[0.5, 2.]):
+    # https://halshs.archives-ouvertes.fr/halshs-01357973/document
+    warp_scales = np.random.choice(scales, x.shape[0])
+    warp_size = np.ceil(window_ratio*x.shape[1]).astype(int)
+    
+    # Handle edge cases: ensure warp_size is at least 1
+    warp_size = max(1, warp_size)
+    window_steps = np.arange(warp_size)
+        
+    ret = np.zeros_like(x)
+    
+    for i, pat in enumerate(x):
+        # Check if we have enough room for warping
+        if x.shape[1] <= warp_size + 2:
+            # Not enough space for warping, return original pattern
+            ret[i] = pat
+            continue
+        
+        # Safely generate window start position
+        try:
+            window_start = np.random.randint(low=1, high=x.shape[1]-warp_size-1)
+        except ValueError:
+            # Fallback if random range is invalid
+            window_start = 1
+            
+        window_end = window_start + warp_size
+            
+        for dim in range(x.shape[2]):
+            start_seg = pat[:window_start, dim]
+            window_seg = np.interp(
+                np.linspace(0, warp_size-1, num=int(warp_size*warp_scales[i])), 
+                window_steps, 
+                pat[window_start:window_end, dim]
+            )
+            end_seg = pat[window_end:, dim]
+            warped = np.concatenate((start_seg, window_seg, end_seg))                
+            ret[i, :, dim] = np.interp(
+                np.arange(x.shape[1]), 
+                np.linspace(0, x.shape[1]-1., num=warped.size), 
+                warped
+            ).T
+    
+    return ret
+
+def spawner(x, labels, sigma=0.05, verbose=0):
+    # https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6983028/
+    # use verbose=-1 to turn off warnings
+    # use verbose=1 to print out figures
+    
+    import dtw as dtw
+    
+    # Fix for the random_points generation
+    if x.shape[1] <= 2:  # Check if there are enough time points
+        if verbose > -1:
+            print("Warning: Time series too short for spawner augmentation")
+        return x  # Return the original data if too short
+    
+    # Generate random points safely for each time series
+    random_points = np.zeros(x.shape[0], dtype=int)
+    for i in range(x.shape[0]):
+        try:
+            random_points[i] = np.random.randint(low=1, high=x.shape[1]-1)
+        except ValueError:
+            # Fallback if random range is invalid
+            random_points[i] = 1
+    
+    window = np.ceil(x.shape[1] / 10.).astype(int)
+    window = max(1, window)  # Ensure window is at least 1
+    
+    orig_steps = np.arange(x.shape[1])
+    l = np.argmax(labels, axis=1) if labels.ndim > 1 else labels
+    
+    ret = np.zeros_like(x)
+    for i, pat in enumerate(tqdm(x) if 'tqdm' in globals() else x):
+        # guarantees that same one isn't selected
+        choices = np.delete(np.arange(x.shape[0]), i)
+        # remove ones of different classes
+        choices = np.where(l[choices] == l[i])[0]
+        if choices.size > 0:
+            random_sample = x[np.random.choice(choices)]
+            
+            # SPAWNER splits the path into two randomly
+            try:
+                # Handle potential edge cases with very small sequences
+                random_point = random_points[i]
+                if random_point <= 0:
+                    random_point = 1
+                if random_point >= x.shape[1]:
+                    random_point = x.shape[1] - 1
+                
+                # Check if window size is appropriate
+                if window >= min(random_point, pat.shape[0] - random_point):
+                    # Adjust window if it's too large
+                    adjusted_window = max(1, min(random_point, pat.shape[0] - random_point) - 1)
+                    if verbose > -1:
+                        print(f"Warning: Adjusting window from {window} to {adjusted_window}")
+                    window = adjusted_window
+                
+                path1 = dtw.dtw(pat[:random_point], random_sample[:random_point], 
+                               dtw.RETURN_PATH, slope_constraint="symmetric", window=window)
+                               
+                path2 = dtw.dtw(pat[random_point:], random_sample[random_point:], 
+                               dtw.RETURN_PATH, slope_constraint="symmetric", window=window)
+                
+                combined = np.concatenate((np.vstack(path1), np.vstack(path2+random_point)), axis=1)
+                
+                if verbose:
+                    print(random_point)
+                    dtw_value, cost, DTW_map, path = dtw.dtw(pat, random_sample, 
+                                                          return_flag=dtw.RETURN_ALL, 
+                                                          slope_constraint="symmetric", 
+                                                          window=window)
+                    dtw.draw_graph1d(cost, DTW_map, path, pat, random_sample)
+                    dtw.draw_graph1d(cost, DTW_map, combined, pat, random_sample)
+                
+                mean = np.mean([pat[combined[0]], random_sample[combined[1]]], axis=0)
+                
+                # Handle potential size mismatch
+                if mean.shape[0] > 0:
+                    for dim in range(x.shape[2]):
+                        ret[i,:,dim] = np.interp(orig_steps, 
+                                               np.linspace(0, x.shape[1]-1., num=mean.shape[0]), 
+                                               mean[:,dim]).T
+                else:
+                    if verbose > -1:
+                        print("Warning: DTW produced empty path, skipping augmentation")
+                    ret[i,:] = pat
+                    
+            except Exception as e:
+                if verbose > -1:
+                    print(f"Error in DTW computation: {e}")
+                ret[i,:] = pat
+        else:
+            if verbose > -1:
+                print(f"There is only one pattern of class {l[i]}, skipping pattern average")
+            ret[i,:] = pat
+    
+    # Assuming jitter is defined elsewhere
+    try:
+        return jitter(ret, sigma=sigma)
+    except:
+        if verbose > -1:
+            print("Warning: jitter function failed or not found, returning unjittered data")
+        return ret
+
+def wdba(x, labels, batch_size=6, slope_constraint="symmetric", use_window=True, verbose=0):
+    # https://ieeexplore.ieee.org/document/8215569
+    # use verbose = -1 to turn off warnings    
+    # slope_constraint is for DTW. "symmetric" or "asymmetric"
+    
+    import dtw as dtw
+    
+    if use_window:
+        window = np.ceil(x.shape[1] / 10.).astype(int)
+    else:
+        window = None
+    orig_steps = np.arange(x.shape[1])
+    l = np.argmax(labels, axis=1) if labels.ndim > 1 else labels
+        
+    ret = np.zeros_like(x)
+    for i in tqdm(range(ret.shape[0])):
+        # get the same class as i
+        choices = np.where(l == l[i])[0]
+        if choices.size > 0:        
+            # pick random intra-class pattern
+            k = min(choices.size, batch_size)
+            random_prototypes = x[np.random.choice(choices, k, replace=False)]
+            
+            # calculate dtw between all
+            dtw_matrix = np.zeros((k, k))
+            for p, prototype in enumerate(random_prototypes):
+                for s, sample in enumerate(random_prototypes):
+                    if p == s:
+                        dtw_matrix[p, s] = 0.
+                    else:
+                        dtw_matrix[p, s] = dtw.dtw(prototype, sample, dtw.RETURN_VALUE, slope_constraint=slope_constraint, window=window)
+                        
+            # get medoid
+            medoid_id = np.argsort(np.sum(dtw_matrix, axis=1))[0]
+            nearest_order = np.argsort(dtw_matrix[medoid_id])
+            medoid_pattern = random_prototypes[medoid_id]
+            
+            # start weighted DBA
+            average_pattern = np.zeros_like(medoid_pattern)
+            weighted_sums = np.zeros((medoid_pattern.shape[0]))
+            for nid in nearest_order:
+                if nid == medoid_id or dtw_matrix[medoid_id, nearest_order[1]] == 0.:
+                    average_pattern += medoid_pattern 
+                    weighted_sums += np.ones_like(weighted_sums) 
+                else:
+                    path = dtw.dtw(medoid_pattern, random_prototypes[nid], dtw.RETURN_PATH, slope_constraint=slope_constraint, window=window)
+                    dtw_value = dtw_matrix[medoid_id, nid]
+                    warped = random_prototypes[nid, path[1]]
+                    weight = np.exp(np.log(0.5)*dtw_value/dtw_matrix[medoid_id, nearest_order[1]])
+                    average_pattern[path[0]] += weight * warped
+                    weighted_sums[path[0]] += weight 
+            
+            ret[i,:] = average_pattern / weighted_sums[:,np.newaxis]
+        else:
+            if verbose > -1:
+                print("There is only one pattern of class %d, skipping pattern average"%l[i])
+            ret[i,:] = x[i]
+    return ret
+
+
+
+def random_guided_warp(x, labels, slope_constraint="symmetric", use_window=True, dtw_type="normal", verbose=0):
+    # use verbose = -1 to turn off warnings
+    # slope_constraint is for DTW. "symmetric" or "asymmetric"
+    # dtw_type is for shapeDTW or DTW. "normal" or "shape"
+    
+    import dtw as dtw
+    
+    if use_window:
+        window = np.ceil(x.shape[1] / 10.).astype(int)
+    else:
+        window = None
+    orig_steps = np.arange(x.shape[1])
+    l = np.argmax(labels, axis=1) if labels.ndim > 1 else labels
+    
+    ret = np.zeros_like(x)
+    for i, pat in enumerate(tqdm(x)):
+        # guarentees that same one isnt selected
+        choices = np.delete(np.arange(x.shape[0]), i)
+        # remove ones of different classes
+        choices = np.where(l[choices] == l[i])[0]
+        if choices.size > 0:        
+            # pick random intra-class pattern
+            random_prototype = x[np.random.choice(choices)]
+            
+            if dtw_type == "shape":
+                path = dtw.shape_dtw(random_prototype, pat, dtw.RETURN_PATH, slope_constraint=slope_constraint, window=window)
+            else:
+                path = dtw.dtw(random_prototype, pat, dtw.RETURN_PATH, slope_constraint=slope_constraint, window=window)
+                            
+            # Time warp
+            warped = pat[path[1]]
+            for dim in range(x.shape[2]):
+                ret[i,:,dim] = np.interp(orig_steps, np.linspace(0, x.shape[1]-1., num=warped.shape[0]), warped[:,dim]).T
+        else:
+            if verbose > -1:
+                print("There is only one pattern of class %d, skipping timewarping"%l[i])
+            ret[i,:] = pat
+    return ret
+
+def random_guided_warp_shape(x, labels, slope_constraint="symmetric", use_window=True):
+    return random_guided_warp(x, labels, slope_constraint, use_window, dtw_type="shape")
+
+def discriminative_guided_warp(x, labels, batch_size=6, slope_constraint="symmetric", use_window=True, dtw_type="normal", use_variable_slice=True, verbose=0):
+    # use verbose = -1 to turn off warnings
+    # slope_constraint is for DTW. "symmetric" or "asymmetric"
+    # dtw_type is for shapeDTW or DTW. "normal" or "shape"
+    
+    import dtw as dtw
+    
+    if use_window:
+        window = np.ceil(x.shape[1] / 10.).astype(int)
+    else:
+        window = None
+    orig_steps = np.arange(x.shape[1])
+    l = np.argmax(labels, axis=1) if labels.ndim > 1 else labels
+    
+    positive_batch = np.ceil(batch_size / 2).astype(int)
+    negative_batch = np.floor(batch_size / 2).astype(int)
+        
+    ret = np.zeros_like(x)
+    warp_amount = np.zeros(x.shape[0])
+    for i, pat in enumerate(tqdm(x)):
+        # guarentees that same one isnt selected
+        choices = np.delete(np.arange(x.shape[0]), i)
+        
+        # remove ones of different classes
+        positive = np.where(l[choices] == l[i])[0]
+        negative = np.where(l[choices] != l[i])[0]
+        
+        if positive.size > 0 and negative.size > 0:
+            pos_k = min(positive.size, positive_batch)
+            neg_k = min(negative.size, negative_batch)
+            positive_prototypes = x[np.random.choice(positive, pos_k, replace=False)]
+            negative_prototypes = x[np.random.choice(negative, neg_k, replace=False)]
+                        
+            # vector embedding and nearest prototype in one
+            pos_aves = np.zeros((pos_k))
+            neg_aves = np.zeros((pos_k))
+            if dtw_type == "shape":
+                for p, pos_prot in enumerate(positive_prototypes):
+                    for ps, pos_samp in enumerate(positive_prototypes):
+                        if p != ps:
+                            pos_aves[p] += (1./(pos_k-1.))*dtw.shape_dtw(pos_prot, pos_samp, dtw.RETURN_VALUE, slope_constraint=slope_constraint, window=window)
+                    for ns, neg_samp in enumerate(negative_prototypes):
+                        neg_aves[p] += (1./neg_k)*dtw.shape_dtw(pos_prot, neg_samp, dtw.RETURN_VALUE, slope_constraint=slope_constraint, window=window)
+                selected_id = np.argmax(neg_aves - pos_aves)
+                path = dtw.shape_dtw(positive_prototypes[selected_id], pat, dtw.RETURN_PATH, slope_constraint=slope_constraint, window=window)
+            else:
+                for p, pos_prot in enumerate(positive_prototypes):
+                    for ps, pos_samp in enumerate(positive_prototypes):
+                        if p != ps:
+                            pos_aves[p] += (1./(pos_k-1.))*dtw.dtw(pos_prot, pos_samp, dtw.RETURN_VALUE, slope_constraint=slope_constraint, window=window)
+                    for ns, neg_samp in enumerate(negative_prototypes):
+                        neg_aves[p] += (1./neg_k)*dtw.dtw(pos_prot, neg_samp, dtw.RETURN_VALUE, slope_constraint=slope_constraint, window=window)
+                selected_id = np.argmax(neg_aves - pos_aves)
+                path = dtw.dtw(positive_prototypes[selected_id], pat, dtw.RETURN_PATH, slope_constraint=slope_constraint, window=window)
+                   
+            # Time warp
+            warped = pat[path[1]]
+            warp_path_interp = np.interp(orig_steps, np.linspace(0, x.shape[1]-1., num=warped.shape[0]), path[1])
+            warp_amount[i] = np.sum(np.abs(orig_steps-warp_path_interp))
+            for dim in range(x.shape[2]):
+                ret[i,:,dim] = np.interp(orig_steps, np.linspace(0, x.shape[1]-1., num=warped.shape[0]), warped[:,dim]).T
+        else:
+            if verbose > -1:
+                print("There is only one pattern of class %d"%l[i])
+            ret[i,:] = pat
+            warp_amount[i] = 0.
+    if use_variable_slice:
+        max_warp = np.max(warp_amount)
+        if max_warp == 0:
+            # unchanged
+            ret = window_slice(ret, reduce_ratio=0.9)
+        else:
+            for i, pat in enumerate(ret):
+                # Variable Sllicing
+                ret[i] = window_slice(pat[np.newaxis,:,:], reduce_ratio=0.9+0.1*warp_amount[i]/max_warp)[0]
+    return ret
+
+def discriminative_guided_warp_shape(x, labels, batch_size=6, slope_constraint="symmetric", use_window=True):
+    return discriminative_guided_warp(x, labels, batch_size, slope_constraint, use_window, dtw_type="shape")

time_series_classification/MultiRocket/dtw.py

@@ -0,0 +1,226 @@
+# Adapted from time_series_augmentation (https://github.com/uchidalab/time_series_augmentation)
+# Original: Apache License 2.0 by Brian Kenji Iwana and Seiichi Uchida
+
+__author__ = 'Brian Iwana'
+
+import numpy as np
+import math
+import sys
+
+RETURN_VALUE = 0
+RETURN_PATH = 1
+RETURN_ALL = -1
+
+# Core DTW
+def _traceback(DTW, slope_constraint):
+    i, j = np.array(DTW.shape) - 1
+    p, q = [i-1], [j-1]
+    
+    if slope_constraint == "asymmetric":
+        while (i > 1):
+            tb = np.argmin((DTW[i-1, j], DTW[i-1, j-1], DTW[i-1, j-2]))
+
+            if (tb == 0):
+                i = i - 1
+            elif (tb == 1):
+                i = i - 1
+                j = j - 1
+            elif (tb == 2):
+                i = i - 1
+                j = j - 2
+
+            p.insert(0, i-1)
+            q.insert(0, j-1)
+    elif slope_constraint == "symmetric":
+        while (i > 1 or j > 1):
+            tb = np.argmin((DTW[i-1, j-1], DTW[i-1, j], DTW[i, j-1]))
+
+            if (tb == 0):
+                i = i - 1
+                j = j - 1
+            elif (tb == 1):
+                i = i - 1
+            elif (tb == 2):
+                j = j - 1
+
+            p.insert(0, i-1)
+            q.insert(0, j-1)
+    else:
+        sys.exit("Unknown slope constraint %s"%slope_constraint)
+        
+    return (np.array(p), np.array(q))
+
+def dtw(prototype, sample, return_flag = RETURN_VALUE, slope_constraint="asymmetric", window=None):
+    """ Computes the DTW of two sequences.
+    :param prototype: np array [0..b]
+    :param sample: np array [0..t]
+    :param extended: bool
+    """
+    p = prototype.shape[0]
+    assert p != 0, "Prototype empty!"
+    s = sample.shape[0]
+    assert s != 0, "Sample empty!"
+    
+    if window is None:
+        window = s
+    
+    cost = np.full((p, s), np.inf)
+    for i in range(p):
+        start = max(0, i-window)
+        end = min(s, i+window)+1
+        cost[i,start:end]=np.linalg.norm(sample[start:end] - prototype[i], axis=1)
+
+    DTW = _cummulative_matrix(cost, slope_constraint, window)
+        
+    if return_flag == RETURN_ALL:
+        return DTW[-1,-1], cost, DTW[1:,1:], _traceback(DTW, slope_constraint)
+    elif return_flag == RETURN_PATH:
+        return _traceback(DTW, slope_constraint)
+    else:
+        return DTW[-1,-1]
+
+def _cummulative_matrix(cost, slope_constraint, window):
+    p = cost.shape[0]
+    s = cost.shape[1]
+    
+    # Note: DTW is one larger than cost and the original patterns
+    DTW = np.full((p+1, s+1), np.inf)
+
+    DTW[0, 0] = 0.0
+
+    if slope_constraint == "asymmetric":
+        for i in range(1, p+1):
+            if i <= window+1:
+                DTW[i,1] = cost[i-1,0] + min(DTW[i-1,0], DTW[i-1,1])
+            for j in range(max(2, i-window), min(s, i+window)+1):
+                DTW[i,j] = cost[i-1,j-1] + min(DTW[i-1,j-2], DTW[i-1,j-1], DTW[i-1,j])
+    elif slope_constraint == "symmetric":
+        for i in range(1, p+1):
+            for j in range(max(1, i-window), min(s, i+window)+1):
+                DTW[i,j] = cost[i-1,j-1] + min(DTW[i-1,j-1], DTW[i,j-1], DTW[i-1,j])
+    else:
+        sys.exit("Unknown slope constraint %s"%slope_constraint)
+        
+    return DTW
+
+def shape_dtw(prototype, sample, return_flag = RETURN_VALUE, slope_constraint="asymmetric", window=None, descr_ratio=0.05):
+    """ Computes the shapeDTW of two sequences.
+    :param prototype: np array [0..b]
+    :param sample: np array [0..t]
+    :param extended: bool
+    """
+    # shapeDTW
+    # https://www.sciencedirect.com/science/article/pii/S0031320317303710
+    
+    p = prototype.shape[0]
+    assert p != 0, "Prototype empty!"
+    s = sample.shape[0]
+    assert s != 0, "Sample empty!"
+    
+    if window is None:
+        window = s
+        
+    p_feature_len = np.clip(np.round(p * descr_ratio), 5, 100).astype(int)
+    s_feature_len = np.clip(np.round(s * descr_ratio), 5, 100).astype(int)
+    
+    # padding
+    p_pad_front = (np.ceil(p_feature_len / 2.)).astype(int)
+    p_pad_back = (np.floor(p_feature_len / 2.)).astype(int)
+    s_pad_front = (np.ceil(s_feature_len / 2.)).astype(int)
+    s_pad_back = (np.floor(s_feature_len / 2.)).astype(int)
+    
+    prototype_pad = np.pad(prototype, ((p_pad_front, p_pad_back), (0, 0)), mode="edge") 
+    sample_pad = np.pad(sample, ((s_pad_front, s_pad_back), (0, 0)), mode="edge") 
+    p_p = prototype_pad.shape[0]
+    s_p = sample_pad.shape[0]
+        
+    cost = np.full((p, s), np.inf)
+    for i in range(p):
+        for j in range(max(0, i-window), min(s, i+window)):
+            cost[i, j] = np.linalg.norm(sample_pad[j:j+s_feature_len] - prototype_pad[i:i+p_feature_len])
+            
+    DTW = _cummulative_matrix(cost, slope_constraint=slope_constraint, window=window)
+    
+    if return_flag == RETURN_ALL:
+        return DTW[-1,-1], cost, DTW[1:,1:], _traceback(DTW, slope_constraint)
+    elif return_flag == RETURN_PATH:
+        return _traceback(DTW, slope_constraint)
+    else:
+        return DTW[-1,-1]
+    
+# Draw helpers
+def draw_graph2d(cost, DTW, path, prototype, sample):
+    import matplotlib.pyplot as plt
+    plt.figure(figsize=(12, 8))
+   # plt.subplots_adjust(left=.02, right=.98, bottom=.001, top=.96, wspace=.05, hspace=.01)
+
+    #cost
+    plt.subplot(2, 3, 1)
+    plt.imshow(cost.T, cmap=plt.cm.gray, interpolation='none', origin='lower')
+    plt.plot(path[0], path[1], 'y')
+    plt.xlim((-0.5, cost.shape[0]-0.5))
+    plt.ylim((-0.5, cost.shape[0]-0.5))
+
+    #dtw
+    plt.subplot(2, 3, 2)
+    plt.imshow(DTW.T, cmap=plt.cm.gray, interpolation='none', origin='lower')
+    plt.plot(path[0]+1, path[1]+1, 'y')
+    plt.xlim((-0.5, DTW.shape[0]-0.5))
+    plt.ylim((-0.5, DTW.shape[0]-0.5))
+
+    #prototype
+    plt.subplot(2, 3, 4)
+    plt.plot(prototype[:,0], prototype[:,1], 'b-o')
+
+    #connection
+    plt.subplot(2, 3, 5)
+    for i in range(0,path[0].shape[0]):
+        plt.plot([prototype[path[0][i],0], sample[path[1][i],0]],[prototype[path[0][i],1], sample[path[1][i],1]], 'y-')
+    plt.plot(sample[:,0], sample[:,1], 'g-o')
+    plt.plot(prototype[:,0], prototype[:,1], 'b-o')
+
+    #sample
+    plt.subplot(2, 3, 6)
+    plt.plot(sample[:,0], sample[:,1], 'g-o')
+
+    plt.tight_layout()
+    plt.show()
+
+def draw_graph1d(cost, DTW, path, prototype, sample):
+    import matplotlib.pyplot as plt
+    plt.figure(figsize=(12, 8))
+   # plt.subplots_adjust(left=.02, right=.98, bottom=.001, top=.96, wspace=.05, hspace=.01)
+    p_steps = np.arange(prototype.shape[0])
+    s_steps = np.arange(sample.shape[0])
+
+    #cost
+    plt.subplot(2, 3, 1)
+    plt.imshow(cost.T, cmap=plt.cm.gray, interpolation='none', origin='lower')
+    plt.plot(path[0], path[1], 'y')
+    plt.xlim((-0.5, cost.shape[0]-0.5))
+    plt.ylim((-0.5, cost.shape[0]-0.5))
+
+    #dtw
+    plt.subplot(2, 3, 2)
+    plt.imshow(DTW.T, cmap=plt.cm.gray, interpolation='none', origin='lower')
+    plt.plot(path[0]+1, path[1]+1, 'y')
+    plt.xlim((-0.5, DTW.shape[0]-0.5))
+    plt.ylim((-0.5, DTW.shape[0]-0.5))
+
+    #prototype
+    plt.subplot(2, 3, 4)
+    plt.plot(p_steps, prototype[:,0], 'b-o')
+
+    #connection
+    plt.subplot(2, 3, 5)
+    for i in range(0,path[0].shape[0]):
+        plt.plot([path[0][i], path[1][i]],[prototype[path[0][i],0], sample[path[1][i],0]], 'y-')
+    plt.plot(p_steps, sample[:,0], 'g-o')
+    plt.plot(s_steps, prototype[:,0], 'b-o')
+
+    #sample
+    plt.subplot(2, 3, 6)
+    plt.plot(s_steps, sample[:,0], 'g-o')
+
+    plt.tight_layout()
+    plt.show()

time_series_classification/MultiRocket/hyperparamter_tune.py

@@ -0,0 +1,478 @@
+# Modified from MultiRocket (https://github.com/ChangWeiTan/MultiRocket)
+# Copyright (C) 2025 Jafar Bakhshaliyev
+# Licensed under GNU General Public License v3.0
+
+
+import argparse
+import os
+import time
+import sys
+import numpy as np
+import pandas as pd
+from sklearn.metrics import accuracy_score, log_loss
+from sklearn.preprocessing import LabelEncoder
+from sklearn.model_selection import train_test_split
+from sktime.utils.data_io import load_from_tsfile_to_dataframe
+from scipy.special import softmax
+
+from multirocket.multirocket_multivariate import MultiRocket
+from utils.data_loader import process_ts_data
+from utils.tools import create_directory
+
+import augmentation as aug 
+
+pd.set_option('display.max_columns', 500)
+
+def run_augmentation(x, y, args):
+    """
+    Apply data augmentation to the input data based on args.
+    
+    Parameters:
+    -----------
+    x : numpy.ndarray
+        Original time series data
+    y : numpy.ndarray
+        Original labels
+    args : argparse.Namespace
+        Command line arguments containing augmentation options
+        
+    Returns:
+    --------
+    x_aug : numpy.ndarray
+        Augmented time series data
+    y_aug : numpy.ndarray
+        Augmented labels
+    augmentation_tags : str
+        String describing the applied augmentations
+    """
+    print("Augmenting data for dataset %s" % args.problem)
+    np.random.seed(args.seed)
+    x_aug = x.copy()
+    y_aug = y.copy()
+    
+    augmentation_tags = ""
+    
+    if args.augmentation_ratio > 0:
+        augmentation_tags = "%d" % args.augmentation_ratio
+        print(f"Original training size: {x.shape[0]} samples")
+        
+        for n in range(args.augmentation_ratio):
+            x_temp, current_tags = augment(x, y, args)
+            
+            if x_temp.shape != x.shape:
+                print(f"Warning: Augmented data shape {x_temp.shape} doesn't match original shape {x.shape}")
+                continue
+                
+            x_aug = np.concatenate((x_aug, x_temp), axis=0)
+            y_aug = np.append(y_aug, y)
+            
+            print(f"Round {n+1}: {current_tags} done - Added {x_temp.shape[0]} samples")
+            
+            if n == 0:
+                augmentation_tags += current_tags
+                
+        print(f"Augmented training size: {x_aug.shape[0]} samples")
+        
+        if args.extra_tag:
+            augmentation_tags += "_" + args.extra_tag
+    else:
+        augmentation_tags = "none"
+        if args.extra_tag:
+            augmentation_tags = args.extra_tag
+            
+    return x_aug, y_aug, augmentation_tags
+
+    
+def augment(x, y, args):
+    """
+    Apply specified augmentations to the multivariate time series data.
+    
+    Parameters:
+    -----------
+    x : numpy.ndarray
+        Original time series data with shape (n_samples, n_dimensions, n_timesteps)
+    y : numpy.ndarray
+        Original labels
+    args : argparse.Namespace
+        Command line arguments containing augmentation options
+        
+    Returns:
+    --------
+    x : numpy.ndarray
+        Augmented time series data
+    augmentation_tags : str
+        String describing the applied augmentations
+    """
+    augmentation_tags = ""
+    
+    x_aug = x.copy()
+    
+
+    if len(x_aug.shape) != 3:
+        if len(x_aug.shape) == 2:
+            x_aug = x_aug.reshape(x_aug.shape[0], 1, x_aug.shape[1])
+            print(f"Reshaped to {x_aug.shape} for processing")
+    
+    if args.jitter:
+        x_aug = aug.jitter(x_aug)
+        augmentation_tags += "_jitter"
+
+    if args.tps and args.patch_len > 0:
+        x_aug = aug.tps(x_aug, y, args.patch_len, args.stride, args.shuffle_rate)
+        augmentation_tags += "_tps"
+
+    if args.scaling:
+        x_aug = aug.scaling(x_aug)
+        augmentation_tags += "_scaling"
+        
+    if args.rotation:
+        x_aug = aug.rotation(x_aug)
+        augmentation_tags += "_rotation"
+        
+    if args.permutation:
+        x_aug = aug.permutation(x_aug)
+        augmentation_tags += "_permutation"
+        
+    if args.randompermutation:
+        x_aug = aug.permutation(x_aug, seg_mode="random")
+        augmentation_tags += "_randomperm"
+        
+    if args.magwarp:
+        x_aug = aug.magnitude_warp(x_aug)
+        augmentation_tags += "_magwarp"
+        
+    if args.timewarp:
+        x_aug = aug.time_warp(x_aug)
+        augmentation_tags += "_timewarp"
+        
+    if args.windowslice:
+        x_aug = aug.window_slice(x_aug)
+        augmentation_tags += "_windowslice"
+        
+    if args.windowwarp:
+        x_aug = aug.window_warp(x_aug)
+        augmentation_tags += "_windowwarp"
+        
+    if args.spawner:
+        x_aug = aug.spawner(x_aug, y)
+        augmentation_tags += "_spawner"
+        
+    if args.dtwwarp:
+        x_aug = aug.random_guided_warp(x_aug, y)
+        augmentation_tags += "_rgw"
+        
+    if args.shapedtwwarp:
+        x_aug = aug.random_guided_warp_shape(x_aug, y)
+        augmentation_tags += "_rgws"
+        
+    if args.wdba:
+        x_aug = aug.wdba(x_aug, y)
+        augmentation_tags += "_wdba"
+        
+    if args.discdtw:
+        x_aug = aug.discriminative_guided_warp(x_aug, y)
+        augmentation_tags += "_dgw"
+        
+    if args.discsdtw:
+        x_aug = aug.discriminative_guided_warp_shape(x_aug, y)
+        augmentation_tags += "_dgws"
+        
+    if not augmentation_tags:
+        augmentation_tags = "_none"
+    
+    return x_aug, augmentation_tags
+
+def run_multirocket_hyperparameter_tuning(args):
+    """
+    Run MultiRocket hyperparameter tuning on a dataset with train/validation split.
+    
+    Parameters:
+    -----------
+    args : argparse.Namespace
+        Command line arguments containing options
+    
+    Returns:
+    --------
+    results_df : pandas.DataFrame
+        DataFrame containing results of the hyperparameter tuning
+    """
+    problem = args.problem
+    data_path = args.datapath
+    data_folder = data_path + problem + "/"
+    
+    # Set output directory
+    output_path = os.getcwd() + "/output/"
+    classifier_name = f"MultiRocket_{args.num_features}"
+    
+    output_dir = "{}/multirocket/hyperparameter_tuning/{}/{}/".format(
+        output_path,
+        classifier_name,
+        problem
+    )
+    
+    if args.save:
+        create_directory(output_dir)
+    
+    train_file = data_folder + problem + "_TRAIN.ts"
+    test_file = data_folder + problem + "_TEST.ts"
+    
+    print("Loading data")
+    X_train_full, y_train_full = load_from_tsfile_to_dataframe(train_file)
+    
+    encoder = LabelEncoder()
+    y_train_full = encoder.fit_transform(y_train_full)
+
+    X_train_full_processed = process_ts_data(X_train_full, normalise=False)
+    
+    # Split the training set into training and validation sets (80/20)
+    try:
+        if len(np.unique(y_train_full)) > 1:
+            class_counts = np.bincount(y_train_full.astype(int))
+            if np.min(class_counts[class_counts > 0]) >= 2:
+                train_indices, val_indices = train_test_split(
+                    np.arange(len(y_train_full)), 
+                    test_size=0.2, 
+                    random_state=args.seed,
+                    stratify=y_train_full
+                )
+            else:
+                print("Warning: Some classes have only 1 sample. Using regular split instead of stratified split.")
+                train_indices, val_indices = train_test_split(
+                    np.arange(len(y_train_full)), 
+                    test_size=0.2, 
+                    random_state=args.seed,
+                    stratify=None
+                )
+        else:
+            train_indices, val_indices = train_test_split(
+                np.arange(len(y_train_full)), 
+                test_size=0.2, 
+                random_state=args.seed,
+                stratify=None
+            )
+    except Exception as e:
+        print(f"Warning: Failed to perform stratified split: {e}")
+        print("Falling back to regular random split.")
+        train_indices, val_indices = train_test_split(
+            np.arange(len(y_train_full)), 
+            test_size=0.2, 
+            random_state=args.seed,
+            stratify=None
+        )
+    
+    y_train = y_train_full[train_indices].copy()
+    y_val = y_train_full[val_indices].copy()
+    
+    X_train = X_train_full_processed[train_indices]
+    X_val = X_train_full_processed[val_indices]
+    
+    print(f"Split training data: Train shape: {X_train.shape}, Validation shape: {X_val.shape}")
+    
+    # Apply augmentation 
+    augmentation_tags = "none"
+    if args.use_augmentation:
+        X_train_aug, y_train_aug, augmentation_tags = run_augmentation(X_train, y_train, args)
+    else:
+        X_train_aug, y_train_aug = X_train.copy(), y_train.copy()
+    
+    train_accuracies = []
+    val_accuracies = []
+    val_cross_entropies = []
+    train_times = []
+    
+    for iteration in range(args.iterations):
+        print(f"Running iteration {iteration+1}/{args.iterations}")
+        
+        start_time = time.perf_counter()
+        
+        np.random.seed(args.seed + iteration)
+        
+        classifier = MultiRocket(
+            num_features=args.num_features,
+            classifier="logistic",
+            verbose=args.verbose
+        )
+        
+        yhat_train = classifier.fit(
+            X_train_aug, y_train_aug,
+            predict_on_train=True 
+        )
+        
+        yhat_val = classifier.predict(X_val)
+
+        train_acc = accuracy_score(y_train_aug, yhat_train)
+        train_accuracies.append(train_acc)
+        
+        val_acc = accuracy_score(y_val, yhat_val)
+        val_accuracies.append(val_acc)
+        
+        try:
+            val_proba = classifier.predict_proba(X_val)
+            
+            try:
+                all_classes = np.unique(np.concatenate((y_train_aug, y_val)))
+                val_cross_entropy = log_loss(y_val, val_proba, labels=all_classes)
+                val_cross_entropies.append(val_cross_entropy)
+            except Exception as e:
+                print(f"Warning: Could not calculate cross-entropy: {e}")
+                val_cross_entropy = np.nan
+                val_cross_entropies.append(val_cross_entropy)
+                
+        except (AttributeError, NotImplementedError) as e:
+            print(f"Warning: Could not get probability estimates: {e}")
+            val_cross_entropy = np.nan
+            val_cross_entropies.append(val_cross_entropy)
+        
+        train_time = classifier.train_duration
+        train_times.append(train_time)
+        
+        print(f"Iteration {iteration+1} - Train Accuracy: {train_acc:.4f}")
+        print(f"Iteration {iteration+1} - Validation Accuracy: {val_acc:.4f}")
+        if not np.isnan(val_cross_entropy):
+            print(f"Iteration {iteration+1} - Validation Cross-Entropy: {val_cross_entropy:.4f}")
+        print(f"Iteration {iteration+1} - Train Time: {train_time:.2f} seconds")
+    
+    # Calculate mean and standard deviation
+    mean_train_accuracy = np.mean(train_accuracies)
+    std_train_accuracy = np.std(train_accuracies)
+    mean_val_accuracy = np.mean(val_accuracies)
+    std_val_accuracy = np.std(val_accuracies)
+    mean_val_cross_entropy = np.nanmean(val_cross_entropies) if not all(np.isnan(val_cross_entropies)) else np.nan
+    std_val_cross_entropy = np.nanstd(val_cross_entropies) if not all(np.isnan(val_cross_entropies)) else np.nan
+    mean_train_time = np.mean(train_times)
+    
+    print(f"\nHyperparameter Tuning Results for {problem} with augmentation: {augmentation_tags}")
+    print(f"Original train size: {X_train.shape[0]} samples")
+    print(f"Augmented train size: {X_train_aug.shape[0]} samples")
+    print(f"Validation size: {X_val.shape[0]} samples")
+    print(f"Mean Train Accuracy: {mean_train_accuracy:.4f} ± {std_train_accuracy:.4f}")
+    print(f"Mean Validation Accuracy: {mean_val_accuracy:.4f} ± {std_val_accuracy:.4f}")
+    if not np.isnan(mean_val_cross_entropy):
+        print(f"Mean Validation Cross-Entropy: {mean_val_cross_entropy:.4f} ± {std_val_cross_entropy:.4f}")
+    print(f"Mean Train Time: {mean_train_time:.2f} seconds")
+    
+    # Create results DataFrame
+    results_df = pd.DataFrame({
+        'dataset': [problem],
+        'augmentation': [augmentation_tags],
+        'train_size': [X_train.shape[0]],
+        'train_size_after_aug': [X_train_aug.shape[0]],
+        'val_size': [X_val.shape[0]],
+        'mean_train_accuracy': [mean_train_accuracy],
+        'train_accuracy_std': [std_train_accuracy],
+        'mean_val_accuracy': [mean_val_accuracy],
+        'val_accuracy_std': [std_val_accuracy],
+        'mean_val_cross_entropy': [mean_val_cross_entropy],
+        'val_cross_entropy_std': [std_val_cross_entropy],
+        'mean_train_time': [mean_train_time],
+        'iterations': [args.iterations],
+        'features': [args.num_features],
+        'individual_train_accuracies': [','.join(map(str, train_accuracies))],
+        'individual_val_accuracies': [','.join(map(str, val_accuracies))],
+        'individual_val_cross_entropies': [','.join(map(str, val_cross_entropies))],
+        'patch_len': [args.patch_len],
+        'stride': [args.stride],
+        'shuffle_rate': [args.shuffle_rate]
+    })
+    
+    if args.save:
+        results_filename = f"{output_dir}/multirocket_hyperparameter_tuning_{problem}_{augmentation_tags}.csv"
+        if os.path.exists(results_filename):
+            try:
+                existing_df = pd.read_csv(results_filename)
+                combined_df = pd.concat([existing_df, results_df], ignore_index=True)
+                combined_df.to_csv(results_filename, index=False)
+                print(f"Results appended to {results_filename}")
+            except Exception as e:
+                print(f"Error appending to existing file: {e}")
+                results_df.to_csv(results_filename, index=False)
+                print(f"Created new file instead: {results_filename}")
+        else:
+            results_df.to_csv(results_filename, index=False)
+            print(f"Results saved to new file {results_filename}")
+    
+    return results_df
+
+def list_available_datasets(args):
+    """
+    List all available datasets in the data path.
+    
+    Parameters:
+    -----------
+    args : argparse.Namespace
+        Command line arguments containing options
+    """
+    data_path = args.datapath
+    try:
+        datasets = [d for d in os.listdir(data_path) if os.path.isdir(os.path.join(data_path, d))]
+        print("Available datasets:")
+        for dataset in sorted(datasets):
+            print(f"  - {dataset}")
+        return sorted(datasets)
+    except Exception as e:
+        print(f"Error listing datasets: {e}")
+        return []
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Hyperparameter Tuning for MultiRocket on Multivariate Time Series')
+    
+    # Dataset selection
+    parser.add_argument("-d", "--datapath", type=str, required=False, default="/home/bakhshaliyev/classification-aug/MultiRocket/data/Multivariate_ts/")
+    parser.add_argument("-p", "--problem", type=str, required=False, default="UWaveGestureLibrary")
+    parser.add_argument("-n", "--num_features", type=int, required=False, default=50000)
+    parser.add_argument("-t", "--num_threads", type=int, required=False, default=-1)
+    parser.add_argument("-s", "--save", type=bool, required=False, default=True)
+    parser.add_argument("-v", "--verbose", type=int, required=False, default=2)
+    
+    # Added arguments for tuning
+    parser.add_argument('--iterations', type=int, default=5, help='Number of iterations for each experiment (default: 5)')
+    parser.add_argument('--seed', type=int, default=42, help='Random seed (default: 42)')
+    parser.add_argument('--list', action='store_true', help='List available datasets')
+    
+    # Augmentation control
+    parser.add_argument('--use-augmentation', action='store_true', help='Use data augmentation')
+    parser.add_argument('--augmentation-ratio', type=int, default=0, 
+                      help='Number of augmented copies to add (default: 0)')
+    parser.add_argument('--extra-tag', type=str, default='', 
+                      help='Extra tag to add to augmentation tags')
+    
+    # Augmentation methods
+    parser.add_argument('--jitter', action='store_true', help='Apply jitter augmentation')
+    parser.add_argument('--scaling', action='store_true', help='Apply scaling augmentation')
+    parser.add_argument('--rotation', action='store_true', help='Apply rotation augmentation')
+    parser.add_argument('--permutation', action='store_true', help='Apply permutation augmentation')
+    parser.add_argument('--randompermutation', action='store_true', help='Apply random permutation augmentation')
+    parser.add_argument('--magwarp', action='store_true', help='Apply magnitude warp augmentation')
+    parser.add_argument('--timewarp', action='store_true', help='Apply time warp augmentation')
+    parser.add_argument('--windowslice', action='store_true', help='Apply window slice augmentation')
+    parser.add_argument('--windowwarp', action='store_true', help='Apply window warp augmentation')
+    parser.add_argument('--spawner', action='store_true', help='Apply spawner augmentation')
+    parser.add_argument('--dtwwarp', action='store_true', help='Apply DTW-based warp augmentation')
+    parser.add_argument('--shapedtwwarp', action='store_true', help='Apply shape DTW warp augmentation')
+    parser.add_argument('--wdba', action='store_true', help='Apply WDBA augmentation')
+    parser.add_argument('--discdtw', action='store_true', help='Apply discriminative DTW augmentation')
+    parser.add_argument('--discsdtw', action='store_true', help='Apply discriminative shape DTW augmentation')
+    parser.add_argument('--tps', action='store_true', help='Apply TPS augmentation')
+
+    # TPS specific parameters
+    parser.add_argument('--stride', type=int, default=0, help='# of patches stride')
+    parser.add_argument('--patch_len', type=int, default=0, help='# of patches')
+    parser.add_argument('--shuffle_rate', type=float, default=0.0, help='shuffle rate')
+    
+    args = parser.parse_args()
+    
+    if args.num_threads > 0:
+        import numba
+        numba.set_num_threads(args.num_threads)
+    
+    if args.list:
+        list_available_datasets(args)
+        sys.exit(0)
+    
+    # Run hyperparameter tuning on specified dataset
+    print(f"Running MultiRocket hyperparameter tuning on {args.problem} dataset")
+    print(f"Using {args.num_features} features and {args.iterations} iterations")
+    if args.use_augmentation:
+        print(f"Using data augmentation with ratio {args.augmentation_ratio}")
+    
+    run_multirocket_hyperparameter_tuning(args)

time_series_classification/MultiRocket/main.py

@@ -0,0 +1,467 @@
+# Modified from MultiRocket (https://github.com/ChangWeiTan/MultiRocket)
+# Copyright (C) 2025 Jafar Bakhshaliyev
+# Licensed under GNU General Public License v3.0
+
+
+import argparse
+import os
+import time
+import sys
+import socket
+import platform
+from datetime import datetime
+
+import numba
+import numpy as np
+import pandas as pd
+import psutil
+import pytz
+from sklearn.metrics import accuracy_score
+from sklearn.preprocessing import LabelEncoder
+from sktime.utils.data_io import load_from_tsfile_to_dataframe
+
+from multirocket.multirocket_multivariate import MultiRocket
+from utils.data_loader import process_ts_data
+from utils.tools import create_directory
+
+import augmentation as aug 
+
+pd.set_option('display.max_columns', 500)
+
+def run_augmentation(x, y, args):
+    """
+    Apply data augmentation to the input data based on args.
+    
+    Parameters:
+    -----------
+    x : numpy.ndarray
+        Original time series data
+    y : numpy.ndarray
+        Original labels
+    args : argparse.Namespace
+        Command line arguments containing augmentation options
+        
+    Returns:
+    --------
+    x_aug : numpy.ndarray
+        Augmented time series data
+    y_aug : numpy.ndarray
+        Augmented labels
+    augmentation_tags : str
+        String describing the applied augmentations
+    """
+    print("Augmenting data for dataset %s" % args.problem)
+    np.random.seed(args.seed)
+    x_aug = x.copy()
+    y_aug = y.copy()
+    
+    augmentation_tags = ""
+    
+    if args.augmentation_ratio > 0:
+        augmentation_tags = "%d" % args.augmentation_ratio
+        print(f"Original training size: {x.shape[0]} samples")
+        
+        for n in range(args.augmentation_ratio):
+            x_temp, current_tags = augment(x, y, args)
+            
+            if x_temp.shape != x.shape:
+                print(f"Warning: Augmented data shape {x_temp.shape} doesn't match original shape {x.shape}")
+                continue
+                
+            x_aug = np.concatenate((x_aug, x_temp), axis=0)
+            y_aug = np.append(y_aug, y)
+            
+            print(f"Round {n+1}: {current_tags} done - Added {x_temp.shape[0]} samples")
+            
+            if n == 0:
+                augmentation_tags += current_tags
+                
+        print(f"Augmented training size: {x_aug.shape[0]} samples")
+        
+        if args.extra_tag:
+            augmentation_tags += "_" + args.extra_tag
+    else:
+        augmentation_tags = "none"
+        if args.extra_tag:
+            augmentation_tags = args.extra_tag
+            
+    return x_aug, y_aug, augmentation_tags
+
+    
+def augment(x, y, args):
+    """
+    Apply specified augmentations to the multivariate time series data.
+    
+    Parameters:
+    -----------
+    x : numpy.ndarray
+        Original time series data with shape (n_samples, n_dimensions, n_timesteps)
+    y : numpy.ndarray
+        Original labels
+    args : argparse.Namespace
+        Command line arguments containing augmentation options
+        
+    Returns:
+    --------
+    x : numpy.ndarray
+        Augmented time series data
+    augmentation_tags : str
+        String describing the applied augmentations
+    """
+    augmentation_tags = ""
+    
+    x_aug = x.copy()
+    
+    if len(x_aug.shape) != 3:
+        if len(x_aug.shape) == 2:
+            x_aug = x_aug.reshape(x_aug.shape[0], 1, x_aug.shape[1])
+            print(f"Reshaped to {x_aug.shape} for processing")
+    
+    if args.jitter:
+        x_aug = aug.jitter(x_aug)
+        augmentation_tags += "_jitter"
+
+    if args.tps and args.patch_len > 0:
+        x_aug = aug.tps(x_aug, y, args.patch_len, args.stride, args.shuffle_rate)
+        augmentation_tags += "_tps"
+        
+    if args.scaling:
+        x_aug = aug.scaling(x_aug)
+        augmentation_tags += "_scaling"
+        
+    if args.rotation:
+        x_aug = aug.rotation(x_aug)
+        augmentation_tags += "_rotation"
+        
+    if args.permutation:
+        x_aug = aug.permutation(x_aug)
+        augmentation_tags += "_permutation"
+        
+    if args.randompermutation:
+        x_aug = aug.permutation(x_aug, seg_mode="random")
+        augmentation_tags += "_randomperm"
+        
+    if args.magwarp:
+        x_aug = aug.magnitude_warp(x_aug)
+        augmentation_tags += "_magwarp"
+        
+    if args.timewarp:
+        x_aug = aug.time_warp(x_aug)
+        augmentation_tags += "_timewarp"
+        
+    if args.windowslice:
+        x_aug = aug.window_slice(x_aug)
+        augmentation_tags += "_windowslice"
+        
+    if args.windowwarp:
+        x_aug = aug.window_warp(x_aug)
+        augmentation_tags += "_windowwarp"
+        
+    if args.spawner:
+        x_aug = aug.spawner(x_aug, y)
+        augmentation_tags += "_spawner"
+        
+    if args.dtwwarp:
+        x_aug = aug.random_guided_warp(x_aug, y)
+        augmentation_tags += "_rgw"
+        
+    if args.shapedtwwarp:
+        x_aug = aug.random_guided_warp_shape(x_aug, y)
+        augmentation_tags += "_rgws"
+        
+    if args.wdba:
+        x_aug = aug.wdba(x_aug, y)
+        augmentation_tags += "_wdba"
+        
+    if args.discdtw:
+        x_aug = aug.discriminative_guided_warp(x_aug, y)
+        augmentation_tags += "_dgw"
+        
+    if args.discsdtw:
+        x_aug = aug.discriminative_guided_warp_shape(x_aug, y)
+        augmentation_tags += "_dgws"
+        
+    if not augmentation_tags:
+        augmentation_tags = "_none"
+    
+    return x_aug, augmentation_tags
+
+def run_multirocket_experiment(args):
+    """
+    Run MultiRocket on a dataset with multiple iterations and optional augmentation.
+    
+    Parameters:
+    -----------
+    args : argparse.Namespace
+        Command line arguments containing options
+    
+    Returns:
+    --------
+    results_df : pandas.DataFrame
+        DataFrame containing results of the experiment
+    """
+    problem = args.problem
+    data_path = args.datapath
+    data_folder = data_path + problem + "/"
+    
+    # Set output directory
+    output_path = os.getcwd() + "/output/"
+    classifier_name = f"MultiRocket_{args.num_features}"
+    
+    output_dir = "{}/multirocket/resample_{}/{}/{}/".format(
+        output_path,
+        args.iter,
+        classifier_name,
+        problem
+    )
+    
+    if args.save:
+        create_directory(output_dir)
+    
+    train_file = data_folder + problem + "_TRAIN.ts"
+    test_file = data_folder + problem + "_TEST.ts"
+    
+    # Loading data
+    X_train, y_train = load_from_tsfile_to_dataframe(train_file)
+    X_test, y_test = load_from_tsfile_to_dataframe(test_file)
+    
+    encoder = LabelEncoder()
+    y_train = encoder.fit_transform(y_train)
+    y_test = encoder.transform(y_test)
+    
+    X_train_processed = process_ts_data(X_train, normalise=False)
+    X_test_processed = process_ts_data(X_test, normalise=False)
+    
+    # Apply augmentation 
+    augmentation_tags = "none"
+    if args.use_augmentation:
+        X_train_processed, y_train, augmentation_tags = run_augmentation(X_train_processed, y_train, args)
+    
+    accuracies = []
+    train_times = []
+    test_times = []
+    
+    for iteration in range(args.iterations):
+        print(f"Running iteration {iteration+1}/{args.iterations}")
+        
+        start_time = time.perf_counter()
+        
+        np.random.seed(args.seed + iteration)
+
+        classifier = MultiRocket(
+            num_features=args.num_features,
+            classifier="logistic",
+            verbose=args.verbose
+        )
+        
+        yhat_train = classifier.fit(
+            X_train_processed, y_train,
+            predict_on_train=False
+        )
+        
+        yhat_test = classifier.predict(X_test_processed)
+
+        test_acc = accuracy_score(y_test, yhat_test)
+        
+        if yhat_train is not None:
+            train_acc = accuracy_score(y_train, yhat_train)
+        else:
+            train_acc = -1
+        
+        accuracies.append(test_acc)
+        train_times.append(classifier.train_duration)
+        test_times.append(classifier.test_duration)
+        
+        print(f"Iteration {iteration+1} - Test Accuracy: {test_acc:.4f}, Train Time: {classifier.train_duration:.2f} seconds")
+    
+    mean_accuracy = np.mean(accuracies)
+    std_accuracy = np.std(accuracies)
+    mean_train_time = np.mean(train_times)
+    mean_test_time = np.mean(test_times)
+    
+    print(f"\nResults for {problem} with augmentation: {augmentation_tags}")
+    print(f"Train size: {X_train_processed.shape[0]} samples")
+    print(f"Mean Accuracy: {mean_accuracy:.4f} ± {std_accuracy:.4f}")
+    print(f"Mean Train Time: {mean_train_time:.2f} seconds")
+    print(f"Mean Test Time: {mean_test_time:.2f} seconds")
+    print(f"Individual Accuracies: {accuracies}")
+    
+    results_df = pd.DataFrame({
+        'dataset': [problem],
+        'augmentation': [augmentation_tags],
+        'train_size': [X_train_processed.shape[0]],
+        'test_size': [X_test_processed.shape[0]],
+        'mean_accuracy': [mean_accuracy],
+        'std_accuracy': [std_accuracy],
+        'mean_train_time': [mean_train_time],
+        'mean_test_time': [mean_test_time],
+        'iterations': [args.iterations],
+        'features': [args.num_features],
+        'individual_accuracies': [','.join(map(str, accuracies))],
+        'patch_len': [args.patch_len] if hasattr(args, 'patch_len') else [0],
+        'stride': [args.stride] if hasattr(args, 'stride') else [0],
+        'shuffle_rate': [args.shuffle_rate] if hasattr(args, 'shuffle_rate') else [0.0],
+    })
+    
+    if args.save:
+        results_filename = f"{output_dir}/multirocket_results_{problem}_{augmentation_tags}.csv"
+        if os.path.exists(results_filename):
+            existing_df = pd.read_csv(results_filename)
+            combined_df = pd.concat([existing_df, results_df], ignore_index=True)
+            combined_df.to_csv(results_filename, index=False)
+            print(f"Results appended to {results_filename}")
+        else:
+            results_df.to_csv(results_filename, index=False)
+            print(f"Results saved to new file {results_filename}")
+    
+    return results_df
+
+def run_all_datasets(args):
+    """
+    Run MultiRocket on all available datasets in the data path.
+    
+    Parameters:
+    -----------
+    args : argparse.Namespace
+        Command line arguments containing options
+    """
+    # list of available datasets
+    data_path = args.datapath
+    datasets = [d for d in os.listdir(data_path) if os.path.isdir(os.path.join(data_path, d))]
+    
+    if not datasets:
+        print(f"No datasets found in {data_path}")
+        return
+    
+    print(f"Found {len(datasets)} datasets: {', '.join(datasets)}")
+    
+    results = []
+    
+    for dataset_name in datasets:
+        print(f"\n{'='*50}")
+        print(f"Processing dataset: {dataset_name}")
+        print(f"{'='*50}")
+        
+        try:
+            args.problem = dataset_name
+            
+            result_df = run_multirocket_experiment(args)
+            results.append(result_df)
+        
+        except Exception as e:
+            print(f"Error processing dataset {dataset_name}: {e}")
+    
+    if results:
+        all_results_df = pd.concat(results, ignore_index=True)
+        overall_mean = all_results_df['mean_accuracy'].mean()
+        
+        print("\n" + "="*80)
+        print("SUMMARY OF RESULTS")
+        print("="*80)
+        print(f"{'Dataset':<25} {'Augmentation':<25} {'Mean Accuracy':<15} {'Std Dev':<10}")
+        print("-"*80)
+        
+        for _, row in all_results_df.iterrows():
+            print(f"{row['dataset']:<25} {row['augmentation']:<25} {row['mean_accuracy']:.4f}{' '*8} {row['std_accuracy']:.4f}")
+        
+        print("-"*80)
+        print(f"{'OVERALL':<25} {'':<25} {overall_mean:.4f}")
+        print("="*80)
+        
+        aug_tag = "none" if not args.use_augmentation else "aug"
+        output_path = os.getcwd() + "/output/"
+        all_results_df.to_csv(f"{output_path}/multirocket_summary_results_{aug_tag}.csv", index=False)
+        print(f"\nSummary results saved to {output_path}/multirocket_summary_results_{aug_tag}.csv")
+
+def list_available_datasets(args):
+    """
+    List all available datasets in the data path.
+    
+    Parameters:
+    -----------
+    args : argparse.Namespace
+        Command line arguments containing options
+    """
+    data_path = args.datapath
+    try:
+        datasets = [d for d in os.listdir(data_path) if os.path.isdir(os.path.join(data_path, d))]
+        print("Available datasets:")
+        for dataset in sorted(datasets):
+            print(f"  - {dataset}")
+    except Exception as e:
+        print(f"Error listing datasets: {e}")
+        return []
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Run MultiRocket on multivariate time series datasets with optional augmentation')
+    
+    # Dataset selection
+    parser.add_argument("-d", "--datapath", type=str, required=False, default="/home/bakhshaliyev/classification-aug/MultiRocket/data/Multivariate_ts/") # change to your data path
+    parser.add_argument("-p", "--problem", type=str, required=False, default="UWaveGestureLibrary")
+    parser.add_argument("-i", "--iter", type=int, required=False, default=0)
+    parser.add_argument("-n", "--num_features", type=int, required=False, default=50000)
+    parser.add_argument("-t", "--num_threads", type=int, required=False, default=-1)
+    parser.add_argument("-s", "--save", type=bool, required=False, default=True)
+    parser.add_argument("-v", "--verbose", type=int, required=False, default=2)
+    
+    # Added arguments for augmentation
+    parser.add_argument('--iterations', type=int, default=5, help='Number of iterations for each experiment (default: 5)')
+    parser.add_argument('--seed', type=int, default=42, help='Random seed (default: 42)')
+    parser.add_argument('--list', action='store_true', help='List available datasets')
+    parser.add_argument('--all', action='store_true', help='Run on all available datasets')
+    
+    # Augmentation control
+    parser.add_argument('--use-augmentation', action='store_true', help='Use data augmentation')
+    parser.add_argument('--augmentation-ratio', type=int, default=0, 
+                      help='Number of augmented copies to add (default: 0)')
+    parser.add_argument('--extra-tag', type=str, default='', 
+                      help='Extra tag to add to augmentation tags')
+    
+    # Augmentation methods
+    parser.add_argument('--jitter', action='store_true', help='Apply jitter augmentation')
+    parser.add_argument('--scaling', action='store_true', help='Apply scaling augmentation')
+    parser.add_argument('--rotation', action='store_true', help='Apply rotation augmentation')
+    parser.add_argument('--permutation', action='store_true', help='Apply permutation augmentation')
+    parser.add_argument('--randompermutation', action='store_true', help='Apply random permutation augmentation')
+    parser.add_argument('--magwarp', action='store_true', help='Apply magnitude warp augmentation')
+    parser.add_argument('--timewarp', action='store_true', help='Apply time warp augmentation')
+    parser.add_argument('--windowslice', action='store_true', help='Apply window slice augmentation')
+    parser.add_argument('--windowwarp', action='store_true', help='Apply window warp augmentation')
+    parser.add_argument('--spawner', action='store_true', help='Apply spawner augmentation')
+    parser.add_argument('--dtwwarp', action='store_true', help='Apply DTW-based warp augmentation')
+    parser.add_argument('--shapedtwwarp', action='store_true', help='Apply shape DTW warp augmentation')
+    parser.add_argument('--wdba', action='store_true', help='Apply WDBA augmentation')
+    parser.add_argument('--discdtw', action='store_true', help='Apply discriminative DTW augmentation')
+    parser.add_argument('--discsdtw', action='store_true', help='Apply discriminative shape DTW augmentation')
+    parser.add_argument('--tps', action='store_true', help='Apply TPS augmentation')
+
+    # TPS specific parameters
+    parser.add_argument('--stride', type=int, default=0, help='# of patches stride')
+    parser.add_argument('--patch_len', type=int, default=0, help='# of patches')
+    parser.add_argument('--shuffle_rate', type=float, default=0.0, help='shuffle rate')
+    
+    args = parser.parse_args()
+    
+    if args.num_threads > 0:
+        numba.set_num_threads(args.num_threads)
+    
+    if args.list:
+        list_available_datasets(args)
+        sys.exit(0)
+    
+    # Run on all datasets 
+    if args.all:
+        print(f"Running MultiRocket on all available datasets")
+        print(f"Using {args.num_features} features and {args.iterations} iterations")
+        if args.use_augmentation:
+            print(f"Using data augmentation with ratio {args.augmentation_ratio}")
+        run_all_datasets(args)
+        sys.exit(0)
+    
+    # Run on specific dataset
+    print(f"Running MultiRocket on {args.problem} dataset")
+    print(f"Using {args.num_features} features and {args.iterations} iterations")
+    if args.use_augmentation:
+        print(f"Using data augmentation with ratio {args.augmentation_ratio}")
+    
+    run_multirocket_experiment(args)

time_series_classification/MultiRocket/multirocket/__init__.py

@@ -0,0 +1,5 @@
+__version__ = "0.0.2"
+
+
+def get_module_version():
+    return __version__

time_series_classification/MultiRocket/multirocket/logistic_regression.py

@@ -0,0 +1,208 @@
+# Modified from MultiRocket (https://github.com/ChangWeiTan/MultiRocket)
+# Copyright (C) 2025 Jafar Bakhshaliyev
+# Licensed under GNU General Public License v3.0
+
+import copy
+
+import numpy as np
+import torch
+import torch.nn.functional
+from sklearn.model_selection import train_test_split
+from sklearn.preprocessing import StandardScaler
+from torch.utils.data import TensorDataset, DataLoader
+from tqdm import tqdm
+
+
+class LogisticRegression:
+
+    def __init__(
+            self,
+            num_features,
+            max_epochs=500,
+            minibatch_size=256,
+            validation_size=2 ** 11,
+            learning_rate=1e-4,
+            patience_lr=5,  # 50 minibatches
+            patience=10,  # 100 minibatches
+            device=torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    ):
+        self.name = "LogisticRegression"
+        self.args = {
+            "num_features": num_features,
+            "validation_size": validation_size,
+            "minibatch_size": minibatch_size,
+            "lr": learning_rate,
+            "max_epochs": max_epochs,
+            "patience_lr": patience_lr,
+            "patience": patience,
+        }
+
+        self.model = None
+        self.device = device
+        self.classes = None
+        self.scaler = None
+        self.num_classes = None
+
+    def fit(self, x_train, y_train):
+        self.classes = np.unique(y_train)
+        self.num_classes = len(self.classes)
+
+        num_outputs = self.num_classes if self.num_classes > 2 else 1
+        train_steps = int(x_train.shape[0] / self.args["minibatch_size"])
+
+        self.scaler = StandardScaler()
+        x_train = self.scaler.fit_transform(x_train)
+
+        model = torch.nn.Sequential(torch.nn.Linear(self.args["num_features"], num_outputs)).to(self.device)
+
+        if num_outputs == 1:
+            loss_function = torch.nn.BCEWithLogitsLoss()
+        else:
+            loss_function = torch.nn.CrossEntropyLoss()
+        optimizer = torch.optim.Adam(model.parameters(), lr=self.args["lr"])
+        scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(
+            optimizer,
+            factor=0.5,
+            min_lr=1e-8,
+            patience=self.args["patience_lr"]
+        )
+
+        training_size = x_train.shape[0]
+        if self.args["validation_size"] < training_size:
+            x_training, x_validation, y_training, y_validation = train_test_split(
+                x_train, y_train,
+                test_size=self.args["validation_size"],
+                stratify=y_train
+            )
+
+           
+            if num_outputs == 1:
+                train_data = TensorDataset(
+                    torch.tensor(x_training, dtype=torch.float32).to(self.device),
+                    torch.tensor(y_training, dtype=torch.float32).to(self.device)
+                )
+                val_data = TensorDataset(
+                    torch.tensor(x_validation, dtype=torch.float32).to(self.device),
+                    torch.tensor(y_validation, dtype=torch.float32).to(self.device)
+                )
+            else:
+                train_data = TensorDataset(
+                    torch.tensor(x_training, dtype=torch.float32).to(self.device),
+                    torch.tensor(y_training, dtype=torch.long).to(self.device)
+                )
+                val_data = TensorDataset(
+                    torch.tensor(x_validation, dtype=torch.float32).to(self.device),
+                    torch.tensor(y_validation, dtype=torch.long).to(self.device)
+                )
+                
+            train_dataloader = DataLoader(train_data, shuffle=True, batch_size=self.args["minibatch_size"])
+            val_dataloader = DataLoader(val_data, batch_size=self.args["minibatch_size"])
+        else:
+            if num_outputs == 1:
+                train_data = TensorDataset(
+                    torch.tensor(x_train, dtype=torch.float32).to(self.device),
+                    torch.tensor(y_train, dtype=torch.float32).to(self.device)
+                )
+            else:
+                train_data = TensorDataset(
+                    torch.tensor(x_train, dtype=torch.float32).to(self.device),
+                    torch.tensor(y_train, dtype=torch.long).to(self.device)
+                )
+                
+            train_dataloader = DataLoader(train_data, shuffle=True, batch_size=self.args["minibatch_size"])
+            val_dataloader = None
+
+        best_loss = np.inf
+        best_model = None
+        stall_count = 0
+        stop = False
+
+        for epoch in range(self.args["max_epochs"]):
+            if epoch > 0 and stop:
+                break
+            model.train()
+
+            # loop over the training set
+            total_train_loss = 0
+            steps = 0
+            for i, data in tqdm(enumerate(train_dataloader), desc=f"epoch: {epoch}", total=train_steps):
+                x, y = data
+
+                y_hat = model(x)
+                if num_outputs == 1:
+                    y_hat = y_hat.squeeze()
+                    loss = loss_function(y_hat, y)
+                else:
+                    loss = loss_function(y_hat, y)
+
+                optimizer.zero_grad()
+                loss.backward()
+                optimizer.step()
+                total_train_loss += loss
+                steps += 1
+
+            total_train_loss = total_train_loss.cpu().detach().numpy() / steps
+
+            if val_dataloader is not None:
+                total_val_loss = 0
+                # switch off autograd for evaluation
+                with torch.no_grad():
+                    # set the model in evaluation mode
+                    model.eval()
+                    for i, data in enumerate(val_dataloader):
+                        x, y = data
+
+                        y_hat = model(x)
+                        if num_outputs == 1:
+                            y_hat = y_hat.squeeze()
+                            total_val_loss += loss_function(y_hat, y)
+                        else:
+                            total_val_loss += loss_function(y_hat, y)
+                total_val_loss = total_val_loss.cpu().detach().numpy() / steps
+                scheduler.step(total_val_loss)
+
+                if total_val_loss >= best_loss:
+                    stall_count += 1
+                    if stall_count >= self.args["patience"]:
+                        stop = True
+                        print(f"\n<Stopped at Epoch {epoch + 1}>")
+                else:
+                    best_loss = total_val_loss
+                    best_model = copy.deepcopy(model)
+                    if not stop:
+                        stall_count = 0
+            else:
+                scheduler.step(total_train_loss)
+                if total_train_loss >= best_loss:
+                    stall_count += 1
+                    if stall_count >= self.args["patience"]:
+                        stop = True
+                        print(f"\n<Stopped at Epoch {epoch + 1}>")
+                else:
+                    best_loss = total_train_loss
+                    best_model = copy.deepcopy(model)
+                    if not stop:
+                        stall_count = 0
+
+        self.model = best_model
+        return self.model
+
+    def predict(self, x):
+        x = self.scaler.transform(x)
+
+        with torch.no_grad():
+            # set the model in evaluation mode
+            self.model.eval()
+
+            y_hat = self.model(torch.tensor(x, dtype=torch.float32).to(self.device))
+
+            if self.num_classes > 2:
+                pred_indices = np.argmax(y_hat.cpu().detach().numpy(), axis=1)
+                predictions = self.classes[pred_indices]
+            else:
+                y_hat = y_hat.squeeze()
+                sigmoid_output = torch.sigmoid(y_hat).cpu().detach().numpy()
+                binary_predictions = np.round(sigmoid_output).astype(int)
+                predictions = np.array([self.classes[int(p)] for p in binary_predictions])
+
+            return predictions

time_series_classification/MultiRocket/multirocket/multirocket.py

@@ -0,0 +1,558 @@
+# Chang Wei Tan, Angus Dempster, Christoph Bergmeir, Geoffrey I Webb
+#
+# MultiRocket: Multiple pooling operators and transformations for fast and effective time series classification
+# https://arxiv.org/abs/2102.00457
+
+import time
+
+import numpy as np
+from numba import njit, prange
+from sklearn.linear_model import RidgeClassifierCV
+from sklearn.pipeline import make_pipeline
+from sklearn.preprocessing import StandardScaler
+
+from multirocket.logistic_regression import LogisticRegression
+
+
+@njit("float32[:](float64[:,:],int32[:],int32[:],float32[:])",
+      fastmath=True, parallel=False, cache=True)
+def _fit_biases(X, dilations, num_features_per_dilation, quantiles):
+    num_examples, input_length = X.shape
+
+    # equivalent to:
+    # >>> from itertools import combinations
+    # >>> indices = np.array([_ for _ in combinations(np.arange(9), 3)], dtype = np.int32)
+    indices = np.array((
+        0, 1, 2, 0, 1, 3, 0, 1, 4, 0, 1, 5, 0, 1, 6, 0, 1, 7, 0, 1, 8,
+        0, 2, 3, 0, 2, 4, 0, 2, 5, 0, 2, 6, 0, 2, 7, 0, 2, 8, 0, 3, 4,
+        0, 3, 5, 0, 3, 6, 0, 3, 7, 0, 3, 8, 0, 4, 5, 0, 4, 6, 0, 4, 7,
+        0, 4, 8, 0, 5, 6, 0, 5, 7, 0, 5, 8, 0, 6, 7, 0, 6, 8, 0, 7, 8,
+        1, 2, 3, 1, 2, 4, 1, 2, 5, 1, 2, 6, 1, 2, 7, 1, 2, 8, 1, 3, 4,
+        1, 3, 5, 1, 3, 6, 1, 3, 7, 1, 3, 8, 1, 4, 5, 1, 4, 6, 1, 4, 7,
+        1, 4, 8, 1, 5, 6, 1, 5, 7, 1, 5, 8, 1, 6, 7, 1, 6, 8, 1, 7, 8,
+        2, 3, 4, 2, 3, 5, 2, 3, 6, 2, 3, 7, 2, 3, 8, 2, 4, 5, 2, 4, 6,
+        2, 4, 7, 2, 4, 8, 2, 5, 6, 2, 5, 7, 2, 5, 8, 2, 6, 7, 2, 6, 8,
+        2, 7, 8, 3, 4, 5, 3, 4, 6, 3, 4, 7, 3, 4, 8, 3, 5, 6, 3, 5, 7,
+        3, 5, 8, 3, 6, 7, 3, 6, 8, 3, 7, 8, 4, 5, 6, 4, 5, 7, 4, 5, 8,
+        4, 6, 7, 4, 6, 8, 4, 7, 8, 5, 6, 7, 5, 6, 8, 5, 7, 8, 6, 7, 8
+    ), dtype=np.int32).reshape(84, 3)
+
+    num_kernels = len(indices)
+    num_dilations = len(dilations)
+
+    num_features = num_kernels * np.sum(num_features_per_dilation)
+
+    biases = np.zeros(num_features, dtype=np.float32)
+
+    feature_index_start = 0
+
+    for dilation_index in range(num_dilations):
+
+        dilation = dilations[dilation_index]
+        padding = ((9 - 1) * dilation) // 2
+
+        num_features_this_dilation = num_features_per_dilation[dilation_index]
+
+        for kernel_index in range(num_kernels):
+
+            feature_index_end = feature_index_start + num_features_this_dilation
+
+            _X = X[np.random.randint(num_examples)]
+
+            A = -_X  # A = alpha * X = -X
+            G = _X + _X + _X  # G = gamma * X = 3X
+
+            C_alpha = np.zeros(input_length, dtype=np.float32)
+            C_alpha[:] = A
+
+            C_gamma = np.zeros((9, input_length), dtype=np.float32)
+            C_gamma[9 // 2] = G
+
+            start = dilation
+            end = input_length - padding
+
+            for gamma_index in range(9 // 2):
+                C_alpha[-end:] = C_alpha[-end:] + A[:end]
+                C_gamma[gamma_index, -end:] = G[:end]
+
+                end += dilation
+
+            for gamma_index in range(9 // 2 + 1, 9):
+                C_alpha[:-start] = C_alpha[:-start] + A[start:]
+                C_gamma[gamma_index, :-start] = G[start:]
+
+                start += dilation
+
+            index_0, index_1, index_2 = indices[kernel_index]
+
+            C = C_alpha + C_gamma[index_0] + C_gamma[index_1] + C_gamma[index_2]
+
+            biases[feature_index_start:feature_index_end] = np.quantile(C, quantiles[
+                                                                           feature_index_start:feature_index_end])
+
+            feature_index_start = feature_index_end
+
+    return biases
+
+
+def _fit_dilations(input_length, num_features, max_dilations_per_kernel):
+    num_kernels = 84
+
+    num_features_per_kernel = num_features // num_kernels
+    true_max_dilations_per_kernel = min(num_features_per_kernel, max_dilations_per_kernel)
+    multiplier = num_features_per_kernel / true_max_dilations_per_kernel
+
+    max_exponent = np.log2((input_length - 1) / (9 - 1))
+    dilations, num_features_per_dilation = \
+        np.unique(np.logspace(0, max_exponent, true_max_dilations_per_kernel, base=2).astype(np.int32),
+                  return_counts=True)
+    num_features_per_dilation = (num_features_per_dilation * multiplier).astype(np.int32)  # this is a vector
+
+    remainder = num_features_per_kernel - np.sum(num_features_per_dilation)
+    i = 0
+    while remainder > 0:
+        num_features_per_dilation[i] += 1
+        remainder -= 1
+        i = (i + 1) % len(num_features_per_dilation)
+
+    return dilations, num_features_per_dilation
+
+
+# low-discrepancy sequence to assign quantiles to kernel/dilation combinations
+def _quantiles(n):
+    return np.array([(_ * ((np.sqrt(5) + 1) / 2)) % 1 for _ in range(1, n + 1)], dtype=np.float32)
+
+
+def fit(X, num_features=10_000, max_dilations_per_kernel=32):
+    _, input_length = X.shape
+
+    num_kernels = 84
+
+    dilations, num_features_per_dilation = _fit_dilations(input_length, num_features, max_dilations_per_kernel)
+
+    num_features_per_kernel = np.sum(num_features_per_dilation)
+
+    quantiles = _quantiles(num_kernels * num_features_per_kernel)
+
+    biases = _fit_biases(X, dilations, num_features_per_dilation, quantiles)
+
+    return dilations, num_features_per_dilation, biases
+
+
+@njit(
+    "float32[:,:](float64[:,:],float64[:,:],Tuple((int32[:],int32[:],float32[:])),Tuple((int32[:],int32[:],float32[:])),int32)",
+    fastmath=True, parallel=True, cache=True)
+def transform(X, X1, parameters, parameters1, n_features_per_kernel=4):
+    num_examples, input_length = X.shape
+
+    dilations, num_features_per_dilation, biases = parameters
+    dilations1, num_features_per_dilation1, biases1 = parameters1
+
+    # equivalent to:
+    # >>> from itertools import combinations
+    # >>> indices = np.array([_ for _ in combinations(np.arange(9), 3)], dtype = np.int32)
+    indices = np.array((
+        0, 1, 2, 0, 1, 3, 0, 1, 4, 0, 1, 5, 0, 1, 6, 0, 1, 7, 0, 1, 8,
+        0, 2, 3, 0, 2, 4, 0, 2, 5, 0, 2, 6, 0, 2, 7, 0, 2, 8, 0, 3, 4,
+        0, 3, 5, 0, 3, 6, 0, 3, 7, 0, 3, 8, 0, 4, 5, 0, 4, 6, 0, 4, 7,
+        0, 4, 8, 0, 5, 6, 0, 5, 7, 0, 5, 8, 0, 6, 7, 0, 6, 8, 0, 7, 8,
+        1, 2, 3, 1, 2, 4, 1, 2, 5, 1, 2, 6, 1, 2, 7, 1, 2, 8, 1, 3, 4,
+        1, 3, 5, 1, 3, 6, 1, 3, 7, 1, 3, 8, 1, 4, 5, 1, 4, 6, 1, 4, 7,
+        1, 4, 8, 1, 5, 6, 1, 5, 7, 1, 5, 8, 1, 6, 7, 1, 6, 8, 1, 7, 8,
+        2, 3, 4, 2, 3, 5, 2, 3, 6, 2, 3, 7, 2, 3, 8, 2, 4, 5, 2, 4, 6,
+        2, 4, 7, 2, 4, 8, 2, 5, 6, 2, 5, 7, 2, 5, 8, 2, 6, 7, 2, 6, 8,
+        2, 7, 8, 3, 4, 5, 3, 4, 6, 3, 4, 7, 3, 4, 8, 3, 5, 6, 3, 5, 7,
+        3, 5, 8, 3, 6, 7, 3, 6, 8, 3, 7, 8, 4, 5, 6, 4, 5, 7, 4, 5, 8,
+        4, 6, 7, 4, 6, 8, 4, 7, 8, 5, 6, 7, 5, 6, 8, 5, 7, 8, 6, 7, 8
+    ), dtype=np.int32).reshape(84, 3)
+
+    num_kernels = len(indices)
+    num_dilations = len(dilations)
+    num_dilations1 = len(dilations1)
+
+    num_features = num_kernels * np.sum(num_features_per_dilation)
+    num_features1 = num_kernels * np.sum(num_features_per_dilation1)
+
+    features = np.zeros((num_examples, (num_features + num_features1) * n_features_per_kernel), dtype=np.float32)
+    n_features_per_transform = np.int64(features.shape[1] / 2)
+
+    for example_index in prange(num_examples):
+
+        _X = X[example_index]
+
+        A = -_X  # A = alpha * X = -X
+        G = _X + _X + _X  # G = gamma * X = 3X
+
+        # Base series
+        feature_index_start = 0
+
+        for dilation_index in range(num_dilations):
+
+            _padding0 = dilation_index % 2
+
+            dilation = dilations[dilation_index]
+            padding = ((9 - 1) * dilation) // 2
+
+            num_features_this_dilation = num_features_per_dilation[dilation_index]
+
+            C_alpha = np.zeros(input_length, dtype=np.float32)
+            C_alpha[:] = A
+
+            C_gamma = np.zeros((9, input_length), dtype=np.float32)
+            C_gamma[9 // 2] = G
+
+            start = dilation
+            end = input_length - padding
+
+            for gamma_index in range(9 // 2):
+                C_alpha[-end:] = C_alpha[-end:] + A[:end]
+                C_gamma[gamma_index, -end:] = G[:end]
+
+                end += dilation
+
+            for gamma_index in range(9 // 2 + 1, 9):
+                C_alpha[:-start] = C_alpha[:-start] + A[start:]
+                C_gamma[gamma_index, :-start] = G[start:]
+
+                start += dilation
+
+            for kernel_index in range(num_kernels):
+
+                feature_index_end = feature_index_start + num_features_this_dilation
+
+                _padding1 = (_padding0 + kernel_index) % 2
+
+                index_0, index_1, index_2 = indices[kernel_index]
+
+                C = C_alpha + \
+                    C_gamma[index_0] + \
+                    C_gamma[index_1] + \
+                    C_gamma[index_2]
+
+                if _padding1 == 0:
+                    for feature_count in range(num_features_this_dilation):
+                        feature_index = feature_index_start + feature_count
+                        _bias = biases[feature_index]
+
+                        ppv = 0
+                        last_val = 0
+                        max_stretch = 0.0
+                        mean_index = 0
+                        mean = 0
+
+                        for j in range(C.shape[0]):
+                            if C[j] > _bias:
+                                ppv += 1
+                                mean_index += j
+                                mean += C[j] + _bias
+                            elif C[j] < _bias:
+                                stretch = j - last_val
+                                if stretch > max_stretch:
+                                    max_stretch = stretch
+                                last_val = j
+                        stretch = C.shape[0] - 1 - last_val
+                        if stretch > max_stretch:
+                            max_stretch = stretch
+
+                        end = feature_index
+                        features[example_index, end] = ppv / C.shape[0]
+                        end = end + num_features
+                        features[example_index, end] = max_stretch
+                        end = end + num_features
+                        features[example_index, end] = mean / ppv if ppv > 0 else 0
+                        end = end + num_features
+                        features[example_index, end] = mean_index / ppv if ppv > 0 else -1
+                else:
+                    _c = C[padding:-padding]
+
+                    for feature_count in range(num_features_this_dilation):
+                        feature_index = feature_index_start + feature_count
+                        _bias = biases[feature_index]
+
+                        ppv = 0
+                        last_val = 0
+                        max_stretch = 0.0
+                        mean_index = 0
+                        mean = 0
+
+                        for j in range(_c.shape[0]):
+                            if _c[j] > _bias:
+                                ppv += 1
+                                mean_index += j
+                                mean += _c[j] + _bias
+                            elif _c[j] < _bias:
+                                stretch = j - last_val
+                                if stretch > max_stretch:
+                                    max_stretch = stretch
+                                last_val = j
+                        stretch = _c.shape[0] - 1 - last_val
+                        if stretch > max_stretch:
+                            max_stretch = stretch
+
+                        end = feature_index
+                        features[example_index, end] = ppv / _c.shape[0]
+                        end = end + num_features
+                        features[example_index, end] = max_stretch
+                        end = end + num_features
+                        features[example_index, end] = mean / ppv if ppv > 0 else 0
+                        end = end + num_features
+                        features[example_index, end] = mean_index / ppv if ppv > 0 else -1
+
+                feature_index_start = feature_index_end
+
+        # First order difference
+        _X1 = X1[example_index]
+        A1 = -_X1  # A = alpha * X = -X
+        G1 = _X1 + _X1 + _X1  # G = gamma * X = 3X
+
+        feature_index_start = 0
+
+        for dilation_index in range(num_dilations1):
+
+            _padding0 = dilation_index % 2
+
+            dilation = dilations1[dilation_index]
+            padding = ((9 - 1) * dilation) // 2
+
+            num_features_this_dilation = num_features_per_dilation1[dilation_index]
+
+            C_alpha = np.zeros(input_length - 1, dtype=np.float32)
+            C_alpha[:] = A1
+
+            C_gamma = np.zeros((9, input_length - 1), dtype=np.float32)
+            C_gamma[9 // 2] = G1
+
+            start = dilation
+            end = input_length - padding
+
+            for gamma_index in range(9 // 2):
+                C_alpha[-end:] = C_alpha[-end:] + A1[:end]
+                C_gamma[gamma_index, -end:] = G1[:end]
+
+                end += dilation
+
+            for gamma_index in range(9 // 2 + 1, 9):
+                C_alpha[:-start] = C_alpha[:-start] + A1[start:]
+                C_gamma[gamma_index, :-start] = G1[start:]
+
+                start += dilation
+
+            for kernel_index in range(num_kernels):
+
+                feature_index_end = feature_index_start + num_features_this_dilation
+
+                _padding1 = (_padding0 + kernel_index) % 2
+
+                index_0, index_1, index_2 = indices[kernel_index]
+
+                C = C_alpha + \
+                    C_gamma[index_0] + \
+                    C_gamma[index_1] + \
+                    C_gamma[index_2]
+
+                if _padding1 == 0:
+                    for feature_count in range(num_features_this_dilation):
+                        feature_index = feature_index_start + feature_count
+                        _bias = biases1[feature_index]
+
+                        ppv = 0
+                        last_val = 0
+                        max_stretch = 0.0
+                        mean_index = 0
+                        mean = 0
+
+                        for j in range(C.shape[0]):
+                            if C[j] > _bias:
+                                ppv += 1
+                                mean_index += j
+                                mean += C[j] + _bias
+                            elif C[j] < _bias:
+                                stretch = j - last_val
+                                if stretch > max_stretch:
+                                    max_stretch = stretch
+                                last_val = j
+                        stretch = C.shape[0] - 1 - last_val
+                        if stretch > max_stretch:
+                            max_stretch = stretch
+
+                        end = feature_index + n_features_per_transform
+                        features[example_index, end] = ppv / C.shape[0]
+                        end = end + num_features
+                        features[example_index, end] = max_stretch
+                        end = end + num_features
+                        features[example_index, end] = mean / ppv if ppv > 0 else 0
+                        end = end + num_features
+                        features[example_index, end] = mean_index / ppv if ppv > 0 else -1
+                else:
+                    _c = C[padding:-padding]
+
+                    for feature_count in range(num_features_this_dilation):
+                        feature_index = feature_index_start + feature_count
+                        _bias = biases1[feature_index]
+
+                        ppv = 0
+                        last_val = 0
+                        max_stretch = 0.0
+                        mean_index = 0
+                        mean = 0
+
+                        for j in range(_c.shape[0]):
+                            if _c[j] > _bias:
+                                ppv += 1
+                                mean_index += j
+                                mean += _c[j] + _bias
+                            elif _c[j] < _bias:
+                                stretch = j - last_val
+                                if stretch > max_stretch:
+                                    max_stretch = stretch
+                                last_val = j
+                        stretch = _c.shape[0] - 1 - last_val
+                        if stretch > max_stretch:
+                            max_stretch = stretch
+
+                        end = feature_index + n_features_per_transform
+                        features[example_index, end] = ppv / _c.shape[0]
+                        end = end + num_features
+                        features[example_index, end] = max_stretch
+                        end = end + num_features
+                        features[example_index, end] = mean / ppv if ppv > 0 else 0
+                        end = end + num_features
+                        features[example_index, end] = mean_index / ppv if ppv > 0 else -1
+
+                feature_index_start = feature_index_end
+
+    return features
+
+
+class MultiRocket:
+
+    def __init__(
+            self,
+            num_features=50000,
+            classifier="ridge",
+            verbose=0
+    ):
+        self.name = "MultiRocket"
+
+        self.base_parameters = None
+        self.diff1_parameters = None
+
+        self.n_features_per_kernel = 4
+        self.num_features = num_features / 2  # 1 per transformation
+        self.num_kernels = int(self.num_features / self.n_features_per_kernel)
+
+        if verbose > 1:
+            print('[{}] Creating {} with {} kernels'.format(self.name, self.name, self.num_kernels))
+
+        self.clf = classifier
+        self.classifier = None
+        self.train_duration = 0
+        self.test_duration = 0
+        self.generate_kernel_duration = 0
+        self.train_transforms_duration = 0
+        self.test_transforms_duration = 0
+        self.apply_kernel_on_train_duration = 0
+        self.apply_kernel_on_test_duration = 0
+
+        self.verbose = verbose
+
+    def fit(self, x_train, y_train, predict_on_train=True):
+        if self.verbose > 1:
+            print('[{}] Training with training set of {}'.format(self.name, x_train.shape))
+
+        self.generate_kernel_duration = 0
+        self.apply_kernel_on_train_duration = 0
+        self.train_transforms_duration = 0
+
+        start_time = time.perf_counter()
+
+        _start_time = time.perf_counter()
+        xx = np.diff(x_train, 1)
+        self.train_transforms_duration += time.perf_counter() - _start_time
+
+        _start_time = time.perf_counter()
+        self.base_parameters = fit(
+            x_train,
+            num_features=self.num_kernels
+        )
+        self.diff1_parameters = fit(
+            xx,
+            num_features=self.num_kernels
+        )
+        self.generate_kernel_duration += time.perf_counter() - _start_time
+
+        _start_time = time.perf_counter()
+        x_train_transform = transform(
+            x_train, xx,
+            self.base_parameters, self.diff1_parameters,
+            self.n_features_per_kernel
+        )
+        self.apply_kernel_on_train_duration += time.perf_counter() - _start_time
+
+        x_train_transform = np.nan_to_num(x_train_transform)
+
+        elapsed_time = time.perf_counter() - start_time
+        if self.verbose > 1:
+            print('[{}] Kernels applied!, took {}s'.format(self.name, elapsed_time))
+            print('[{}] Transformed Shape {}'.format(self.name, x_train_transform.shape))
+
+        if self.verbose > 1:
+            print('[{}] Training'.format(self.name))
+
+        if self.clf.lower() == "ridge":
+            self.classifier = make_pipeline(
+                StandardScaler(),
+                RidgeClassifierCV(
+                    alphas=np.logspace(-3, 3, 10),
+                    normalize=False
+                )
+            )
+        else:
+            self.classifier = LogisticRegression(
+                num_features=x_train_transform.shape[1],
+                max_epochs=200,
+            )
+        _start_time = time.perf_counter()
+        self.classifier.fit(x_train_transform, y_train)
+        self.train_duration = time.perf_counter() - _start_time
+
+        if self.verbose > 1:
+            print('[{}] Training done!, took {:.3f}s'.format(self.name, self.train_duration))
+        if predict_on_train:
+            yhat = self.classifier.predict(x_train_transform)
+        else:
+            yhat = None
+
+        return yhat
+
+    def predict(self, x):
+        if self.verbose > 1:
+            print('[{}] Predicting'.format(self.name))
+
+        self.apply_kernel_on_test_duration = 0
+        self.test_transforms_duration = 0
+
+        _start_time = time.perf_counter()
+        xx = np.diff(x, 1)
+        self.test_transforms_duration += time.perf_counter() - _start_time
+
+        _start_time = time.perf_counter()
+        x_transform = transform(
+            x, xx,
+            self.base_parameters, self.diff1_parameters,
+            self.n_features_per_kernel
+        )
+        self.apply_kernel_on_test_duration += time.perf_counter() - _start_time
+
+        x_transform = np.nan_to_num(x_transform)
+        if self.verbose > 1:
+            print('Kernels applied!, took {:.3f}s. Transformed shape: {}.'.format(self.apply_kernel_on_test_duration,
+                                                                                  x_transform.shape))
+
+        start_time = time.perf_counter()
+        yhat = self.classifier.predict(x_transform)
+        self.test_duration = time.perf_counter() - start_time
+        if self.verbose > 1:
+            print("[{}] Predicting completed, took {:.3f}s".format(self.name, self.test_duration))
+
+        return yhat

time_series_classification/MultiRocket/multirocket/multirocket_multivariate.py

@@ -0,0 +1,622 @@
+# Chang Wei Tan, Angus Dempster, Christoph Bergmeir, Geoffrey I Webb
+#
+# MultiRocket: Multiple pooling operators and transformations for fast and effective time series classification
+# https://arxiv.org/abs/2102.00457
+
+import time
+
+import numpy as np
+from numba import njit, prange
+from sklearn.linear_model import RidgeClassifierCV
+from sklearn.pipeline import make_pipeline
+from sklearn.preprocessing import StandardScaler
+
+from multirocket.logistic_regression import LogisticRegression
+
+
+@njit("float32[:](float64[:,:,:],int32[:],int32[:],int32[:],int32[:],float32[:])",
+      fastmath=True, parallel=False, cache=True)
+def _fit_biases(X, num_channels_per_combination, channel_indices, dilations, num_features_per_dilation, quantiles):
+    num_examples, num_channels, input_length = X.shape
+
+    # equivalent to:
+    # >>> from itertools import combinations
+    # >>> indices = np.array([_ for _ in combinations(np.arange(9), 3)], dtype = np.int32)
+    indices = np.array((
+        0, 1, 2, 0, 1, 3, 0, 1, 4, 0, 1, 5, 0, 1, 6, 0, 1, 7, 0, 1, 8,
+        0, 2, 3, 0, 2, 4, 0, 2, 5, 0, 2, 6, 0, 2, 7, 0, 2, 8, 0, 3, 4,
+        0, 3, 5, 0, 3, 6, 0, 3, 7, 0, 3, 8, 0, 4, 5, 0, 4, 6, 0, 4, 7,
+        0, 4, 8, 0, 5, 6, 0, 5, 7, 0, 5, 8, 0, 6, 7, 0, 6, 8, 0, 7, 8,
+        1, 2, 3, 1, 2, 4, 1, 2, 5, 1, 2, 6, 1, 2, 7, 1, 2, 8, 1, 3, 4,
+        1, 3, 5, 1, 3, 6, 1, 3, 7, 1, 3, 8, 1, 4, 5, 1, 4, 6, 1, 4, 7,
+        1, 4, 8, 1, 5, 6, 1, 5, 7, 1, 5, 8, 1, 6, 7, 1, 6, 8, 1, 7, 8,
+        2, 3, 4, 2, 3, 5, 2, 3, 6, 2, 3, 7, 2, 3, 8, 2, 4, 5, 2, 4, 6,
+        2, 4, 7, 2, 4, 8, 2, 5, 6, 2, 5, 7, 2, 5, 8, 2, 6, 7, 2, 6, 8,
+        2, 7, 8, 3, 4, 5, 3, 4, 6, 3, 4, 7, 3, 4, 8, 3, 5, 6, 3, 5, 7,
+        3, 5, 8, 3, 6, 7, 3, 6, 8, 3, 7, 8, 4, 5, 6, 4, 5, 7, 4, 5, 8,
+        4, 6, 7, 4, 6, 8, 4, 7, 8, 5, 6, 7, 5, 6, 8, 5, 7, 8, 6, 7, 8
+    ), dtype=np.int32).reshape(84, 3)
+
+    num_kernels = len(indices)
+    num_dilations = len(dilations)
+
+    num_features = num_kernels * np.sum(num_features_per_dilation)
+
+    biases = np.zeros(num_features, dtype=np.float32)
+
+    feature_index_start = 0
+
+    combination_index = 0
+    num_channels_start = 0
+
+    for dilation_index in range(num_dilations):
+
+        dilation = dilations[dilation_index]
+        padding = ((9 - 1) * dilation) // 2
+
+        num_features_this_dilation = num_features_per_dilation[dilation_index]
+
+        for kernel_index in range(num_kernels):
+
+            feature_index_end = feature_index_start + num_features_this_dilation
+
+            num_channels_this_combination = num_channels_per_combination[combination_index]
+
+            num_channels_end = num_channels_start + num_channels_this_combination
+
+            channels_this_combination = channel_indices[num_channels_start:num_channels_end]
+
+            _X = X[np.random.randint(num_examples)][channels_this_combination]
+
+            A = -_X  # A = alpha * X = -X
+            G = _X + _X + _X  # G = gamma * X = 3X
+
+            C_alpha = np.zeros((num_channels_this_combination, input_length), dtype=np.float32)
+            C_alpha[:] = A
+
+            C_gamma = np.zeros((9, num_channels_this_combination, input_length), dtype=np.float32)
+            C_gamma[9 // 2] = G
+
+            start = dilation
+            end = input_length - padding
+
+            for gamma_index in range(9 // 2):
+                C_alpha[:, -end:] = C_alpha[:, -end:] + A[:, :end]
+                C_gamma[gamma_index, :, -end:] = G[:, :end]
+
+                end += dilation
+
+            for gamma_index in range(9 // 2 + 1, 9):
+                C_alpha[:, :-start] = C_alpha[:, :-start] + A[:, start:]
+                C_gamma[gamma_index, :, :-start] = G[:, start:]
+
+                start += dilation
+
+            index_0, index_1, index_2 = indices[kernel_index]
+
+            C = C_alpha + C_gamma[index_0] + C_gamma[index_1] + C_gamma[index_2]
+            C = np.sum(C, axis=0)
+
+            biases[feature_index_start:feature_index_end] = np.quantile(C, quantiles[
+                                                                           feature_index_start:feature_index_end])
+
+            feature_index_start = feature_index_end
+
+            combination_index += 1
+            num_channels_start = num_channels_end
+
+    return biases
+
+
+def _fit_dilations(input_length, num_features, max_dilations_per_kernel):
+    num_kernels = 84
+
+    num_features_per_kernel = num_features // num_kernels
+    true_max_dilations_per_kernel = min(num_features_per_kernel, max_dilations_per_kernel)
+    multiplier = num_features_per_kernel / true_max_dilations_per_kernel
+
+    max_exponent = np.log2((input_length - 1) / (9 - 1))
+    dilations, num_features_per_dilation = \
+        np.unique(np.logspace(0, max_exponent, true_max_dilations_per_kernel, base=2).astype(np.int32),
+                  return_counts=True)
+    num_features_per_dilation = (num_features_per_dilation * multiplier).astype(np.int32)  # this is a vector
+
+    remainder = num_features_per_kernel - np.sum(num_features_per_dilation)
+    i = 0
+    while remainder > 0:
+        num_features_per_dilation[i] += 1
+        remainder -= 1
+        i = (i + 1) % len(num_features_per_dilation)
+
+    return dilations, num_features_per_dilation
+
+
+# low-discrepancy sequence to assign quantiles to kernel/dilation combinations
+def _quantiles(n):
+    return np.array([(_ * ((np.sqrt(5) + 1) / 2)) % 1 for _ in range(1, n + 1)], dtype=np.float32)
+
+
+def fit(X, num_features=10_000, max_dilations_per_kernel=32):
+    _, num_channels, input_length = X.shape
+
+    num_kernels = 84
+
+    dilations, num_features_per_dilation = _fit_dilations(input_length, num_features, max_dilations_per_kernel)
+
+    num_features_per_kernel = np.sum(num_features_per_dilation)
+
+    quantiles = _quantiles(num_kernels * num_features_per_kernel)
+
+    num_dilations = len(dilations)
+    num_combinations = num_kernels * num_dilations
+
+    max_num_channels = min(num_channels, 9)
+    max_exponent = np.log2(max_num_channels + 1)
+
+    num_channels_per_combination = (2 ** np.random.uniform(0, max_exponent, num_combinations)).astype(np.int32)
+
+    channel_indices = np.zeros(num_channels_per_combination.sum(), dtype=np.int32)
+
+    num_channels_start = 0
+    for combination_index in range(num_combinations):
+        num_channels_this_combination = num_channels_per_combination[combination_index]
+        num_channels_end = num_channels_start + num_channels_this_combination
+        channel_indices[num_channels_start:num_channels_end] = np.random.choice(num_channels,
+                                                                                num_channels_this_combination,
+                                                                                replace=False)
+
+        num_channels_start = num_channels_end
+
+    biases = _fit_biases(X, num_channels_per_combination, channel_indices,
+                         dilations, num_features_per_dilation, quantiles)
+
+    return num_channels_per_combination, channel_indices, dilations, num_features_per_dilation, biases
+
+
+@njit(
+    "float32[:,:](float64[:,:,:],float64[:,:,:],Tuple((int32[:],int32[:],int32[:],int32[:],float32[:])),Tuple((int32[:],int32[:],int32[:],int32[:],float32[:])),int32)",
+    fastmath=True, parallel=True, cache=True)
+def transform(X, X1, parameters, parameters1, n_features_per_kernel=4):
+    num_examples, num_channels, input_length = X.shape
+
+    num_channels_per_combination, channel_indices, dilations, num_features_per_dilation, biases = parameters
+    _, _, dilations1, num_features_per_dilation1, biases1 = parameters1
+
+    # equivalent to:
+    # >>> from itertools import combinations
+    # >>> indices = np.array([_ for _ in combinations(np.arange(9), 3)], dtype = np.int32)
+    indices = np.array((
+        0, 1, 2, 0, 1, 3, 0, 1, 4, 0, 1, 5, 0, 1, 6, 0, 1, 7, 0, 1, 8,
+        0, 2, 3, 0, 2, 4, 0, 2, 5, 0, 2, 6, 0, 2, 7, 0, 2, 8, 0, 3, 4,
+        0, 3, 5, 0, 3, 6, 0, 3, 7, 0, 3, 8, 0, 4, 5, 0, 4, 6, 0, 4, 7,
+        0, 4, 8, 0, 5, 6, 0, 5, 7, 0, 5, 8, 0, 6, 7, 0, 6, 8, 0, 7, 8,
+        1, 2, 3, 1, 2, 4, 1, 2, 5, 1, 2, 6, 1, 2, 7, 1, 2, 8, 1, 3, 4,
+        1, 3, 5, 1, 3, 6, 1, 3, 7, 1, 3, 8, 1, 4, 5, 1, 4, 6, 1, 4, 7,
+        1, 4, 8, 1, 5, 6, 1, 5, 7, 1, 5, 8, 1, 6, 7, 1, 6, 8, 1, 7, 8,
+        2, 3, 4, 2, 3, 5, 2, 3, 6, 2, 3, 7, 2, 3, 8, 2, 4, 5, 2, 4, 6,
+        2, 4, 7, 2, 4, 8, 2, 5, 6, 2, 5, 7, 2, 5, 8, 2, 6, 7, 2, 6, 8,
+        2, 7, 8, 3, 4, 5, 3, 4, 6, 3, 4, 7, 3, 4, 8, 3, 5, 6, 3, 5, 7,
+        3, 5, 8, 3, 6, 7, 3, 6, 8, 3, 7, 8, 4, 5, 6, 4, 5, 7, 4, 5, 8,
+        4, 6, 7, 4, 6, 8, 4, 7, 8, 5, 6, 7, 5, 6, 8, 5, 7, 8, 6, 7, 8
+    ), dtype=np.int32).reshape(84, 3)
+
+    num_kernels = len(indices)
+    num_dilations = len(dilations)
+    num_dilations1 = len(dilations1)
+
+    num_features = num_kernels * np.sum(num_features_per_dilation)
+    num_features1 = num_kernels * np.sum(num_features_per_dilation1)
+
+    features = np.zeros((num_examples, (num_features + num_features1) * n_features_per_kernel), dtype=np.float32)
+    n_features_per_transform = np.int64(features.shape[1] / 2)
+
+    for example_index in prange(num_examples):
+
+        _X = X[example_index]
+
+        A = -_X  # A = alpha * X = -X
+        G = _X + _X + _X  # G = gamma * X = 3X
+
+        # Base series
+        feature_index_start = 0
+
+        combination_index = 0
+        num_channels_start = 0
+
+        for dilation_index in range(num_dilations):
+
+            _padding0 = dilation_index % 2
+
+            dilation = dilations[dilation_index]
+            padding = ((9 - 1) * dilation) // 2
+
+            num_features_this_dilation = num_features_per_dilation[dilation_index]
+
+            C_alpha = np.zeros((num_channels, input_length), dtype=np.float32)
+            C_alpha[:] = A
+
+            C_gamma = np.zeros((9, num_channels, input_length), dtype=np.float32)
+            C_gamma[9 // 2] = G
+
+            start = dilation
+            end = input_length - padding
+
+            for gamma_index in range(9 // 2):
+                C_alpha[:, -end:] = C_alpha[:, -end:] + A[:, :end]
+                C_gamma[gamma_index, :, -end:] = G[:, :end]
+
+                end += dilation
+
+            for gamma_index in range(9 // 2 + 1, 9):
+                C_alpha[:, :-start] = C_alpha[:, :-start] + A[:, start:]
+                C_gamma[gamma_index, :, :-start] = G[:, start:]
+
+                start += dilation
+
+            for kernel_index in range(num_kernels):
+
+                feature_index_end = feature_index_start + num_features_this_dilation
+
+                num_channels_this_combination = num_channels_per_combination[combination_index]
+
+                num_channels_end = num_channels_start + num_channels_this_combination
+
+                channels_this_combination = channel_indices[num_channels_start:num_channels_end]
+
+                _padding1 = (_padding0 + kernel_index) % 2
+
+                index_0, index_1, index_2 = indices[kernel_index]
+
+                C = C_alpha[channels_this_combination] + \
+                    C_gamma[index_0][channels_this_combination] + \
+                    C_gamma[index_1][channels_this_combination] + \
+                    C_gamma[index_2][channels_this_combination]
+                C = np.sum(C, axis=0)
+
+                if _padding1 == 0:
+                    for feature_count in range(num_features_this_dilation):
+                        feature_index = feature_index_start + feature_count
+                        _bias = biases[feature_index]
+
+                        ppv = 0
+                        last_val = 0
+                        max_stretch = 0.0
+                        mean_index = 0
+                        mean = 0
+
+                        for j in range(C.shape[0]):
+                            if C[j] > _bias:
+                                ppv += 1
+                                mean_index += j
+                                mean += C[j] + _bias
+                            elif C[j] < _bias:
+                                stretch = j - last_val
+                                if stretch > max_stretch:
+                                    max_stretch = stretch
+                                last_val = j
+                        stretch = C.shape[0] - 1 - last_val
+                        if stretch > max_stretch:
+                            max_stretch = stretch
+
+                        end = feature_index
+                        features[example_index, end] = ppv / C.shape[0]
+                        end = end + num_features
+                        features[example_index, end] = max_stretch
+                        end = end + num_features
+                        features[example_index, end] = mean / ppv if ppv > 0 else 0
+                        end = end + num_features
+                        features[example_index, end] = mean_index / ppv if ppv > 0 else -1
+                else:
+                    _c = C[padding:-padding]
+
+                    for feature_count in range(num_features_this_dilation):
+                        feature_index = feature_index_start + feature_count
+                        _bias = biases[feature_index]
+
+                        ppv = 0
+                        last_val = 0
+                        max_stretch = 0.0
+                        mean_index = 0
+                        mean = 0
+
+                        for j in range(_c.shape[0]):
+                            if _c[j] > _bias:
+                                ppv += 1
+                                mean_index += j
+                                mean += _c[j] + _bias
+                            elif _c[j] < _bias:
+                                stretch = j - last_val
+                                if stretch > max_stretch:
+                                    max_stretch = stretch
+                                last_val = j
+                        stretch = _c.shape[0] - 1 - last_val
+                        if stretch > max_stretch:
+                            max_stretch = stretch
+
+                        end = feature_index
+                        features[example_index, end] = ppv / _c.shape[0]
+                        end = end + num_features
+                        features[example_index, end] = max_stretch
+                        end = end + num_features
+                        features[example_index, end] = mean / ppv if ppv > 0 else 0
+                        end = end + num_features
+                        features[example_index, end] = mean_index / ppv if ppv > 0 else -1
+
+                feature_index_start = feature_index_end
+
+                combination_index += 1
+                num_channels_start = num_channels_end
+
+        # First order difference
+        _X1 = X1[example_index]
+        A1 = -_X1  # A = alpha * X = -X
+        G1 = _X1 + _X1 + _X1  # G = gamma * X = 3X
+
+        feature_index_start = 0
+
+        combination_index = 0
+        num_channels_start = 0
+
+        for dilation_index in range(num_dilations1):
+
+            _padding0 = dilation_index % 2
+
+            dilation = dilations1[dilation_index]
+            padding = ((9 - 1) * dilation) // 2
+
+            num_features_this_dilation = num_features_per_dilation1[dilation_index]
+
+            C_alpha = np.zeros((num_channels, input_length - 1), dtype=np.float32)
+            C_alpha[:] = A1
+
+            C_gamma = np.zeros((9, num_channels, input_length - 1), dtype=np.float32)
+            C_gamma[9 // 2] = G1
+
+            start = dilation
+            end = input_length - padding
+
+            for gamma_index in range(9 // 2):
+                C_alpha[:, -end:] = C_alpha[:, -end:] + A1[:, :end]
+                C_gamma[gamma_index, :, -end:] = G1[:, :end]
+
+                end += dilation
+
+            for gamma_index in range(9 // 2 + 1, 9):
+                C_alpha[:, :-start] = C_alpha[:, :-start] + A1[:, start:]
+                C_gamma[gamma_index, :, :-start] = G1[:, start:]
+
+                start += dilation
+
+            for kernel_index in range(num_kernels):
+
+                feature_index_end = feature_index_start + num_features_this_dilation
+
+                num_channels_this_combination = num_channels_per_combination[combination_index]
+
+                num_channels_end = num_channels_start + num_channels_this_combination
+
+                channels_this_combination = channel_indices[num_channels_start:num_channels_end]
+
+                _padding1 = (_padding0 + kernel_index) % 2
+
+                index_0, index_1, index_2 = indices[kernel_index]
+
+                C = C_alpha[channels_this_combination] + \
+                    C_gamma[index_0][channels_this_combination] + \
+                    C_gamma[index_1][channels_this_combination] + \
+                    C_gamma[index_2][channels_this_combination]
+                C = np.sum(C, axis=0)
+
+                if _padding1 == 0:
+                    for feature_count in range(num_features_this_dilation):
+                        feature_index = feature_index_start + feature_count
+                        _bias = biases1[feature_index]
+
+                        ppv = 0
+                        last_val = 0
+                        max_stretch = 0.0
+                        mean_index = 0
+                        mean = 0
+
+                        for j in range(C.shape[0]):
+                            if C[j] > _bias:
+                                ppv += 1
+                                mean_index += j
+                                mean += C[j] + _bias
+                            elif C[j] < _bias:
+                                stretch = j - last_val
+                                if stretch > max_stretch:
+                                    max_stretch = stretch
+                                last_val = j
+                        stretch = C.shape[0] - 1 - last_val
+                        if stretch > max_stretch:
+                            max_stretch = stretch
+
+                        end = feature_index + n_features_per_transform
+                        features[example_index, end] = ppv / C.shape[0]
+                        end = end + num_features
+                        features[example_index, end] = max_stretch
+                        end = end + num_features
+                        features[example_index, end] = mean / ppv if ppv > 0 else 0
+                        end = end + num_features
+                        features[example_index, end] = mean_index / ppv if ppv > 0 else -1
+                else:
+                    _c = C[padding:-padding]
+
+                    for feature_count in range(num_features_this_dilation):
+                        feature_index = feature_index_start + feature_count
+                        _bias = biases1[feature_index]
+
+                        ppv = 0
+                        last_val = 0
+                        max_stretch = 0.0
+                        mean_index = 0
+                        mean = 0
+
+                        for j in range(_c.shape[0]):
+                            if _c[j] > _bias:
+                                ppv += 1
+                                mean_index += j
+                                mean += _c[j] + _bias
+                            elif _c[j] < _bias:
+                                stretch = j - last_val
+                                if stretch > max_stretch:
+                                    max_stretch = stretch
+                                last_val = j
+                        stretch = _c.shape[0] - 1 - last_val
+                        if stretch > max_stretch:
+                            max_stretch = stretch
+
+                        end = feature_index + n_features_per_transform
+                        features[example_index, end] = ppv / _c.shape[0]
+                        end = end + num_features
+                        features[example_index, end] = max_stretch
+                        end = end + num_features
+                        features[example_index, end] = mean / ppv if ppv > 0 else 0
+                        end = end + num_features
+                        features[example_index, end] = mean_index / ppv if ppv > 0 else -1
+
+                feature_index_start = feature_index_end
+
+    return features
+
+
+class MultiRocket:
+
+    def __init__(
+            self,
+            num_features=50000,
+            classifier="ridge",
+            verbose=0
+    ):
+        self.name = "MultiRocket"
+
+        self.base_parameters = None
+        self.diff1_parameters = None
+
+        self.n_features_per_kernel = 4
+        self.num_features = num_features / 2  # 1 per transformation
+        self.num_kernels = int(self.num_features / self.n_features_per_kernel)
+
+        if verbose > 1:
+            print('[{}] Creating {} with {} kernels'.format(self.name, self.name, self.num_kernels))
+
+        self.clf = classifier
+        self.classifier = None
+        self.train_duration = 0
+        self.test_duration = 0
+        self.generate_kernel_duration = 0
+        self.train_transforms_duration = 0
+        self.test_transforms_duration = 0
+        self.apply_kernel_on_train_duration = 0
+        self.apply_kernel_on_test_duration = 0
+
+        self.verbose = verbose
+
+    def fit(self, x_train, y_train, predict_on_train=True):
+        if self.verbose > 1:
+            print('[{}] Training with training set of {}'.format(self.name, x_train.shape))
+        if x_train.shape[2] < 10:
+            # handling very short series (like PensDigit from the MTSC archive)
+            # series have to be at least a length of 10 (including differencing)
+            _x_train = np.zeros((x_train.shape[0], x_train.shape[1], 10), dtype=x_train.dtype)
+            _x_train[:, :, :x_train.shape[2]] = x_train
+            x_train = _x_train
+            del _x_train
+
+        self.generate_kernel_duration = 0
+        self.apply_kernel_on_train_duration = 0
+        self.train_transforms_duration = 0
+
+        start_time = time.perf_counter()
+
+        _start_time = time.perf_counter()
+        xx = np.diff(x_train, 1)
+        self.train_transforms_duration += time.perf_counter() - _start_time
+
+        _start_time = time.perf_counter()
+        self.base_parameters = fit(
+            x_train,
+            num_features=self.num_kernels
+        )
+        self.diff1_parameters = fit(
+            xx,
+            num_features=self.num_kernels
+        )
+        self.generate_kernel_duration += time.perf_counter() - _start_time
+
+        _start_time = time.perf_counter()
+        x_train_transform = transform(
+            x_train, xx,
+            self.base_parameters, self.diff1_parameters,
+            self.n_features_per_kernel
+        )
+        self.apply_kernel_on_train_duration += time.perf_counter() - _start_time
+
+        x_train_transform = np.nan_to_num(x_train_transform)
+
+        elapsed_time = time.perf_counter() - start_time
+        if self.verbose > 1:
+            print('[{}] Kernels applied!, took {}s'.format(self.name, elapsed_time))
+            print('[{}] Transformed Shape {}'.format(self.name, x_train_transform.shape))
+
+        if self.verbose > 1:
+            print('[{}] Training'.format(self.name))
+
+        if self.clf.lower() == "ridge":
+            self.classifier = make_pipeline(
+                StandardScaler(),
+                RidgeClassifierCV(
+                    alphas=np.logspace(-3, 3, 10),
+                    normalize=False
+                )
+            )
+        else:
+            self.classifier = LogisticRegression(
+                num_features=x_train_transform.shape[1],
+                max_epochs=200,
+            )
+        _start_time = time.perf_counter()
+        self.classifier.fit(x_train_transform, y_train)
+        self.train_duration = time.perf_counter() - _start_time
+
+        if self.verbose > 1:
+            print('[{}] Training done!, took {:.3f}s'.format(self.name, self.train_duration))
+        if predict_on_train:
+            yhat = self.classifier.predict(x_train_transform)
+        else:
+            yhat = None
+
+        return yhat
+
+    def predict(self, x):
+        if self.verbose > 1:
+            print('[{}] Predicting'.format(self.name))
+
+        self.apply_kernel_on_test_duration = 0
+        self.test_transforms_duration = 0
+
+        _start_time = time.perf_counter()
+        xx = np.diff(x, 1)
+        self.test_transforms_duration += time.perf_counter() - _start_time
+
+        _start_time = time.perf_counter()
+        x_transform = transform(
+            x, xx,
+            self.base_parameters, self.diff1_parameters,
+            self.n_features_per_kernel
+        )
+        self.apply_kernel_on_test_duration += time.perf_counter() - _start_time
+
+        x_transform = np.nan_to_num(x_transform)
+        if self.verbose > 1:
+            print('Kernels applied!, took {:.3f}s. Transformed shape: {}.'.format(self.apply_kernel_on_test_duration,
+                                                                                  x_transform.shape))
+
+        start_time = time.perf_counter()
+        yhat = self.classifier.predict(x_transform)
+        self.test_duration = time.perf_counter() - start_time
+        if self.verbose > 1:
+            print("[{}] Predicting completed, took {:.3f}s".format(self.name, self.test_duration))
+
+        return yhat

time_series_classification/MultiRocket/requirements.txt

@@ -0,0 +1,6 @@
+numba==0.50.1
+numpy==1.18.5
+pandas == 1.0.5
+scikit_learn>=0.23.1
+sktime==0.4.3
+torch==1.11.0+cu113

time_series_classification/MultiRocket/scripts/example.sh

@@ -0,0 +1,76 @@
+#!/bin/bash
+#SBATCH --job-name=face_detection
+#SBATCH --output=fd.out
+#SBATCH --error=fd.err
+#SBATCH --partition=CPU
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=40 
+#SBATCH --chdir=/home/bakhshaliyev/classification-aug/MultiRocket # change to the directory
+
+
+echo "Activating environment..."
+source ~/venvs/sktime-env/bin/activate # change to your virtual environment path
+
+echo "Running script..."
+
+
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1
+
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --jitter
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --scaling
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --rotation
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --permutation
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --randompermutation
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --magwarp
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --timewarp
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --windowslice
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --windowwarp
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --spawner
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --dtwwarp
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --shapedtwwarp
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --wdba
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --discdtw
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --discsdtw
+
+
+
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 420 --stride 1 --shuffle_rate 0.8 
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 120 --stride 1 --shuffle_rate 0.7
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 4 --stride 4 --shuffle_rate 0.6
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 64 --stride 2 --shuffle_rate 0.6
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 32 --stride 1 --shuffle_rate 0.6
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 8 --stride 4 --shuffle_rate 1.0
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 6 --stride 4 --shuffle_rate 1.0
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 120 --stride 1 --shuffle_rate 1.0
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 96 --stride 8 --shuffle_rate 0.8
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 48 --stride 36 --shuffle_rate 0.9
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 24 --stride 36 --shuffle_rate 0.8
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 240 --stride 1 --shuffle_rate 0.8
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 240 --stride 1 --shuffle_rate 0.4
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 240 --stride 1 --shuffle_rate 0.2
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 96 --stride 96 --shuffle_rate 1.0
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 96 --stride 96 --shuffle_rate 0.6
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 120 --stride 48 --shuffle_rate 0.8
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 240 --stride 96 --shuffle_rate 1.0
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 12 --stride 96 --shuffle_rate 0.8
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 2 --stride 1 --shuffle_rate 0.4
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 480 --stride 1 --shuffle_rate 0.7
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 900 --stride 1 --shuffle_rate 0.8
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 64 --stride 24 --shuffle_rate 0.8
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 64 --stride 24 --shuffle_rate 1.0
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 148 --stride 24 --shuffle_rate 1.0
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 148 --stride 96 --shuffle_rate 1.0
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 148 --stride 48 --shuffle_rate 1.0
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 180 --stride 12 --shuffle_rate 0.8
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 180 --stride 24 --shuffle_rate 0.8
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 180 --stride 12 --shuffle_rate 0.5
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 152 --stride 96 --shuffle_rate 1.0
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 360 --stride 1 --shuffle_rate 0.1
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 72 --stride 2 --shuffle_rate 0.4
+srun python3 main.py --problem FaceDetection --iter 5 --verbose 1 --use-augmentation --augmentation-ratio 1 --tps --patch_len 120 --stride 1 --shuffle_rate 0.2
+
+
+
+
+echo "Job finished"

time_series_classification/MultiRocket/utils/data_loader.py

@@ -0,0 +1,319 @@
+# Chang Wei Tan, Angus Dempster, Christoph Bergmeir, Geoffrey I Webb
+#
+# MultiRocket: Multiple pooling operators and transformations for fast and effective time series classification
+# https://arxiv.org/abs/2102.00457
+
+import os
+import random
+
+import numpy as np
+import pandas as pd
+from sklearn.preprocessing import StandardScaler
+
+non_109_datasets = ["HandOutlines",
+                    "NonInvasiveFetalECGThorax1",
+                    "NonInvasiveFetalECGThorax2",
+                    "AllGestureWiimoteX",
+                    "AllGestureWiimoteY",
+                    "AllGestureWiimoteZ",
+                    "DodgerLoopDay",
+                    "DodgerLoopGame",
+                    "DodgerLoopWeekend",
+                    "Fungi",
+                    "GestureMidAirD1",
+                    "GestureMidAirD2",
+                    "GestureMidAirD3",
+                    "GesturePebbleZ1",
+                    "GesturePebbleZ2",
+                    "MelbournePedestrian",
+                    "PickupGestureWiimoteZ",
+                    "PLAID",
+                    "ShakeGestureWiimoteZ"]
+
+classification_datasets = ["Adiac",  # 390,391,37,176,0.3887,0.3913 (3),0.3964
+                           "ArrowHead",  # 36,175,3,251,0.2,0.2000 (0),0.2971
+                           "Beef",  # 30,30,5,470,0.3333,0.3333 (0),0.3667
+                           "BeetleFly",  # 20,20,2,512,0.25,0.3000 (7),0.3
+                           "BirdChicken",  # 20,20,2,512,0.45,0.3000 (6),0.25
+                           "Car",  # 60,60,4,577,0.2667,0.2333 (1),0.2667
+                           "CBF",  # 30,900,3,128,0.1478,0.0044 (11),0.0033
+                           "ChlorineConcentration",  # 467,3840,3,166,0.35,0.3500 (0),0.3516
+                           "CinCECGTorso",  # 40,1380,4,1639,0.1029,0.0696 (1),0.3493
+                           "Coffee",  # 28,28,2,286,0,0.0000 (0),0
+                           "Computers",  # 250,250,2,720,0.424,0.3800 (12),0.3
+                           "CricketX",  # 390,390,12,300,0.4231,0.2282 (10),0.2462
+                           "CricketY",  # 390,390,12,300,0.4333,0.2410 (17),0.2564
+                           "CricketZ",  # 390,390,12,300,0.4128,0.2538 (5),0.2462
+                           "DiatomSizeReduction",  # 16,306,4,345,0.0654,0.0654 (0),0.0327
+                           "DistalPhalanxOutlineAgeGroup",  # 400,139,3,80,0.3741,0.3741 (0),0.2302
+                           "DistalPhalanxOutlineCorrect",  # 600,276,2,80,0.2826,0.2754 (1),0.2826
+                           "DistalPhalanxTW",  # 400,139,6,80,0.3669,0.3669 (0),0.4101
+                           "Earthquakes",  # 322,139,2,512,0.2878,0.2734 (6),0.2806
+                           "ECG200",  # 100,100,2,96,0.12,0.1200 (0),0.23
+                           "ECG5000",  # 500,4500,5,140,0.0751,0.0749 (1),0.0756
+                           "ECGFiveDays",  # 23,861,2,136,0.2033,0.2033 (0),0.2323
+                           "ElectricDevices",  # 8926,7711,7,96,0.4492,0.3806 (14),0.3988
+                           "FaceAll",  # 560,1690,14,131,0.2864,0.1917 (3),0.1923
+                           "FaceFour",  # 24,88,4,350,0.2159,0.1136 (2),0.1705
+                           "FacesUCR",  # 200,2050,14,131,0.2307,0.0878 (12),0.0951
+                           "FiftyWords",  # 450,455,50,270,0.3692,0.2418 (6),0.3099
+                           "Fish",  # 175,175,7,463,0.2171,0.1543 (4),0.1771
+                           "FordA",  # 3601,1320,2,500,0.3348,0.3091 (1),0.4455
+                           "FordB",  # 3636,810,2,500,0.3938,0.3926 (1),0.3802
+                           "GunPoint",  # 50,150,2,150,0.0867,0.0867 (0) ,0.0933
+                           "Ham",  # 109,105,2,431,0.4,0.4000 (0),0.5333
+                           "HandOutlines",  # 1000,370,2,2709,0.1378,0.1378 (0),0.1189
+                           "Haptics",  # 155,308,5,1092,0.6299,0.5877 (2),0.6234
+                           "Herring",  # 64,64,2,512,0.4844,0.4688 (5),0.4688
+                           "InlineSkate",  # 100,550,7,1882,0.6582,0.6127 (14),0.6164
+                           "InsectWingbeatSound",  # 220,1980,11,256,0.4384,0.4152 (1),0.6449
+                           "ItalyPowerDemand",  # 67,1029,2,24,0.0447,0.0447 (0),0.0496
+                           "LargeKitchenAppliances",  # 375,375,3,720,0.5067,0.2053 (94),0.2053
+                           "Lightning2",  # 60,61,2,637,0.2459,0.1311 (6),0.1311
+                           "Lightning7",  # 70,73,7,319,0.4247,0.2877 (5),0.274
+                           "Mallat",  # 55,2345,8,1024,0.0857,0.0857 (0),0.0661
+                           "Meat",  # 60,60,3,448,0.0667,0.0667 (0),0.0667
+                           "MedicalImages",  # 381,760,10,99,0.3158,0.2526 (20),0.2632
+                           "MiddlePhalanxOutlineAgeGroup",  # 400,154,3,80,0.4805,0.4805 (0),0.5
+                           "MiddlePhalanxOutlineCorrect",  # 600,291,2,80,0.2337,0.2337 (0),0.3024
+                           "MiddlePhalanxTW",  # 399,154,6,80,0.487,0.4935 (3),0.4935
+                           "MoteStrain",  # 20,1252,2,84,0.1214,0.1342 (1),0.1653
+                           "NonInvasiveFetalECGThorax1",  # 1800,1965,42,750,0.171,0.1893 (1),0.2097
+                           "NonInvasiveFetalECGThorax2",  # 1800,1965,42,750,0.1201,0.1290 (1),0.1354
+                           "OliveOil",  # 30,30,4,570,0.1333,0.1333 (0),0.1667
+                           "OSULeaf",  # 200,242,6,427,0.4793,0.3884 (7),0.4091
+                           "PhalangesOutlinesCorrect",  # 1800,858,2,80,0.2389,0.2389 (0),0.2716
+                           "Phoneme",  # 214,1896,39,1024,0.8908,0.7727 (14),0.7716
+                           "Plane",  # 105,105,7,144,0.0381,0.0000 (5),0
+                           "ProximalPhalanxOutlineAgeGroup",  # 400,205,3,80,0.2146,0.2146 (0),0.1951
+                           "ProximalPhalanxOutlineCorrect",  # 600,291,2,80,0.1924,0.2096 (1),0.2165
+                           "ProximalPhalanxTW",  # 400,205,6,80,0.2927,0.2439 (2),0.2439
+                           "RefrigerationDevices",  # 375,375,3,720,0.6053,0.5600 (8),0.536
+                           "ScreenType",  # 375,375,3,720,0.64,0.5893 (17),0.6027
+                           "ShapeletSim",  # 20,180,2,500,0.4611,0.3000 (3),0.35
+                           "ShapesAll",  # 600,600,60,512,0.2483,0.1980 (4),0.2317
+                           "SmallKitchenAppliances",  # 375,375,3,720,0.6587,0.3280 (15),0.3573
+                           "SonyAIBORobotSurface1",  # 20,601,2,70,0.3045,0.3045 (0),0.2745
+                           "SonyAIBORobotSurface2",  # 27,953,2,65,0.1406,0.1406 (0),0.1689
+                           "StarLightCurves",  # 1000,8236,3,1024,0.1512,0.0947 (16),0.0934
+                           "Strawberry",  # 613,370,2,235,0.0541,0.0541 (0),0.0595
+                           "SwedishLeaf",  # 500,625,15,128,0.2112,0.1536 (2),0.208
+                           "Symbols",  # 25,995,6,398,0.1005,0.0623 (8),0.0503
+                           "SyntheticControl",  # 300,300,6,60,0.12,0.0167 (6),0.0067
+                           "ToeSegmentation1",  # 40,228,2,277,0.3202,0.2500 (8),0.2281
+                           "ToeSegmentation2",  # 36,130,2,343,0.1923,0.0923 (5),0.1615
+                           "Trace",  # 100,100,4,275,0.24,0.0100 (3),0
+                           "TwoLeadECG",  # 23,1139,2,82,0.2529,0.1317 (4),0.0957
+                           "TwoPatterns",  # 1000,4000,4,128,0.0932,0.0015 (4),0
+                           "UWaveGestureLibraryAll",  # 896,3582,8,945,0.0519,0.0343 (4),0.1083
+                           "UWaveGestureLibraryX",  # 896,3582,8,315,0.2607,0.2267 (4),0.2725
+                           "UWaveGestureLibraryY",  # 896,3582,8,315,0.338,0.3009 (4),0.366
+                           "UWaveGestureLibraryZ",  # 896,3582,8,315,0.3504,0.3222 (6),0.3417
+                           "Wafer",  # 1000,6164,2,152,0.0045,0.0045 (1),0.0201
+                           "Wine",  # 57,54,2,234,0.3889,0.3889 (0),0.4259
+                           "WordSynonyms",  # 267,638,25,270,0.3824,0.2618 (9),0.3511
+                           "Worms",  # 181,77,5,900,0.5455,0.4675 (9),0.4156
+                           "WormsTwoClass",  # 181,77,2,900,0.3896,0.4156 (7),0.3766
+                           "Yoga",  # 300,3000,2,426,0.1697,0.1560 (7),0.1637
+                           "ACSF1",  # 100,100,10,1460,0.46,0.3800 (4),0.36
+                           "AllGestureWiimoteX",  # 300,700,10,Vary,0.4843, 0.2829 (14),0.2843
+                           "AllGestureWiimoteY",  # 300,700,10,Vary,0.4314, 0.2700 (9),0.2714
+                           "AllGestureWiimoteZ",  # 300,700,10,Vary,0.5457,0.3486 (11),0.3571
+                           "BME",  # 30,150,3,128,0.1667,0.0200 (4),0.1
+                           "Chinatown",  # 20,345,2,24,0.0464,0.0464 (0),0.0435
+                           "Crop",  # 7200,16800,24,46,0.2883,0.2883 (0),0.3348
+                           "DodgerLoopDay",  # 78,80,7,288,0.45, 0.4125 (1),0.5
+                           "DodgerLoopGame",  # 20,138,2,288,0.1159, 0.0725 (1),0.1232
+                           "DodgerLoopWeekend",  # 20,138,2,288,0.0145, 0.0217 (1),0.0507
+                           "EOGHorizontalSignal",  # 362,362,12,1250,0.5829, 0.5249 (1),0.4972
+                           "EOGVerticalSignal",  # 362,362,12,1250,0.558, 0.5249 (2),0.5525
+                           "EthanolLevel",  # 504,500,4,1751,0.726,0.7180 (1),0.724
+                           "FreezerRegularTrain",  # 150,2850,2,301,0.1951,0.0930 (1),0.1011
+                           "FreezerSmallTrain",  # 28,2850,2,301,0.3302,0.3302 (0),0.2467
+                           "Fungi",  # 18,186,18,201,0.1774,0.1774 (0),0.1613
+                           "GestureMidAirD1",  # 208,130,26,Vary,0.4231, 0.3615 (5),0.4308
+                           "GestureMidAirD2",  # 208,130,26,Vary,0.5077, 0.4000 (6),0.3923
+                           "GestureMidAirD3",  # 208,130,26,Vary,0.6538, 0.6231 (1),0.6769
+                           "GesturePebbleZ1",  # 132,172,6,Vary,0.2674,0.1744 (2),0.2093
+                           "GesturePebbleZ2",  # 146,158,6,Vary,0.3291,0.2215 (6),0.3291
+                           "GunPointAgeSpan",  # 135,316,2,150,0.1013,0.0348 (3),0.0823
+                           "GunPointMaleVersusFemale",  # 135,316,2,150,0.0253,0.0253 (0),0.0032
+                           "GunPointOldVersusYoung",  # 135,316,2,150,0.0476,0.0349 (4),0.1619
+                           "HouseTwenty",  # 40,119,2,2000,0.3361, 0.0588 (33),0.0756
+                           "InsectEPGRegularTrain",  # 62,249,3,601,0.3213,0.1727 (11),0.1285
+                           "InsectEPGSmallTrain",  # 17,249,3,601,0.3373,0.3052 (1),0.2651
+                           "MelbournePedestrian",  # 1200,2450,10,24,0.1518,0.1518 (0),0.2094
+                           "MixedShapesRegularTrain",  # 500,2425,5,1024,0.1027, 0.0911 (4),0.1584
+                           "MixedShapesSmallTrain",  # 100,2425,5,1024,0.1645, 0.1674 (7),0.2202
+                           "PickupGestureWiimoteZ",  # 50,50,10,Vary,0.44,0.3400 (17),0.34
+                           "PigAirwayPressure",  # 104,208,52,2000,0.9423,0.9038 (1),0.8942
+                           "PigArtPressure",  # 104,208,52,2000,0.875,0.8029 (1),0.7548
+                           "PigCVP",  # 104,208,52,2000,0.9183,0.8413 (11),0.8462
+                           "PLAID",  # 537,537,11,Vary,0.4786,0.1862 (3),0.1601
+                           "PowerCons",  # 180,180,2,144,0.0667,0.0778 (3),0.1222
+                           "Rock",  # 20,50,4,2844,0.16, 0.1600 (0),0.4
+                           "SemgHandGenderCh2",  # 300,600,2,1500,0.2383,0.1550 (1),0.1983
+                           "SemgHandMovementCh2",  # 450,450,6,1500,0.6311,0.3622 (1),0.4156
+                           "SemgHandSubjectCh2",  # 450,450,5,1500,0.5956,0.2000 (3),0.2733
+                           "ShakeGestureWiimoteZ",  # 50,50,10,Vary,0.4,0.1600 (6),0.14
+                           "SmoothSubspace",  # 150,150,3,15,0.0933,0.0533 (1),0.1733
+                           "UMD",  # 36,144,3,150,0.2361,0.0278 (6),0.0069
+                           ]
+
+
+def get_classification_datasets_summary(dataset=None, subset="full"):
+    if subset == "109":
+        if os.path.exists("../data/classification_datasets_109.csv"):
+            df = pd.read_csv("../data/classification_datasets_109.csv")
+        else:
+            df = pd.read_csv(os.getcwd() + "/data/classification_datasets_109.csv")
+        df.columns = [x.strip() for x in df.columns]
+        if dataset is None:
+            return df
+    elif subset == "bakeoff":
+        if os.path.exists("../data/classification_datasets_bakeoff.csv"):
+            df = pd.read_csv("../data/classification_datasets_bakeoff.csv")
+        else:
+            df = pd.read_csv(os.getcwd() + "/data/classification_datasets_bakeoff.csv")
+        df.columns = [x.strip() for x in df.columns]
+        if dataset is None:
+            return df
+    elif subset == "development":
+        if os.path.exists("../data/classification_datasets_development.csv"):
+            df = pd.read_csv("../data/classification_datasets_development.csv")
+        else:
+            df = pd.read_csv(os.getcwd() + "/data/classification_datasets_development.csv")
+        df.columns = [x.strip() for x in df.columns]
+        if dataset is None:
+            return df
+    elif subset == "holdout":
+        if os.path.exists("../data/classification_datasets_development.csv"):
+            df_dev = pd.read_csv("../data/classification_datasets_development.csv")
+        else:
+            df_dev = pd.read_csv(os.getcwd() + "/data/classification_datasets_development.csv")
+        if os.path.exists("../data/classification_datasets_bakeoff.csv"):
+            df = pd.read_csv("../data/classification_datasets_bakeoff.csv")
+        else:
+            df = pd.read_csv(os.getcwd() + "/data/classification_datasets_bakeoff.csv")
+        df = df.loc[~df["Name"].isin(df_dev["Name"])].reset_index(drop=True)
+        df.columns = [x.strip() for x in df.columns]
+        if dataset is None:
+            return df
+    else:
+        if os.path.exists("../data/classification_datasets.csv"):
+            df = pd.read_csv("../data/classification_datasets.csv")
+        else:
+            df = pd.read_csv(os.getcwd() + "/data/classification_datasets.csv")
+        df.columns = [x.strip() for x in df.columns]
+        if dataset is None:
+            return df
+
+    return df.loc[df.Name == dataset].reset_index(drop=True)
+
+
+def read_univariate_ucr(filename, normalise=True):
+    if "csv" in filename:
+        data = np.loadtxt(filename, delimiter=',')
+    else:
+        data = np.loadtxt(filename, delimiter='\t')
+    Y = data[:, 0]
+    X = data[:, 1:]
+
+    scaler = StandardScaler()
+    for i in range(len(X)):
+        for j in range(len(X[i])):
+            if np.isnan(X[i, j]):
+                X[i, j] = random.random() / 1000
+        # scale it later
+        if normalise:
+            tmp = scaler.fit_transform(X[i].reshape(-1, 1))
+            X[i] = tmp[:, 0]
+    X = X.reshape((X.shape[0], X.shape[1], 1))
+    return X, Y
+
+
+def fill_missing(x: np.array,
+                 max_len: int,
+                 vary_len: str = "suffix-noise",
+                 normalise: bool = True):
+    if vary_len == "zero":
+        if normalise:
+            x = StandardScaler().fit_transform(x)
+        x = np.nan_to_num(x)
+    elif vary_len == 'prefix-suffix-noise':
+        for i in range(len(x)):
+            series = list()
+            for a in x[i, :]:
+                if np.isnan(a):
+                    break
+                series.append(a)
+            series = np.array(series)
+            seq_len = len(series)
+            diff_len = int(0.5 * (max_len - seq_len))
+
+            for j in range(diff_len):
+                x[i, j] = random.random() / 1000
+
+            for j in range(diff_len, seq_len):
+                x[i, j] = series[j - seq_len]
+
+            for j in range(seq_len, max_len):
+                x[i, j] = random.random() / 1000
+
+            if normalise:
+                tmp = StandardScaler().fit_transform(x[i].reshape(-1, 1))
+                x[i] = tmp[:, 0]
+    elif vary_len == 'uniform-scaling':
+        for i in range(len(x)):
+            series = list()
+            for a in x[i, :]:
+                if np.isnan(a):
+                    break
+                series.append(a)
+            series = np.array(series)
+            seq_len = len(series)
+
+            for j in range(max_len):
+                scaling_factor = int(j * seq_len / max_len)
+                x[i, j] = series[scaling_factor]
+            if normalise:
+                tmp = StandardScaler().fit_transform(x[i].reshape(-1, 1))
+                x[i] = tmp[:, 0]
+    else:
+        for i in range(len(x)):
+            for j in range(len(x[i])):
+                if np.isnan(x[i, j]):
+                    x[i, j] = random.random() / 1000
+
+            if normalise:
+                tmp = StandardScaler().fit_transform(x[i].reshape(-1, 1))
+                x[i] = tmp[:, 0]
+
+    return x
+
+
+def process_ts_data(X,
+                    vary_len: str = "suffix-noise",
+                    normalise: bool = False):
+    """
+    This is a function to process the data, i.e. convert dataframe to numpy array
+    :param X:
+    :param normalise:
+    :return:
+    """
+    num_instances, num_dim = X.shape
+    columns = X.columns
+    max_len = np.max([len(X[columns[0]][i]) for i in range(num_instances)])
+    output = np.zeros((num_instances, num_dim, max_len), dtype=np.float64)
+
+    for i in range(num_dim):
+        for j in range(num_instances):
+            output[j, i, :] = X[columns[i]][j].values
+        output[:, i, :] = fill_missing(
+            output[:, i, :],
+            max_len,
+            vary_len,
+            normalise
+        )
+
+    return output

time_series_classification/MultiRocket/utils/tools.py

@@ -0,0 +1,154 @@
+# Chang Wei Tan, Angus Dempster, Christoph Bergmeir, Geoffrey I Webb
+#
+# MultiRocket: Multiple pooling operators and transformations for fast and effective time series classification
+# https://arxiv.org/abs/2102.00457
+
+import cmath
+import os
+
+import numpy as np
+from numba import njit
+
+
+# =======================================================================================================
+# Simple functions that worked in Numba
+# =======================================================================================================
+@njit(fastmath=True, cache=True)
+def downsample(x, n):
+    len_y = int(np.ceil(len(x) / n))
+    y = np.zeros(len_y, dtype=np.float64)
+    for i in range(len_y):
+        y[i] = x[i * n]
+
+    return y
+
+
+@njit(fastmath=True, cache=True)
+def histc(X, bins):
+    # https://stackoverflow.com/questions/32765333/how-do-i-replicate-this-matlab-function-in-numpy
+    map_to_bins = np.digitize(X, bins)
+    r = np.zeros(bins.shape, dtype=np.int32)
+    for i in map_to_bins:
+        r[i - 1] += 1
+    return r, map_to_bins
+
+
+@njit(fastmath=True, cache=True)
+def numba_dft(x=None, sign=-1):
+    N = x.shape[0]
+    if np.log2(N) % 1 > 0:
+        nFFT = int(2 ** (np.ceil(np.log2(np.abs(N))) + 1))
+        z = np.zeros(nFFT, dtype=x.dtype)
+        z[:N] = x
+        x = z
+        N = nFFT
+
+    dft = np.zeros(N, dtype=np.complex128)
+    for i in range(N):
+        series_element = 0
+        for n in range(N):
+            series_element += x[n] * cmath.exp(sign * 2j * cmath.pi * i * n * (1 / N))
+        dft[i] = series_element
+    if sign == 1:
+        dft = dft / N
+    return dft
+
+
+@njit(fastmath=True, cache=True)
+def numba_fft_v(x, sign=-1):
+    N = x.shape[0]
+    if np.log2(N) % 1 > 0:
+        nFFT = int(2 ** (np.ceil(np.log2(np.abs(N))) + 1))
+        z = np.zeros(nFFT, dtype=x.dtype)
+        z[:N] = x
+        x = z
+        N = nFFT
+
+    x = np.asarray(x, dtype=np.complex128)
+
+    N_min = min(N, 2)
+
+    n = np.arange(N_min, dtype=np.float64)
+    k = n.T.reshape(-1, 1)
+    M = np.exp(sign * 2j * np.pi * n * k / N_min)
+    X = np.dot(M, x.reshape((N_min, -1)))
+    while X.shape[0] < N:
+        X_even = X[:, :int(X.shape[1] / 2)]
+        X_odd = X[:, int(X.shape[1] / 2):]
+        terms = np.exp(sign * 1j * np.pi * np.arange(X.shape[0]) / X.shape[0]).T.reshape(-1, 1)
+        X = np.vstack((X_even + terms * X_odd,
+                       X_even - terms * X_odd))
+    if sign == 1:
+        return X.ravel() / N
+    return X.ravel()
+
+
+@njit(fastmath=True, cache=True)
+def autocorr(y, fft):
+    l = len(y)
+
+    fft = fft * np.conjugate(fft)
+    acf = numba_fft_v(fft, sign=1)
+
+    acf = acf.real
+    acf = acf / acf[0]
+    acf = acf[:l]
+
+    return acf
+
+
+@njit(fastmath=True, cache=True)
+def numba_std(values, mean):
+    if len(values) == 1:
+        return 0
+    sum_squares_diff = 0
+    for v in values:
+        diff = v - mean
+        sum_squares_diff += diff * diff
+
+    return np.sqrt(sum_squares_diff / (len(values) - 1))
+
+
+@njit(fastmath=True, cache=True)
+def numba_min(a, b):
+    if a < b:
+        return a
+    return b
+
+
+@njit(fastmath=True, cache=True)
+def numba_max(a, b):
+    if a < b:
+        return b
+    return a
+
+
+@njit(fastmath=True, cache=True)
+def numba_linear_regression(x, y, n, lag):
+    co = np.zeros(2, dtype=np.float64)
+    sumx, sumx2, sumxy, sumy = 0, 0, 0, 0
+
+    for i in range(lag, n + lag):
+        sumx += x[i]
+        sumx2 += x[i] * x[i]
+        sumxy += x[i] * y[i]
+        sumy += y[i]
+
+    denom = n * sumx2 - sumx * sumx
+    if denom != 0:
+        co[0] = (n * sumxy - sumx * sumy) / denom
+        co[1] = (sumy * sumx2 - sumx * sumxy) / denom
+
+    return co
+
+
+def create_directory(directory_path):
+    if os.path.exists(directory_path):
+        return None
+    else:
+        try:
+            os.makedirs(directory_path)
+        except:
+            # in case another machine created the path meanwhile !:(
+            return None
+        return directory_path

time_series_classification/minirocket/LICENSE

@@ -0,0 +1,692 @@
+This project is licensed under the GNU General Public License v3.0 (GPL-3.0)
+
+Original Authors:
+  - Angus Dempster
+  - Daniel F. Schmidt
+  - Geoffrey I. Webb
+  Original repository: https://github.com/angus924/minirocket  
+  Paper: Dempster, A., Schmidt, D. F., & Webb, G. I. (2021).  
+  "MiniRocket: A Very Fast (Almost) Deterministic Transform for Time Series Classification",  
+  *Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining*, pp. 248–257.
+
+Modifications by:
+  - Jafar Bakhshaliyev, 2025
+
+This project includes modifications to the original MiniRocket codebase.  
+All changes are released under the same GPL-3.0 license.
+-----------------------------------------------------------------------
+
+                    GNU GENERAL PUBLIC LICENSE
+                       Version 3, 29 June 2007
+
+ Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The GNU General Public License is a free, copyleft license for
+software and other kinds of works.
+
+  The licenses for most software and other practical works are designed
+to take away your freedom to share and change the works.  By contrast,
+the GNU General Public License is intended to guarantee your freedom to
+share and change all versions of a program--to make sure it remains free
+software for all its users.  We, the Free Software Foundation, use the
+GNU General Public License for most of our software; it applies also to
+any other work released this way by its authors.  You can apply it to
+your programs, too.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+them if you wish), that you receive source code or can get it if you
+want it, that you can change the software or use pieces of it in new
+free programs, and that you know you can do these things.
+
+  To protect your rights, we need to prevent others from denying you
+these rights or asking you to surrender the rights.  Therefore, you have
+certain responsibilities if you distribute copies of the software, or if
+you modify it: responsibilities to respect the freedom of others.
+
+  For example, if you distribute copies of such a program, whether
+gratis or for a fee, you must pass on to the recipients the same
+freedoms that you received.  You must make sure that they, too, receive
+or can get the source code.  And you must show them these terms so they
+know their rights.
+
+  Developers that use the GNU GPL protect your rights with two steps:
+(1) assert copyright on the software, and (2) offer you this License
+giving you legal permission to copy, distribute and/or modify it.
+
+  For the developers' and authors' protection, the GPL clearly explains
+that there is no warranty for this free software.  For both users' and
+authors' sake, the GPL requires that modified versions be marked as
+changed, so that their problems will not be attributed erroneously to
+authors of previous versions.
+
+  Some devices are designed to deny users access to install or run
+modified versions of the software inside them, although the manufacturer
+can do so.  This is fundamentally incompatible with the aim of
+protecting users' freedom to change the software.  The systematic
+pattern of such abuse occurs in the area of products for individuals to
+use, which is precisely where it is most unacceptable.  Therefore, we
+have designed this version of the GPL to prohibit the practice for those
+products.  If such problems arise substantially in other domains, we
+stand ready to extend this provision to those domains in future versions
+of the GPL, as needed to protect the freedom of users.
+
+  Finally, every program is threatened constantly by software patents.
+States should not allow patents to restrict development and use of
+software on general-purpose computers, but in those that do, we wish to
+avoid the special danger that patents applied to a free program could
+make it effectively proprietary.  To prevent this, the GPL assures that
+patents cannot be used to render the program non-free.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                       TERMS AND CONDITIONS
+
+  0. Definitions.
+
+  "This License" refers to version 3 of the GNU General Public License.
+
+  "Copyright" also means copyright-like laws that apply to other kinds of
+works, such as semiconductor masks.
+
+  "The Program" refers to any copyrightable work licensed under this
+License.  Each licensee is addressed as "you".  "Licensees" and
+"recipients" may be individuals or organizations.
+
+  To "modify" a work means to copy from or adapt all or part of the work
+in a fashion requiring copyright permission, other than the making of an
+exact copy.  The resulting work is called a "modified version" of the
+earlier work or a work "based on" the earlier work.
+
+  A "covered work" means either the unmodified Program or a work based
+on the Program.
+
+  To "propagate" a work means to do anything with it that, without
+permission, would make you directly or secondarily liable for
+infringement under applicable copyright law, except executing it on a
+computer or modifying a private copy.  Propagation includes copying,
+distribution (with or without modification), making available to the
+public, and in some countries other activities as well.
+
+  To "convey" a work means any kind of propagation that enables other
+parties to make or receive copies.  Mere interaction with a user through
+a computer network, with no transfer of a copy, is not conveying.
+
+  An interactive user interface displays "Appropriate Legal Notices"
+to the extent that it includes a convenient and prominently visible
+feature that (1) displays an appropriate copyright notice, and (2)
+tells the user that there is no warranty for the work (except to the
+extent that warranties are provided), that licensees may convey the
+work under this License, and how to view a copy of this License.  If
+the interface presents a list of user commands or options, such as a
+menu, a prominent item in the list meets this criterion.
+
+  1. Source Code.
+
+  The "source code" for a work means the preferred form of the work
+for making modifications to it.  "Object code" means any non-source
+form of a work.
+
+  A "Standard Interface" means an interface that either is an official
+standard defined by a recognized standards body, or, in the case of
+interfaces specified for a particular programming language, one that
+is widely used among developers working in that language.
+
+  The "System Libraries" of an executable work include anything, other
+than the work as a whole, that (a) is included in the normal form of
+packaging a Major Component, but which is not part of that Major
+Component, and (b) serves only to enable use of the work with that
+Major Component, or to implement a Standard Interface for which an
+implementation is available to the public in source code form.  A
+"Major Component", in this context, means a major essential component
+(kernel, window system, and so on) of the specific operating system
+(if any) on which the executable work runs, or a compiler used to
+produce the work, or an object code interpreter used to run it.
+
+  The "Corresponding Source" for a work in object code form means all
+the source code needed to generate, install, and (for an executable
+work) run the object code and to modify the work, including scripts to
+control those activities.  However, it does not include the work's
+System Libraries, or general-purpose tools or generally available free
+programs which are used unmodified in performing those activities but
+which are not part of the work.  For example, Corresponding Source
+includes interface definition files associated with source files for
+the work, and the source code for shared libraries and dynamically
+linked subprograms that the work is specifically designed to require,
+such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+  The Corresponding Source need not include anything that users
+can regenerate automatically from other parts of the Corresponding
+Source.
+
+  The Corresponding Source for a work in source code form is that
+same work.
+
+  2. Basic Permissions.
+
+  All rights granted under this License are granted for the term of
+copyright on the Program, and are irrevocable provided the stated
+conditions are met.  This License explicitly affirms your unlimited
+permission to run the unmodified Program.  The output from running a
+covered work is covered by this License only if the output, given its
+content, constitutes a covered work.  This License acknowledges your
+rights of fair use or other equivalent, as provided by copyright law.
+
+  You may make, run and propagate covered works that you do not
+convey, without conditions so long as your license otherwise remains
+in force.  You may convey covered works to others for the sole purpose
+of having them make modifications exclusively for you, or provide you
+with facilities for running those works, provided that you comply with
+the terms of this License in conveying all material for which you do
+not control copyright.  Those thus making or running the covered works
+for you must do so exclusively on your behalf, under your direction
+and control, on terms that prohibit them from making any copies of
+your copyrighted material outside their relationship with you.
+
+  Conveying under any other circumstances is permitted solely under
+the conditions stated below.  Sublicensing is not allowed; section 10
+makes it unnecessary.
+
+  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+
+  No covered work shall be deemed part of an effective technological
+measure under any applicable law fulfilling obligations under article
+11 of the WIPO copyright treaty adopted on 20 December 1996, or
+similar laws prohibiting or restricting circumvention of such
+measures.
+
+  When you convey a covered work, you waive any legal power to forbid
+circumvention of technological measures to the extent such circumvention
+is effected by exercising rights under this License with respect to
+the covered work, and you disclaim any intention to limit operation or
+modification of the work as a means of enforcing, against the work's
+users, your or third parties' legal rights to forbid circumvention of
+technological measures.
+
+  4. Conveying Verbatim Copies.
+
+  You may convey verbatim copies of the Program's source code as you
+receive it, in any medium, provided that you conspicuously and
+appropriately publish on each copy an appropriate copyright notice;
+keep intact all notices stating that this License and any
+non-permissive terms added in accord with section 7 apply to the code;
+keep intact all notices of the absence of any warranty; and give all
+recipients a copy of this License along with the Program.
+
+  You may charge any price or no price for each copy that you convey,
+and you may offer support or warranty protection for a fee.
+
+  5. Conveying Modified Source Versions.
+
+  You may convey a work based on the Program, or the modifications to
+produce it from the Program, in the form of source code under the
+terms of section 4, provided that you also meet all of these conditions:
+
+    a) The work must carry prominent notices stating that you modified
+    it, and giving a relevant date.
+
+    b) The work must carry prominent notices stating that it is
+    released under this License and any conditions added under section
+    7.  This requirement modifies the requirement in section 4 to
+    "keep intact all notices".
+
+    c) You must license the entire work, as a whole, under this
+    License to anyone who comes into possession of a copy.  This
+    License will therefore apply, along with any applicable section 7
+    additional terms, to the whole of the work, and all its parts,
+    regardless of how they are packaged.  This License gives no
+    permission to license the work in any other way, but it does not
+    invalidate such permission if you have separately received it.
+
+    d) If the work has interactive user interfaces, each must display
+    Appropriate Legal Notices; however, if the Program has interactive
+    interfaces that do not display Appropriate Legal Notices, your
+    work need not make them do so.
+
+  A compilation of a covered work with other separate and independent
+works, which are not by their nature extensions of the covered work,
+and which are not combined with it such as to form a larger program,
+in or on a volume of a storage or distribution medium, is called an
+"aggregate" if the compilation and its resulting copyright are not
+used to limit the access or legal rights of the compilation's users
+beyond what the individual works permit.  Inclusion of a covered work
+in an aggregate does not cause this License to apply to the other
+parts of the aggregate.
+
+  6. Conveying Non-Source Forms.
+
+  You may convey a covered work in object code form under the terms
+of sections 4 and 5, provided that you also convey the
+machine-readable Corresponding Source under the terms of this License,
+in one of these ways:
+
+    a) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by the
+    Corresponding Source fixed on a durable physical medium
+    customarily used for software interchange.
+
+    b) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by a
+    written offer, valid for at least three years and valid for as
+    long as you offer spare parts or customer support for that product
+    model, to give anyone who possesses the object code either (1) a
+    copy of the Corresponding Source for all the software in the
+    product that is covered by this License, on a durable physical
+    medium customarily used for software interchange, for a price no
+    more than your reasonable cost of physically performing this
+    conveying of source, or (2) access to copy the
+    Corresponding Source from a network server at no charge.
+
+    c) Convey individual copies of the object code with a copy of the
+    written offer to provide the Corresponding Source.  This
+    alternative is allowed only occasionally and noncommercially, and
+    only if you received the object code with such an offer, in accord
+    with subsection 6b.
+
+    d) Convey the object code by offering access from a designated
+    place (gratis or for a charge), and offer equivalent access to the
+    Corresponding Source in the same way through the same place at no
+    further charge.  You need not require recipients to copy the
+    Corresponding Source along with the object code.  If the place to
+    copy the object code is a network server, the Corresponding Source
+    may be on a different server (operated by you or a third party)
+    that supports equivalent copying facilities, provided you maintain
+    clear directions next to the object code saying where to find the
+    Corresponding Source.  Regardless of what server hosts the
+    Corresponding Source, you remain obligated to ensure that it is
+    available for as long as needed to satisfy these requirements.
+
+    e) Convey the object code using peer-to-peer transmission, provided
+    you inform other peers where the object code and Corresponding
+    Source of the work are being offered to the general public at no
+    charge under subsection 6d.
+
+  A separable portion of the object code, whose source code is excluded
+from the Corresponding Source as a System Library, need not be
+included in conveying the object code work.
+
+  A "User Product" is either (1) a "consumer product", which means any
+tangible personal property which is normally used for personal, family,
+or household purposes, or (2) anything designed or sold for incorporation
+into a dwelling.  In determining whether a product is a consumer product,
+doubtful cases shall be resolved in favor of coverage.  For a particular
+product received by a particular user, "normally used" refers to a
+typical or common use of that class of product, regardless of the status
+of the particular user or of the way in which the particular user
+actually uses, or expects or is expected to use, the product.  A product
+is a consumer product regardless of whether the product has substantial
+commercial, industrial or non-consumer uses, unless such uses represent
+the only significant mode of use of the product.
+
+  "Installation Information" for a User Product means any methods,
+procedures, authorization keys, or other information required to install
+and execute modified versions of a covered work in that User Product from
+a modified version of its Corresponding Source.  The information must
+suffice to ensure that the continued functioning of the modified object
+code is in no case prevented or interfered with solely because
+modification has been made.
+
+  If you convey an object code work under this section in, or with, or
+specifically for use in, a User Product, and the conveying occurs as
+part of a transaction in which the right of possession and use of the
+User Product is transferred to the recipient in perpetuity or for a
+fixed term (regardless of how the transaction is characterized), the
+Corresponding Source conveyed under this section must be accompanied
+by the Installation Information.  But this requirement does not apply
+if neither you nor any third party retains the ability to install
+modified object code on the User Product (for example, the work has
+been installed in ROM).
+
+  The requirement to provide Installation Information does not include a
+requirement to continue to provide support service, warranty, or updates
+for a work that has been modified or installed by the recipient, or for
+the User Product in which it has been modified or installed.  Access to a
+network may be denied when the modification itself materially and
+adversely affects the operation of the network or violates the rules and
+protocols for communication across the network.
+
+  Corresponding Source conveyed, and Installation Information provided,
+in accord with this section must be in a format that is publicly
+documented (and with an implementation available to the public in
+source code form), and must require no special password or key for
+unpacking, reading or copying.
+
+  7. Additional Terms.
+
+  "Additional permissions" are terms that supplement the terms of this
+License by making exceptions from one or more of its conditions.
+Additional permissions that are applicable to the entire Program shall
+be treated as though they were included in this License, to the extent
+that they are valid under applicable law.  If additional permissions
+apply only to part of the Program, that part may be used separately
+under those permissions, but the entire Program remains governed by
+this License without regard to the additional permissions.
+
+  When you convey a copy of a covered work, you may at your option
+remove any additional permissions from that copy, or from any part of
+it.  (Additional permissions may be written to require their own
+removal in certain cases when you modify the work.)  You may place
+additional permissions on material, added by you to a covered work,
+for which you have or can give appropriate copyright permission.
+
+  Notwithstanding any other provision of this License, for material you
+add to a covered work, you may (if authorized by the copyright holders of
+that material) supplement the terms of this License with terms:
+
+    a) Disclaiming warranty or limiting liability differently from the
+    terms of sections 15 and 16 of this License; or
+
+    b) Requiring preservation of specified reasonable legal notices or
+    author attributions in that material or in the Appropriate Legal
+    Notices displayed by works containing it; or
+
+    c) Prohibiting misrepresentation of the origin of that material, or
+    requiring that modified versions of such material be marked in
+    reasonable ways as different from the original version; or
+
+    d) Limiting the use for publicity purposes of names of licensors or
+    authors of the material; or
+
+    e) Declining to grant rights under trademark law for use of some
+    trade names, trademarks, or service marks; or
+
+    f) Requiring indemnification of licensors and authors of that
+    material by anyone who conveys the material (or modified versions of
+    it) with contractual assumptions of liability to the recipient, for
+    any liability that these contractual assumptions directly impose on
+    those licensors and authors.
+
+  All other non-permissive additional terms are considered "further
+restrictions" within the meaning of section 10.  If the Program as you
+received it, or any part of it, contains a notice stating that it is
+governed by this License along with a term that is a further
+restriction, you may remove that term.  If a license document contains
+a further restriction but permits relicensing or conveying under this
+License, you may add to a covered work material governed by the terms
+of that license document, provided that the further restriction does
+not survive such relicensing or conveying.
+
+  If you add terms to a covered work in accord with this section, you
+must place, in the relevant source files, a statement of the
+additional terms that apply to those files, or a notice indicating
+where to find the applicable terms.
+
+  Additional terms, permissive or non-permissive, may be stated in the
+form of a separately written license, or stated as exceptions;
+the above requirements apply either way.
+
+  8. Termination.
+
+  You may not propagate or modify a covered work except as expressly
+provided under this License.  Any attempt otherwise to propagate or
+modify it is void, and will automatically terminate your rights under
+this License (including any patent licenses granted under the third
+paragraph of section 11).
+
+  However, if you cease all violation of this License, then your
+license from a particular copyright holder is reinstated (a)
+provisionally, unless and until the copyright holder explicitly and
+finally terminates your license, and (b) permanently, if the copyright
+holder fails to notify you of the violation by some reasonable means
+prior to 60 days after the cessation.
+
+  Moreover, your license from a particular copyright holder is
+reinstated permanently if the copyright holder notifies you of the
+violation by some reasonable means, this is the first time you have
+received notice of violation of this License (for any work) from that
+copyright holder, and you cure the violation prior to 30 days after
+your receipt of the notice.
+
+  Termination of your rights under this section does not terminate the
+licenses of parties who have received copies or rights from you under
+this License.  If your rights have been terminated and not permanently
+reinstated, you do not qualify to receive new licenses for the same
+material under section 10.
+
+  9. Acceptance Not Required for Having Copies.
+
+  You are not required to accept this License in order to receive or
+run a copy of the Program.  Ancillary propagation of a covered work
+occurring solely as a consequence of using peer-to-peer transmission
+to receive a copy likewise does not require acceptance.  However,
+nothing other than this License grants you permission to propagate or
+modify any covered work.  These actions infringe copyright if you do
+not accept this License.  Therefore, by modifying or propagating a
+covered work, you indicate your acceptance of this License to do so.
+
+  10. Automatic Licensing of Downstream Recipients.
+
+  Each time you convey a covered work, the recipient automatically
+receives a license from the original licensors, to run, modify and
+propagate that work, subject to this License.  You are not responsible
+for enforcing compliance by third parties with this License.
+
+  An "entity transaction" is a transaction transferring control of an
+organization, or substantially all assets of one, or subdividing an
+organization, or merging organizations.  If propagation of a covered
+work results from an entity transaction, each party to that
+transaction who receives a copy of the work also receives whatever
+licenses to the work the party's predecessor in interest had or could
+give under the previous paragraph, plus a right to possession of the
+Corresponding Source of the work from the predecessor in interest, if
+the predecessor has it or can get it with reasonable efforts.
+
+  You may not impose any further restrictions on the exercise of the
+rights granted or affirmed under this License.  For example, you may
+not impose a license fee, royalty, or other charge for exercise of
+rights granted under this License, and you may not initiate litigation
+(including a cross-claim or counterclaim in a lawsuit) alleging that
+any patent claim is infringed by making, using, selling, offering for
+sale, or importing the Program or any portion of it.
+
+  11. Patents.
+
+  A "contributor" is a copyright holder who authorizes use under this
+License of the Program or a work on which the Program is based.  The
+work thus licensed is called the contributor's "contributor version".
+
+  A contributor's "essential patent claims" are all patent claims
+owned or controlled by the contributor, whether already acquired or
+hereafter acquired, that would be infringed by some manner, permitted
+by this License, of making, using, or selling its contributor version,
+but do not include claims that would be infringed only as a
+consequence of further modification of the contributor version.  For
+purposes of this definition, "control" includes the right to grant
+patent sublicenses in a manner consistent with the requirements of
+this License.
+
+  Each contributor grants you a non-exclusive, worldwide, royalty-free
+patent license under the contributor's essential patent claims, to
+make, use, sell, offer for sale, import and otherwise run, modify and
+propagate the contents of its contributor version.
+
+  In the following three paragraphs, a "patent license" is any express
+agreement or commitment, however denominated, not to enforce a patent
+(such as an express permission to practice a patent or covenant not to
+sue for patent infringement).  To "grant" such a patent license to a
+party means to make such an agreement or commitment not to enforce a
+patent against the party.
+
+  If you convey a covered work, knowingly relying on a patent license,
+and the Corresponding Source of the work is not available for anyone
+to copy, free of charge and under the terms of this License, through a
+publicly available network server or other readily accessible means,
+then you must either (1) cause the Corresponding Source to be so
+available, or (2) arrange to deprive yourself of the benefit of the
+patent license for this particular work, or (3) arrange, in a manner
+consistent with the requirements of this License, to extend the patent
+license to downstream recipients.  "Knowingly relying" means you have
+actual knowledge that, but for the patent license, your conveying the
+covered work in a country, or your recipient's use of the covered work
+in a country, would infringe one or more identifiable patents in that
+country that you have reason to believe are valid.
+
+  If, pursuant to or in connection with a single transaction or
+arrangement, you convey, or propagate by procuring conveyance of, a
+covered work, and grant a patent license to some of the parties
+receiving the covered work authorizing them to use, propagate, modify
+or convey a specific copy of the covered work, then the patent license
+you grant is automatically extended to all recipients of the covered
+work and works based on it.
+
+  A patent license is "discriminatory" if it does not include within
+the scope of its coverage, prohibits the exercise of, or is
+conditioned on the non-exercise of one or more of the rights that are
+specifically granted under this License.  You may not convey a covered
+work if you are a party to an arrangement with a third party that is
+in the business of distributing software, under which you make payment
+to the third party based on the extent of your activity of conveying
+the work, and under which the third party grants, to any of the
+parties who would receive the covered work from you, a discriminatory
+patent license (a) in connection with copies of the covered work
+conveyed by you (or copies made from those copies), or (b) primarily
+for and in connection with specific products or compilations that
+contain the covered work, unless you entered into that arrangement,
+or that patent license was granted, prior to 28 March 2007.
+
+  Nothing in this License shall be construed as excluding or limiting
+any implied license or other defenses to infringement that may
+otherwise be available to you under applicable patent law.
+
+  12. No Surrender of Others' Freedom.
+
+  If conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot convey a
+covered work so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you may
+not convey it at all.  For example, if you agree to terms that obligate you
+to collect a royalty for further conveying from those to whom you convey
+the Program, the only way you could satisfy both those terms and this
+License would be to refrain entirely from conveying the Program.
+
+  13. Use with the GNU Affero General Public License.
+
+  Notwithstanding any other provision of this License, you have
+permission to link or combine any covered work with a work licensed
+under version 3 of the GNU Affero General Public License into a single
+combined work, and to convey the resulting work.  The terms of this
+License will continue to apply to the part which is the covered work,
+but the special requirements of the GNU Affero General Public License,
+section 13, concerning interaction through a network will apply to the
+combination as such.
+
+  14. Revised Versions of this License.
+
+  The Free Software Foundation may publish revised and/or new versions of
+the GNU General Public License from time to time.  Such new versions will
+be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+  Each version is given a distinguishing version number.  If the
+Program specifies that a certain numbered version of the GNU General
+Public License "or any later version" applies to it, you have the
+option of following the terms and conditions either of that numbered
+version or of any later version published by the Free Software
+Foundation.  If the Program does not specify a version number of the
+GNU General Public License, you may choose any version ever published
+by the Free Software Foundation.
+
+  If the Program specifies that a proxy can decide which future
+versions of the GNU General Public License can be used, that proxy's
+public statement of acceptance of a version permanently authorizes you
+to choose that version for the Program.
+
+  Later license versions may give you additional or different
+permissions.  However, no additional obligations are imposed on any
+author or copyright holder as a result of your choosing to follow a
+later version.
+
+  15. Disclaimer of Warranty.
+
+  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
+APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
+OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
+IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
+ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+  16. Limitation of Liability.
+
+  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
+THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
+GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
+USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
+DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
+PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
+EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
+SUCH DAMAGES.
+
+  17. Interpretation of Sections 15 and 16.
+
+  If the disclaimer of warranty and limitation of liability provided
+above cannot be given local legal effect according to their terms,
+reviewing courts shall apply local law that most closely approximates
+an absolute waiver of all civil liability in connection with the
+Program, unless a warranty or assumption of liability accompanies a
+copy of the Program in return for a fee.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+state the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software: you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation, either version 3 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+Also add information on how to contact you by electronic and paper mail.
+
+  If the program does terminal interaction, make it output a short
+notice like this when it starts in an interactive mode:
+
+    <program>  Copyright (C) <year>  <name of author>
+    This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
+    This is free software, and you are welcome to redistribute it
+    under certain conditions; type `show c' for details.
+
+The hypothetical commands `show w' and `show c' should show the appropriate
+parts of the General Public License.  Of course, your program's commands
+might be different; for a GUI interface, you would use an "about box".
+
+  You should also get your employer (if you work as a programmer) or school,
+if any, to sign a "copyright disclaimer" for the program, if necessary.
+For more information on this, and how to apply and follow the GNU GPL, see
+<https://www.gnu.org/licenses/>.
+
+  The GNU General Public License does not permit incorporating your program
+into proprietary programs.  If your program is a subroutine library, you
+may consider it more useful to permit linking proprietary applications with
+the library.  If this is what you want to do, use the GNU Lesser General
+Public License instead of this License.  But first, please read
+<https://www.gnu.org/licenses/why-not-lgpl.html>.

time_series_classification/minirocket/THIRD_PARTY_LICENSES.txt

@@ -0,0 +1,221 @@
+THIRD PARTY LICENSES
+
+This file contains the licenses for third-party components used in this project.
+
+================================================================================
+
+Time Series Augmentation Components
+
+Some augmentation methods in this repository are derived from or inspired by the 
+time_series_augmentation repository:
+
+Source: https://github.com/uchidalab/time_series_augmentation
+Authors: Brian Kenji Iwana and Seiichi Uchida
+License: Apache License 2.0
+Files affected: ./time_series_classification/minirocket/src/augmentation.py
+
+================================================================================
+
+
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

time_series_classification/minirocket/scripts/example.sh

@@ -0,0 +1,65 @@
+#!/bin/bash
+#SBATCH --job-name=univariate_tsc
+#SBATCH --output=tsc_univariate.out
+#SBATCH --error=tsc_univariate.err
+#SBATCH --partition=CPU
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=40 
+#SBATCH --chdir= # Change to your working directory
+
+
+
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup
+
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --jitter
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --scaling
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --rotation
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --permutation
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --randompermutation
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --magwarp
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --timewarp
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --windowslice
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --windowwarp
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --spawner
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --dtwwarp
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --shapedtwwarp
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --wdba
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --discdtw
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --discsdtw
+
+
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 420 --stride 1 --shuffle_rate 0.8 
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 120 --stride 1 --shuffle_rate 0.7
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 4 --stride 4 --shuffle_rate 0.6
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 64 --stride 2 --shuffle_rate 0.6
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 32 --stride 1 --shuffle_rate 0.6
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 8 --stride 4 --shuffle_rate 1.0
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 6 --stride 4 --shuffle_rate 1.0
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 120 --stride 1 --shuffle_rate 1.0
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 96 --stride 8 --shuffle_rate 0.8
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 48 --stride 36 --shuffle_rate 0.9
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 24 --stride 36 --shuffle_rate 0.8
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 240 --stride 1 --shuffle_rate 0.8
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 240 --stride 1 --shuffle_rate 0.4
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 240 --stride 1 --shuffle_rate 0.2
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 96 --stride 96 --shuffle_rate 1.0
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 96 --stride 96 --shuffle_rate 0.6
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 120 --stride 48 --shuffle_rate 0.8
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 240 --stride 96 --shuffle_rate 1.0
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 12 --stride 96 --shuffle_rate 0.8
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 2 --stride 1 --shuffle_rate 0.4
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 480 --stride 1 --shuffle_rate 0.7
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 900 --stride 1 --shuffle_rate 0.8
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 64 --stride 24 --shuffle_rate 0.8
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 64 --stride 24 --shuffle_rate 1.0
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 148 --stride 24 --shuffle_rate 1.0
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 148 --stride 96 --shuffle_rate 1.0
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 148 --stride 48 --shuffle_rate 1.0
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 180 --stride 12 --shuffle_rate 0.8
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 180 --stride 24 --shuffle_rate 0.8
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 180 --stride 12 --shuffle_rate 0.5
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 152 --stride 96 --shuffle_rate 1.0
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 360 --stride 1 --shuffle_rate 0.1
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 72 --stride 2 --shuffle_rate 0.4
+srun python3 ./src/main.py --dataset MiddlePhalanxOutlineAgeGroup --use-augmentation --augmentation-ratio 1 --tps --patch_len 120 --stride 1 --shuffle_rate 0.2

time_series_classification/minirocket/src/augmentation.py

@@ -0,0 +1,415 @@
+# Adapted from time_series_augmentation (https://github.com/uchidalab/time_series_augmentation)
+# Original: Apache License 2.0 by Brian Kenji Iwana and Seiichi Uchida
+# Modified by Jafar Bakhshaliyev (2025) - Licensed under GPL v3.0
+
+
+import numpy as np
+from tqdm import tqdm
+
+
+
+
+def tps(x, y, patch_len=0, stride=0, shuffle_rate=0.0):
+    """
+   Temporal Patch Shuffle (TPS) augmentation for time series classification.
+
+    Parameters:
+    -----------
+    x : numpy.ndarray
+        Input time series data of shape (n_samples, timesteps, n_features)
+    patch_len : int
+        Length of each patch
+    stride : int
+        Stride between patches
+    shuffle_rate : float
+        Proportion of patches to shuffle (between 0 and 1)
+        
+    Returns:
+    --------
+    numpy.ndarray
+        Augmented time series data with same shape as input
+    """
+    n_samples, T, n_features = x.shape
+        
+
+    ret = np.zeros_like(x)
+    
+
+    total_patches = (T - patch_len + stride - 1) // stride + 1
+    total_len = (total_patches - 1) * stride + patch_len
+    padding_needed = total_len - T
+
+    for i in range(n_samples):
+
+        sample = x[i]  # shape: (timesteps, n_features)
+        
+        if padding_needed > 0:
+            padded_sample = np.pad(sample, ((0, padding_needed), (0, 0)), mode='edge')
+            T_padded = T + padding_needed
+        else:
+            padded_sample = sample
+            T_padded = T
+     
+        num_patches = ((T_padded - patch_len) // stride) + 1
+
+        patches = np.zeros((num_patches, patch_len, n_features))
+        for j in range(num_patches):
+            start = j * stride
+            patches[j] = padded_sample[start:start + patch_len]
+            
+   
+        variance_scores = np.var(patches, axis=(1, 2))
+        
+
+        num_to_shuffle = int(num_patches * shuffle_rate)
+        
+        if num_to_shuffle > 0:
+ 
+            shuffle_indices = np.argsort(variance_scores)[:num_to_shuffle]
+            
+   
+            patches_to_shuffle = patches[shuffle_indices].copy()
+            shuffled_order = np.random.permutation(num_to_shuffle)
+            
+            for idx, new_idx in enumerate(shuffled_order):
+                patch_idx = shuffle_indices[idx]
+                new_patch = patches_to_shuffle[new_idx]
+                patches[patch_idx] = new_patch
+            
+        # Reconstruct the time series
+        reconstructed = np.zeros((T_padded, n_features))
+        counts = np.zeros((T_padded, n_features))
+        
+        for j in range(num_patches):
+            start = j * stride
+            end = start + patch_len
+            reconstructed[start:end] += patches[j]
+            counts[start:end] += 1
+  
+        counts[counts == 0] = 1
+        reconstructed = reconstructed / counts
+        
+
+        ret[i] = reconstructed[:T]
+        
+    return ret
+
+
+def jitter(x, sigma=0.03):
+    # https://arxiv.org/pdf/1706.00527.pdf
+    return x + np.random.normal(loc=0., scale=sigma, size=x.shape)
+
+def scaling(x, sigma=0.1):
+    # https://arxiv.org/pdf/1706.00527.pdf
+    factor = np.random.normal(loc=1., scale=sigma, size=(x.shape[0],x.shape[2]))
+    return np.multiply(x, factor[:,np.newaxis,:])
+
+def rotation(x):
+    flip = np.random.choice([-1, 1], size=(x.shape[0],x.shape[2]))
+    rotate_axis = np.arange(x.shape[2])
+    np.random.shuffle(rotate_axis)    
+    return flip[:,np.newaxis,:] * x[:,:,rotate_axis]
+
+def permutation(x, max_segments=5, seg_mode="equal"):
+    orig_steps = np.arange(x.shape[1])
+    num_segs = np.random.randint(1, max_segments, size=(x.shape[0]))
+
+    ret = np.zeros_like(x)
+    for i, pat in enumerate(x):
+        if num_segs[i] > 1:
+            if seg_mode == "random":
+                split_points = np.random.choice(x.shape[1] - 2, num_segs[i] - 1, replace=False)
+                split_points.sort()
+                splits = np.split(orig_steps, split_points)
+            else:
+                splits = np.array_split(orig_steps, num_segs[i])
+
+    
+            perm = np.random.permutation(len(splits))
+            warp = np.concatenate([splits[j] for j in perm]).ravel()
+
+            ret[i] = pat[warp]
+        else:
+            ret[i] = pat
+    return ret
+
+
+def magnitude_warp(x, sigma=0.2, knot=4):
+    from scipy.interpolate import CubicSpline
+    orig_steps = np.arange(x.shape[1])
+    
+    random_warps = np.random.normal(loc=1.0, scale=sigma, size=(x.shape[0], knot+2, x.shape[2]))
+    warp_steps = (np.ones((x.shape[2],1))*(np.linspace(0, x.shape[1]-1., num=knot+2))).T
+    ret = np.zeros_like(x)
+    for i, pat in enumerate(x):
+        warper = np.array([CubicSpline(warp_steps[:,dim], random_warps[i,:,dim])(orig_steps) for dim in range(x.shape[2])]).T
+        ret[i] = pat * warper
+
+    return ret
+
+def time_warp(x, sigma=0.2, knot=4):
+    from scipy.interpolate import CubicSpline
+    orig_steps = np.arange(x.shape[1])
+    
+    random_warps = np.random.normal(loc=1.0, scale=sigma, size=(x.shape[0], knot+2, x.shape[2]))
+    warp_steps = (np.ones((x.shape[2],1))*(np.linspace(0, x.shape[1]-1., num=knot+2))).T
+    
+    ret = np.zeros_like(x)
+    for i, pat in enumerate(x):
+        for dim in range(x.shape[2]):
+            time_warp = CubicSpline(warp_steps[:,dim], warp_steps[:,dim] * random_warps[i,:,dim])(orig_steps)
+            scale = (x.shape[1]-1)/time_warp[-1]
+            ret[i,:,dim] = np.interp(orig_steps, np.clip(scale*time_warp, 0, x.shape[1]-1), pat[:,dim]).T
+    return ret
+
+def window_slice(x, reduce_ratio=0.9):
+    # https://halshs.archives-ouvertes.fr/halshs-01357973/document
+    target_len = np.ceil(reduce_ratio*x.shape[1]).astype(int)
+    if target_len >= x.shape[1]:
+        return x
+    starts = np.random.randint(low=0, high=x.shape[1]-target_len, size=(x.shape[0])).astype(int)
+    ends = (target_len + starts).astype(int)
+    
+    ret = np.zeros_like(x)
+    for i, pat in enumerate(x):
+        for dim in range(x.shape[2]):
+            ret[i,:,dim] = np.interp(np.linspace(0, target_len, num=x.shape[1]), np.arange(target_len), pat[starts[i]:ends[i],dim]).T
+    return ret
+
+def window_warp(x, window_ratio=0.1, scales=[0.5, 2.]):
+    # https://halshs.archives-ouvertes.fr/halshs-01357973/document
+    warp_scales = np.random.choice(scales, x.shape[0])
+    warp_size = np.ceil(window_ratio*x.shape[1]).astype(int)
+    window_steps = np.arange(warp_size)
+        
+    window_starts = np.random.randint(low=1, high=x.shape[1]-warp_size-1, size=(x.shape[0])).astype(int)
+    window_ends = (window_starts + warp_size).astype(int)
+            
+    ret = np.zeros_like(x)
+    for i, pat in enumerate(x):
+        for dim in range(x.shape[2]):
+            start_seg = pat[:window_starts[i],dim]
+            window_seg = np.interp(np.linspace(0, warp_size-1, num=int(warp_size*warp_scales[i])), window_steps, pat[window_starts[i]:window_ends[i],dim])
+            end_seg = pat[window_ends[i]:,dim]
+            warped = np.concatenate((start_seg, window_seg, end_seg))                
+            ret[i,:,dim] = np.interp(np.arange(x.shape[1]), np.linspace(0, x.shape[1]-1., num=warped.size), warped).T
+    return ret
+
+def spawner(x, labels, sigma=0.05, verbose=0):
+    # https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6983028/
+    # use verbose=-1 to turn off warnings
+    # use verbose=1 to print out figures
+    
+    import dtw as dtw
+    random_points = np.random.randint(low=1, high=x.shape[1]-1, size=x.shape[0])
+    window = np.ceil(x.shape[1] / 10.).astype(int)
+    orig_steps = np.arange(x.shape[1])
+    l = np.argmax(labels, axis=1) if labels.ndim > 1 else labels
+    
+    ret = np.zeros_like(x)
+    for i, pat in enumerate(tqdm(x)):
+        # guarentees that same one isnt selected
+        choices = np.delete(np.arange(x.shape[0]), i)
+        # remove ones of different classes
+        choices = np.where(l[choices] == l[i])[0]
+        if choices.size > 0:     
+            random_sample = x[np.random.choice(choices)]
+            # SPAWNER splits the path into two randomly
+            path1 = dtw.dtw(pat[:random_points[i]], random_sample[:random_points[i]], dtw.RETURN_PATH, slope_constraint="symmetric", window=window)
+            path2 = dtw.dtw(pat[random_points[i]:], random_sample[random_points[i]:], dtw.RETURN_PATH, slope_constraint="symmetric", window=window)
+            combined = np.concatenate((np.vstack(path1), np.vstack(path2+random_points[i])), axis=1)
+            if verbose:
+                print(random_points[i])
+                dtw_value, cost, DTW_map, path = dtw.dtw(pat, random_sample, return_flag = dtw.RETURN_ALL, slope_constraint=slope_constraint, window=window)
+                dtw.draw_graph1d(cost, DTW_map, path, pat, random_sample)
+                dtw.draw_graph1d(cost, DTW_map, combined, pat, random_sample)
+            mean = np.mean([pat[combined[0]], random_sample[combined[1]]], axis=0)
+            for dim in range(x.shape[2]):
+                ret[i,:,dim] = np.interp(orig_steps, np.linspace(0, x.shape[1]-1., num=mean.shape[0]), mean[:,dim]).T
+        else:
+            if verbose > -1:
+                print("There is only one pattern of class %d, skipping pattern average"%l[i])
+            ret[i,:] = pat
+    return jitter(ret, sigma=sigma)
+
+def wdba(x, labels, batch_size=6, slope_constraint="symmetric", use_window=True, verbose=0):
+    # https://ieeexplore.ieee.org/document/8215569
+    # use verbose = -1 to turn off warnings    
+    # slope_constraint is for DTW. "symmetric" or "asymmetric"
+    
+    import dtw as dtw
+    
+    if use_window:
+        window = np.ceil(x.shape[1] / 10.).astype(int)
+    else:
+        window = None
+    orig_steps = np.arange(x.shape[1])
+    l = np.argmax(labels, axis=1) if labels.ndim > 1 else labels
+        
+    ret = np.zeros_like(x)
+    for i in tqdm(range(ret.shape[0])):
+        # get the same class as i
+        choices = np.where(l == l[i])[0]
+        if choices.size > 0:        
+            # pick random intra-class pattern
+            k = min(choices.size, batch_size)
+            random_prototypes = x[np.random.choice(choices, k, replace=False)]
+            
+            # calculate dtw between all
+            dtw_matrix = np.zeros((k, k))
+            for p, prototype in enumerate(random_prototypes):
+                for s, sample in enumerate(random_prototypes):
+                    if p == s:
+                        dtw_matrix[p, s] = 0.
+                    else:
+                        dtw_matrix[p, s] = dtw.dtw(prototype, sample, dtw.RETURN_VALUE, slope_constraint=slope_constraint, window=window)
+                        
+            # get medoid
+            medoid_id = np.argsort(np.sum(dtw_matrix, axis=1))[0]
+            nearest_order = np.argsort(dtw_matrix[medoid_id])
+            medoid_pattern = random_prototypes[medoid_id]
+            
+            # start weighted DBA
+            average_pattern = np.zeros_like(medoid_pattern)
+            weighted_sums = np.zeros((medoid_pattern.shape[0]))
+            for nid in nearest_order:
+                if nid == medoid_id or dtw_matrix[medoid_id, nearest_order[1]] == 0.:
+                    average_pattern += medoid_pattern 
+                    weighted_sums += np.ones_like(weighted_sums) 
+                else:
+                    path = dtw.dtw(medoid_pattern, random_prototypes[nid], dtw.RETURN_PATH, slope_constraint=slope_constraint, window=window)
+                    dtw_value = dtw_matrix[medoid_id, nid]
+                    warped = random_prototypes[nid, path[1]]
+                    weight = np.exp(np.log(0.5)*dtw_value/dtw_matrix[medoid_id, nearest_order[1]])
+                    average_pattern[path[0]] += weight * warped
+                    weighted_sums[path[0]] += weight 
+            
+            ret[i,:] = average_pattern / weighted_sums[:,np.newaxis]
+        else:
+            if verbose > -1:
+                print("There is only one pattern of class %d, skipping pattern average"%l[i])
+            ret[i,:] = x[i]
+    return ret
+
+
+def random_guided_warp(x, labels, slope_constraint="symmetric", use_window=True, dtw_type="normal", verbose=0):
+    # use verbose = -1 to turn off warnings
+    # slope_constraint is for DTW. "symmetric" or "asymmetric"
+    # dtw_type is for shapeDTW or DTW. "normal" or "shape"
+    
+    import dtw as dtw
+    
+    if use_window:
+        window = np.ceil(x.shape[1] / 10.).astype(int)
+    else:
+        window = None
+    orig_steps = np.arange(x.shape[1])
+    l = np.argmax(labels, axis=1) if labels.ndim > 1 else labels
+    
+    ret = np.zeros_like(x)
+    for i, pat in enumerate(tqdm(x)):
+        # guarentees that same one isnt selected
+        choices = np.delete(np.arange(x.shape[0]), i)
+        # remove ones of different classes
+        choices = np.where(l[choices] == l[i])[0]
+        if choices.size > 0:        
+            # pick random intra-class pattern
+            random_prototype = x[np.random.choice(choices)]
+            
+            if dtw_type == "shape":
+                path = dtw.shape_dtw(random_prototype, pat, dtw.RETURN_PATH, slope_constraint=slope_constraint, window=window)
+            else:
+                path = dtw.dtw(random_prototype, pat, dtw.RETURN_PATH, slope_constraint=slope_constraint, window=window)
+                            
+            # Time warp
+            warped = pat[path[1]]
+            for dim in range(x.shape[2]):
+                ret[i,:,dim] = np.interp(orig_steps, np.linspace(0, x.shape[1]-1., num=warped.shape[0]), warped[:,dim]).T
+        else:
+            if verbose > -1:
+                print("There is only one pattern of class %d, skipping timewarping"%l[i])
+            ret[i,:] = pat
+    return ret
+
+def random_guided_warp_shape(x, labels, slope_constraint="symmetric", use_window=True):
+    return random_guided_warp(x, labels, slope_constraint, use_window, dtw_type="shape")
+
+def discriminative_guided_warp(x, labels, batch_size=6, slope_constraint="symmetric", use_window=True, dtw_type="normal", use_variable_slice=True, verbose=0):
+    # use verbose = -1 to turn off warnings
+    # slope_constraint is for DTW. "symmetric" or "asymmetric"
+    # dtw_type is for shapeDTW or DTW. "normal" or "shape"
+    
+    import dtw as dtw
+    
+    if use_window:
+        window = np.ceil(x.shape[1] / 10.).astype(int)
+    else:
+        window = None
+    orig_steps = np.arange(x.shape[1])
+    l = np.argmax(labels, axis=1) if labels.ndim > 1 else labels
+    
+    positive_batch = np.ceil(batch_size / 2).astype(int)
+    negative_batch = np.floor(batch_size / 2).astype(int)
+        
+    ret = np.zeros_like(x)
+    warp_amount = np.zeros(x.shape[0])
+    for i, pat in enumerate(tqdm(x)):
+        # guarentees that same one isnt selected
+        choices = np.delete(np.arange(x.shape[0]), i)
+        
+        # remove ones of different classes
+        positive = np.where(l[choices] == l[i])[0]
+        negative = np.where(l[choices] != l[i])[0]
+        
+        if positive.size > 0 and negative.size > 0:
+            pos_k = min(positive.size, positive_batch)
+            neg_k = min(negative.size, negative_batch)
+            positive_prototypes = x[np.random.choice(positive, pos_k, replace=False)]
+            negative_prototypes = x[np.random.choice(negative, neg_k, replace=False)]
+                        
+            # vector embedding and nearest prototype in one
+            pos_aves = np.zeros((pos_k))
+            neg_aves = np.zeros((pos_k))
+            if dtw_type == "shape":
+                for p, pos_prot in enumerate(positive_prototypes):
+                    for ps, pos_samp in enumerate(positive_prototypes):
+                        if p != ps:
+                            pos_aves[p] += (1./(pos_k-1.))*dtw.shape_dtw(pos_prot, pos_samp, dtw.RETURN_VALUE, slope_constraint=slope_constraint, window=window)
+                    for ns, neg_samp in enumerate(negative_prototypes):
+                        neg_aves[p] += (1./neg_k)*dtw.shape_dtw(pos_prot, neg_samp, dtw.RETURN_VALUE, slope_constraint=slope_constraint, window=window)
+                selected_id = np.argmax(neg_aves - pos_aves)
+                path = dtw.shape_dtw(positive_prototypes[selected_id], pat, dtw.RETURN_PATH, slope_constraint=slope_constraint, window=window)
+            else:
+                for p, pos_prot in enumerate(positive_prototypes):
+                    for ps, pos_samp in enumerate(positive_prototypes):
+                        if p != ps:
+                            pos_aves[p] += (1./(pos_k-1.))*dtw.dtw(pos_prot, pos_samp, dtw.RETURN_VALUE, slope_constraint=slope_constraint, window=window)
+                    for ns, neg_samp in enumerate(negative_prototypes):
+                        neg_aves[p] += (1./neg_k)*dtw.dtw(pos_prot, neg_samp, dtw.RETURN_VALUE, slope_constraint=slope_constraint, window=window)
+                selected_id = np.argmax(neg_aves - pos_aves)
+                path = dtw.dtw(positive_prototypes[selected_id], pat, dtw.RETURN_PATH, slope_constraint=slope_constraint, window=window)
+                   
+            # Time warp
+            warped = pat[path[1]]
+            warp_path_interp = np.interp(orig_steps, np.linspace(0, x.shape[1]-1., num=warped.shape[0]), path[1])
+            warp_amount[i] = np.sum(np.abs(orig_steps-warp_path_interp))
+            for dim in range(x.shape[2]):
+                ret[i,:,dim] = np.interp(orig_steps, np.linspace(0, x.shape[1]-1., num=warped.shape[0]), warped[:,dim]).T
+        else:
+            if verbose > -1:
+                print("There is only one pattern of class %d"%l[i])
+            ret[i,:] = pat
+            warp_amount[i] = 0.
+    if use_variable_slice:
+        max_warp = np.max(warp_amount)
+        if max_warp == 0:
+            # unchanged
+            ret = window_slice(ret, reduce_ratio=0.9)
+        else:
+            for i, pat in enumerate(ret):
+                # Variable Sllicing
+                ret[i] = window_slice(pat[np.newaxis,:,:], reduce_ratio=0.9+0.1*warp_amount[i]/max_warp)[0]
+    return ret
+
+def discriminative_guided_warp_shape(x, labels, batch_size=6, slope_constraint="symmetric", use_window=True):
+    return discriminative_guided_warp(x, labels, batch_size, slope_constraint, use_window, dtw_type="shape")

time_series_classification/minirocket/src/dtw.py

@@ -0,0 +1,226 @@
+# Adapted from time_series_augmentation (https://github.com/uchidalab/time_series_augmentation)
+# Original: Apache License 2.0 by Brian Kenji Iwana and Seiichi Uchida
+
+__author__ = 'Brian Iwana'
+
+import numpy as np
+import math
+import sys
+
+RETURN_VALUE = 0
+RETURN_PATH = 1
+RETURN_ALL = -1
+
+# Core DTW
+def _traceback(DTW, slope_constraint):
+    i, j = np.array(DTW.shape) - 1
+    p, q = [i-1], [j-1]
+    
+    if slope_constraint == "asymmetric":
+        while (i > 1):
+            tb = np.argmin((DTW[i-1, j], DTW[i-1, j-1], DTW[i-1, j-2]))
+
+            if (tb == 0):
+                i = i - 1
+            elif (tb == 1):
+                i = i - 1
+                j = j - 1
+            elif (tb == 2):
+                i = i - 1
+                j = j - 2
+
+            p.insert(0, i-1)
+            q.insert(0, j-1)
+    elif slope_constraint == "symmetric":
+        while (i > 1 or j > 1):
+            tb = np.argmin((DTW[i-1, j-1], DTW[i-1, j], DTW[i, j-1]))
+
+            if (tb == 0):
+                i = i - 1
+                j = j - 1
+            elif (tb == 1):
+                i = i - 1
+            elif (tb == 2):
+                j = j - 1
+
+            p.insert(0, i-1)
+            q.insert(0, j-1)
+    else:
+        sys.exit("Unknown slope constraint %s"%slope_constraint)
+        
+    return (np.array(p), np.array(q))
+
+def dtw(prototype, sample, return_flag = RETURN_VALUE, slope_constraint="asymmetric", window=None):
+    """ Computes the DTW of two sequences.
+    :param prototype: np array [0..b]
+    :param sample: np array [0..t]
+    :param extended: bool
+    """
+    p = prototype.shape[0]
+    assert p != 0, "Prototype empty!"
+    s = sample.shape[0]
+    assert s != 0, "Sample empty!"
+    
+    if window is None:
+        window = s
+    
+    cost = np.full((p, s), np.inf)
+    for i in range(p):
+        start = max(0, i-window)
+        end = min(s, i+window)+1
+        cost[i,start:end]=np.linalg.norm(sample[start:end] - prototype[i], axis=1)
+
+    DTW = _cummulative_matrix(cost, slope_constraint, window)
+        
+    if return_flag == RETURN_ALL:
+        return DTW[-1,-1], cost, DTW[1:,1:], _traceback(DTW, slope_constraint)
+    elif return_flag == RETURN_PATH:
+        return _traceback(DTW, slope_constraint)
+    else:
+        return DTW[-1,-1]
+
+def _cummulative_matrix(cost, slope_constraint, window):
+    p = cost.shape[0]
+    s = cost.shape[1]
+    
+    # Note: DTW is one larger than cost and the original patterns
+    DTW = np.full((p+1, s+1), np.inf)
+
+    DTW[0, 0] = 0.0
+
+    if slope_constraint == "asymmetric":
+        for i in range(1, p+1):
+            if i <= window+1:
+                DTW[i,1] = cost[i-1,0] + min(DTW[i-1,0], DTW[i-1,1])
+            for j in range(max(2, i-window), min(s, i+window)+1):
+                DTW[i,j] = cost[i-1,j-1] + min(DTW[i-1,j-2], DTW[i-1,j-1], DTW[i-1,j])
+    elif slope_constraint == "symmetric":
+        for i in range(1, p+1):
+            for j in range(max(1, i-window), min(s, i+window)+1):
+                DTW[i,j] = cost[i-1,j-1] + min(DTW[i-1,j-1], DTW[i,j-1], DTW[i-1,j])
+    else:
+        sys.exit("Unknown slope constraint %s"%slope_constraint)
+        
+    return DTW
+
+def shape_dtw(prototype, sample, return_flag = RETURN_VALUE, slope_constraint="asymmetric", window=None, descr_ratio=0.05):
+    """ Computes the shapeDTW of two sequences.
+    :param prototype: np array [0..b]
+    :param sample: np array [0..t]
+    :param extended: bool
+    """
+    # shapeDTW
+    # https://www.sciencedirect.com/science/article/pii/S0031320317303710
+    
+    p = prototype.shape[0]
+    assert p != 0, "Prototype empty!"
+    s = sample.shape[0]
+    assert s != 0, "Sample empty!"
+    
+    if window is None:
+        window = s
+        
+    p_feature_len = np.clip(np.round(p * descr_ratio), 5, 100).astype(int)
+    s_feature_len = np.clip(np.round(s * descr_ratio), 5, 100).astype(int)
+    
+    # padding
+    p_pad_front = (np.ceil(p_feature_len / 2.)).astype(int)
+    p_pad_back = (np.floor(p_feature_len / 2.)).astype(int)
+    s_pad_front = (np.ceil(s_feature_len / 2.)).astype(int)
+    s_pad_back = (np.floor(s_feature_len / 2.)).astype(int)
+    
+    prototype_pad = np.pad(prototype, ((p_pad_front, p_pad_back), (0, 0)), mode="edge") 
+    sample_pad = np.pad(sample, ((s_pad_front, s_pad_back), (0, 0)), mode="edge") 
+    p_p = prototype_pad.shape[0]
+    s_p = sample_pad.shape[0]
+        
+    cost = np.full((p, s), np.inf)
+    for i in range(p):
+        for j in range(max(0, i-window), min(s, i+window)):
+            cost[i, j] = np.linalg.norm(sample_pad[j:j+s_feature_len] - prototype_pad[i:i+p_feature_len])
+            
+    DTW = _cummulative_matrix(cost, slope_constraint=slope_constraint, window=window)
+    
+    if return_flag == RETURN_ALL:
+        return DTW[-1,-1], cost, DTW[1:,1:], _traceback(DTW, slope_constraint)
+    elif return_flag == RETURN_PATH:
+        return _traceback(DTW, slope_constraint)
+    else:
+        return DTW[-1,-1]
+    
+# Draw helpers
+def draw_graph2d(cost, DTW, path, prototype, sample):
+    import matplotlib.pyplot as plt
+    plt.figure(figsize=(12, 8))
+   # plt.subplots_adjust(left=.02, right=.98, bottom=.001, top=.96, wspace=.05, hspace=.01)
+
+    #cost
+    plt.subplot(2, 3, 1)
+    plt.imshow(cost.T, cmap=plt.cm.gray, interpolation='none', origin='lower')
+    plt.plot(path[0], path[1], 'y')
+    plt.xlim((-0.5, cost.shape[0]-0.5))
+    plt.ylim((-0.5, cost.shape[0]-0.5))
+
+    #dtw
+    plt.subplot(2, 3, 2)
+    plt.imshow(DTW.T, cmap=plt.cm.gray, interpolation='none', origin='lower')
+    plt.plot(path[0]+1, path[1]+1, 'y')
+    plt.xlim((-0.5, DTW.shape[0]-0.5))
+    plt.ylim((-0.5, DTW.shape[0]-0.5))
+
+    #prototype
+    plt.subplot(2, 3, 4)
+    plt.plot(prototype[:,0], prototype[:,1], 'b-o')
+
+    #connection
+    plt.subplot(2, 3, 5)
+    for i in range(0,path[0].shape[0]):
+        plt.plot([prototype[path[0][i],0], sample[path[1][i],0]],[prototype[path[0][i],1], sample[path[1][i],1]], 'y-')
+    plt.plot(sample[:,0], sample[:,1], 'g-o')
+    plt.plot(prototype[:,0], prototype[:,1], 'b-o')
+
+    #sample
+    plt.subplot(2, 3, 6)
+    plt.plot(sample[:,0], sample[:,1], 'g-o')
+
+    plt.tight_layout()
+    plt.show()
+
+def draw_graph1d(cost, DTW, path, prototype, sample):
+    import matplotlib.pyplot as plt
+    plt.figure(figsize=(12, 8))
+   # plt.subplots_adjust(left=.02, right=.98, bottom=.001, top=.96, wspace=.05, hspace=.01)
+    p_steps = np.arange(prototype.shape[0])
+    s_steps = np.arange(sample.shape[0])
+
+    #cost
+    plt.subplot(2, 3, 1)
+    plt.imshow(cost.T, cmap=plt.cm.gray, interpolation='none', origin='lower')
+    plt.plot(path[0], path[1], 'y')
+    plt.xlim((-0.5, cost.shape[0]-0.5))
+    plt.ylim((-0.5, cost.shape[0]-0.5))
+
+    #dtw
+    plt.subplot(2, 3, 2)
+    plt.imshow(DTW.T, cmap=plt.cm.gray, interpolation='none', origin='lower')
+    plt.plot(path[0]+1, path[1]+1, 'y')
+    plt.xlim((-0.5, DTW.shape[0]-0.5))
+    plt.ylim((-0.5, DTW.shape[0]-0.5))
+
+    #prototype
+    plt.subplot(2, 3, 4)
+    plt.plot(p_steps, prototype[:,0], 'b-o')
+
+    #connection
+    plt.subplot(2, 3, 5)
+    for i in range(0,path[0].shape[0]):
+        plt.plot([path[0][i], path[1][i]],[prototype[path[0][i],0], sample[path[1][i],0]], 'y-')
+    plt.plot(p_steps, sample[:,0], 'g-o')
+    plt.plot(s_steps, prototype[:,0], 'b-o')
+
+    #sample
+    plt.subplot(2, 3, 6)
+    plt.plot(s_steps, sample[:,0], 'g-o')
+
+    plt.tight_layout()
+    plt.show()

time_series_classification/minirocket/src/hyperparameter_tune.py

@@ -0,0 +1,472 @@
+# Modified from MiniRocket (https://github.com/angus924/minirocket)
+# Original authors: Angus Dempster, Daniel F. Schmidt, Geoffrey I. Webb
+# Copyright (C) 2025 Jafar Bakhshaliyev
+# Licensed under GNU General Public License v3.0
+
+import numpy as np
+import pandas as pd
+import os
+import time
+import sys
+import argparse
+from sklearn.linear_model import RidgeClassifierCV
+from sklearn.metrics import accuracy_score
+import augmentation as aug
+from scipy.special import softmax
+from sklearn.metrics import log_loss
+from sklearn.model_selection import train_test_split
+from minirocket import fit, transform
+
+UCR_PATH = "" # Update this path to your environment
+
+
+def run_augmentation(x, y, args):
+    """
+    Apply data augmentation to the input data based on args.
+    
+    Parameters:
+    -----------
+    x : numpy.ndarray
+        Original time series data
+    y : numpy.ndarray
+        Original labels
+    args : argparse.Namespace
+        Command line arguments containing augmentation options
+        
+    Returns:
+    --------
+    x_aug : numpy.ndarray
+        Augmented time series data
+    y_aug : numpy.ndarray
+        Augmented labels
+    augmentation_tags : str
+        String describing the applied augmentations
+    """
+    print("Augmenting data for dataset %s" % args.dataset)
+
+    np.random.seed(args.seed)
+    x_aug = x.copy()
+    y_aug = y.copy()
+    
+    augmentation_tags = ""
+    
+    if args.augmentation_ratio > 0:
+        augmentation_tags = "%d" % args.augmentation_ratio
+        print(f"Original training size: {x.shape[0]} samples")
+        
+        for n in range(args.augmentation_ratio):
+            x_temp, current_tags = augment(x, y, args)
+            
+            x_temp = x_temp.astype(np.float32)
+            x_aug = np.vstack((x_aug, x_temp))
+            y_aug = np.append(y_aug, y)
+            
+            print(f"Round {n+1}: {current_tags} done - Added {x_temp.shape[0]} samples")
+            
+            if n == 0:
+                augmentation_tags += current_tags
+                
+        print(f"Augmented training size: {x_aug.shape[0]} samples")
+        print(f"Augmented data type: {x_aug.dtype}")
+        
+        if args.extra_tag:
+            augmentation_tags += "_" + args.extra_tag
+    else:
+        augmentation_tags = "none"
+        if args.extra_tag:
+            augmentation_tags = args.extra_tag
+    
+    x_aug = x_aug.astype(np.float32)
+
+    return x_aug, y_aug, augmentation_tags
+
+
+    
+def augment(x, y, args):
+    """
+    Apply specified augmentations to the data including SFCC.
+    
+    Parameters:
+    -----------
+    x : numpy.ndarray
+        Original time series data
+    y : numpy.ndarray
+        Original labels
+    args : argparse.Namespace
+        Command line arguments containing augmentation options
+        
+    Returns:
+    --------
+    x : numpy.ndarray
+        Augmented time series data
+    augmentation_tags : str
+        String describing the applied augmentations
+    """
+    augmentation_tags = ""
+    x_aug = x.copy()
+    
+    needs_reshape = False
+    original_shape = x_aug.shape
+    
+    if len(x_aug.shape) == 2:
+        # Reshape from (n_samples, timesteps) to (n_samples, timesteps, 1)
+        x_aug = x_aug.reshape(x_aug.shape[0], x_aug.shape[1], 1)
+        needs_reshape = True
+    
+    print('after needs reshape', x_aug.shape)
+    
+    
+    if args.jitter:
+        x_aug = aug.jitter(x_aug)
+        augmentation_tags += "_jitter"
+
+    if args.tps:
+        x_aug = aug.tps(x_aug, y, args.patch_len, args.stride, args.shuffle_rate)
+        augmentation_tags += "_tps"
+
+    if args.scaling:
+        x_aug = aug.scaling(x_aug)
+        augmentation_tags += "_scaling"
+        
+    if args.rotation:
+        x_aug = aug.rotation(x_aug)
+        augmentation_tags += "_rotation"
+        
+    if args.permutation:
+        x_aug = aug.permutation(x_aug)
+        augmentation_tags += "_permutation"
+        
+    if args.randompermutation:
+        x_aug = aug.permutation(x_aug, seg_mode="random")
+        augmentation_tags += "_randomperm"
+        
+    if args.magwarp:
+        x_aug = aug.magnitude_warp(x_aug)
+        augmentation_tags += "_magwarp"
+        
+    if args.timewarp:
+        x_aug = aug.time_warp(x_aug)
+        augmentation_tags += "_timewarp"
+        
+    if args.windowslice:
+        x_aug = aug.window_slice(x_aug)
+        augmentation_tags += "_windowslice"
+        
+    if args.windowwarp:
+        x_aug = aug.window_warp(x_aug)
+        augmentation_tags += "_windowwarp"
+        
+    if args.spawner:
+        x_aug = aug.spawner(x_aug, y)
+        augmentation_tags += "_spawner"
+        
+    if args.dtwwarp:
+        x_aug = aug.random_guided_warp(x_aug, y)
+        augmentation_tags += "_rgw"
+        
+    if args.shapedtwwarp:
+        x_aug = aug.random_guided_warp_shape(x_aug, y)
+        augmentation_tags += "_rgws"
+        
+    if args.wdba:
+        x_aug = aug.wdba(x_aug, y)
+        augmentation_tags += "_wdba"
+        
+    if args.discdtw:
+        x_aug = aug.discriminative_guided_warp(x_aug, y)
+        augmentation_tags += "_dgw"
+        
+    if args.discsdtw:
+        x_aug = aug.discriminative_guided_warp_shape(x_aug, y)
+        augmentation_tags += "_dgws"
+        
+    if needs_reshape:
+         x_aug = x_aug.reshape(original_shape)
+
+        
+    if not augmentation_tags:
+        augmentation_tags = "_none"
+    
+    return x_aug.astype(np.float32), augmentation_tags
+
+def load_ucr_dataset(dataset_name):
+    """
+    Load a UCR dataset from TSV files.
+    
+    Parameters:
+    -----------
+    dataset_name : str
+        Name of the dataset (e.g., 'FordB')
+    
+    Returns:
+    --------
+    X_train : numpy.ndarray
+        Training data (time series)
+    y_train : numpy.ndarray
+        Training labels
+    X_test : numpy.ndarray
+        Test data (time series)
+    y_test : numpy.ndarray
+        Test labels
+    """
+    # training data
+    train_file = os.path.join(UCR_PATH, dataset_name, f"{dataset_name}_TRAIN.tsv")
+    if not os.path.exists(train_file):
+        train_file = os.path.join(UCR_PATH, dataset_name, f"{dataset_name}_Train.tsv")
+    
+    # testing data
+    test_file = os.path.join(UCR_PATH, dataset_name, f"{dataset_name}_TEST.tsv")
+    if not os.path.exists(test_file):
+        test_file = os.path.join(UCR_PATH, dataset_name, f"{dataset_name}_Test.tsv")
+
+    if not os.path.exists(train_file) or not os.path.exists(test_file):
+        raise FileNotFoundError(f"Dataset files for {dataset_name} not found: {train_file}, {test_file}")
+    
+    print(f"Loading files: {train_file}, {test_file}")
+    
+    # Load data 
+    train_df = pd.read_csv(train_file, sep='\t', header=None)
+    test_df = pd.read_csv(test_file, sep='\t', header=None)
+    
+    y_train = train_df.iloc[:, 0].values
+    X_train = train_df.iloc[:, 1:].values
+    
+    y_test = test_df.iloc[:, 0].values
+    X_test = test_df.iloc[:, 1:].values
+    
+    unique_train, counts_train = np.unique(y_train, return_counts=True)
+    unique_test, counts_test = np.unique(y_test, return_counts=True)
+    
+    print(f"Train class distribution: {dict(zip(unique_train, counts_train))}")
+    print(f"Test class distribution: {dict(zip(unique_test, counts_test))}")
+    
+    X_train = X_train.astype(np.float32)
+    X_test = X_test.astype(np.float32)
+    
+    print(f"Data loaded successfully. Train shape: {X_train.shape}, Test shape: {X_test.shape}")
+    print(f"Train data type: {X_train.dtype}, Test data type: {X_test.dtype}")
+    
+    return X_train, y_train, X_test, y_test
+
+def run_minirocket_experiment(dataset_name, args):
+    """
+    Run MiniRocket on a UCR dataset with validation split for hyperparameter tuning.
+    
+    Parameters:
+    -----------
+    dataset_name : str
+        Name of the dataset (e.g., 'FordB')
+    args : argparse.Namespace
+        Command line arguments containing options
+    
+    Returns:
+    --------
+    mean_train_accuracy : float
+        Mean training accuracy across iterations
+    std_train_accuracy : float
+        Standard deviation of training accuracy across iterations
+    mean_val_accuracy : float
+        Mean validation accuracy across iterations
+    std_val_accuracy : float
+        Standard deviation of validation accuracy across iterations
+    """
+    # Load dataset
+    print(f"Loading dataset: {dataset_name}")
+    X_train_full, y_train_full, X_test, y_test = load_ucr_dataset(dataset_name)
+    
+    # Split original training set into train (80%) and validation (20%)
+    X_train, X_val, y_train, y_val = train_test_split(
+        X_train_full, y_train_full, 
+        test_size=0.2, 
+        random_state=args.seed,
+        stratify=y_train_full if len(np.unique(y_train_full)) > 1 else None
+    )
+    
+    print(f"Split training data: Train shape: {X_train.shape}, Validation shape: {X_val.shape}")
+    
+
+    if args.use_augmentation:
+        X_train_aug, y_train_aug, augmentation_tags = run_augmentation(X_train, y_train, args)
+    else:
+        X_train_aug, y_train_aug = X_train.copy(), y_train.copy()
+        augmentation_tags = "none"
+    
+    train_accuracies = []
+    val_accuracies = []
+    val_cross_entropies = []
+    runtimes = []
+    
+    for iteration in range(args.iterations):
+        print(f"Running iteration {iteration+1}/{args.iterations}")
+        
+        start_time = time.time()
+        
+        np.random.seed(args.seed + iteration)
+
+        parameters = fit(X_train_aug, num_features=args.features)
+
+        X_train_transform = transform(X_train_aug, parameters)
+        classifier = RidgeClassifierCV(alphas=np.logspace(-3, 3, 10))
+        classifier.fit(X_train_transform, y_train_aug)
+
+        train_predictions = classifier.predict(X_train_transform)
+        train_accuracy = accuracy_score(y_train_aug, train_predictions)
+        train_accuracies.append(train_accuracy)
+
+        X_val_transform = transform(X_val, parameters)
+        
+        val_predictions = classifier.predict(X_val_transform)
+        val_accuracy = accuracy_score(y_val, val_predictions)
+        val_accuracies.append(val_accuracy)
+        
+        # Get decision function values for validation set
+        val_decision_scores = classifier.decision_function(X_val_transform)
+        
+        if len(np.unique(y_val)) > 2:  # Multi-class
+            val_probabilities = softmax(val_decision_scores, axis=1)
+        else:  # Binary case
+            
+            val_scores = np.vstack([-val_decision_scores, val_decision_scores]).T
+            val_probabilities = softmax(val_scores, axis=1)
+        
+        try:
+            all_classes = np.unique(np.concatenate((y_train_aug, y_val)))
+            val_cross_entropy = log_loss(y_val, val_probabilities, labels=all_classes)
+            val_cross_entropies.append(val_cross_entropy)
+        except Exception as e:
+            print(f"Warning: Could not calculate cross-entropy: {e}")
+            val_cross_entropy = np.nan
+            val_cross_entropies.append(val_cross_entropy)
+        
+        runtime = time.time() - start_time
+        runtimes.append(runtime)
+        
+        print(f"Iteration {iteration+1} - Train Accuracy: {train_accuracy:.4f}")
+        print(f"Iteration {iteration+1} - Validation Accuracy: {val_accuracy:.4f}, Cross-Entropy: {val_cross_entropy:.4f}")
+        print(f"Iteration {iteration+1} - Runtime: {runtime:.2f} seconds")
+
+    mean_train_accuracy = np.mean(train_accuracies)
+    std_train_accuracy = np.std(train_accuracies)
+    mean_val_accuracy = np.mean(val_accuracies)
+    std_val_accuracy = np.std(val_accuracies)
+    mean_val_cross_entropy = np.mean(val_cross_entropies)
+    std_val_cross_entropy = np.std(val_cross_entropies)
+    mean_runtime = np.mean(runtimes)
+
+    print(f"\nResults for {dataset_name} with augmentation: {augmentation_tags}")
+    print(f"Train size (after augmentation): {X_train_aug.shape[0]} samples")
+    print(f"Validation size: {X_val.shape[0]} samples")
+    print(f"Mean Train Accuracy: {mean_train_accuracy:.4f} ± {std_train_accuracy:.4f}")
+    print(f"Mean Validation Accuracy: {mean_val_accuracy:.4f} ± {std_val_accuracy:.4f}")
+    print(f"Mean Validation Cross-Entropy: {mean_val_cross_entropy:.4f} ± {std_val_cross_entropy:.4f}")
+    print(f"Mean Runtime: {mean_runtime:.2f} seconds")
+
+    results_df = pd.DataFrame({
+        'Dataset': [dataset_name],
+        'Augmentation': [augmentation_tags],
+        'Train_Size': [X_train.shape[0]],
+        'Train_Size_After_Aug': [X_train_aug.shape[0]],
+        'Val_Size': [X_val.shape[0]],
+        'Mean_Train_Accuracy': [mean_train_accuracy],
+        'Train_Accuracy_STD': [std_train_accuracy],
+        'Mean_Val_Accuracy': [mean_val_accuracy],
+        'Val_Accuracy_STD': [std_val_accuracy],
+        'Mean_Val_Cross_Entropy': [mean_val_cross_entropy],
+        'Val_Cross_Entropy_STD': [std_val_cross_entropy],
+        'Mean_Runtime': [mean_runtime],
+        'Iterations': [args.iterations],
+        'Features': [args.features],
+        'Individual_Train_Accuracies': [','.join(map(str, train_accuracies))],
+        'Individual_Val_Accuracies': [','.join(map(str, val_accuracies))],
+        'Individual_Val_Cross_Entropies': [','.join(map(str, val_cross_entropies))],
+        'patch_len': [args.patch_len],
+        'stride': [args.stride],
+        'shuffle_rate': [args.shuffle_rate],
+        'sfcc_groups': [args.sfcc_groups]
+    })
+    
+    results_filename = f"hyperparameter_tuning_{dataset_name}_{augmentation_tags}.csv"
+    
+    if os.path.exists(results_filename):
+        existing_df = pd.read_csv(results_filename)
+        combined_df = pd.concat([existing_df, results_df], ignore_index=True)
+        combined_df.to_csv(results_filename, index=False)
+        print(f"Results appended to {results_filename}")
+    else:
+        results_df.to_csv(results_filename, index=False)
+        print(f"Results saved to new file {results_filename}")
+        
+    return mean_train_accuracy, std_train_accuracy, mean_val_accuracy, std_val_accuracy, mean_val_cross_entropy, std_val_cross_entropy, runtimes, augmentation_tags
+
+def list_ucr_datasets():
+    """List all available UCR datasets in the UCR_PATH directory"""
+    try:
+        datasets = [d for d in os.listdir(UCR_PATH) if os.path.isdir(os.path.join(UCR_PATH, d))]
+        return sorted(datasets)
+    except Exception as e:
+        print(f"Error listing datasets: {e}")
+        return []
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description='Hyperparameter Tuning for Time Series Augmentation')
+    
+    # Dataset selection
+    parser.add_argument('--dataset', type=str, help='Dataset name (default: FordB)')
+    parser.add_argument('--list', action='store_true', help='List available datasets')
+    
+    # MiniRocket parameters
+    parser.add_argument('--features', type=int, default=10000, help='Number of features (default: 10000)')
+    parser.add_argument('--iterations', type=int, default=5, help='Number of iterations (default: 5)')
+    parser.add_argument('--seed', type=int, default=42, help='Random seed (default: 42)')
+    
+    # Augmentation control
+    parser.add_argument('--use-augmentation', action='store_true', help='Use data augmentation')
+    parser.add_argument('--augmentation-ratio', type=int, default=0, 
+                        help='Number of augmented copies to add (default: 0)')
+    parser.add_argument('--extra-tag', type=str, default='', 
+                        help='Extra tag to add to augmentation tags')
+    
+    # Augmentation methods
+    parser.add_argument('--sfcc', action='store_true', help='Apply SFCC augmentation')
+    parser.add_argument('--sfcc_groups', type=int, default=4, help='Number of groups for SFCC augmentation (default: 4)')
+    parser.add_argument('--jitter', action='store_true', help='Apply jitter augmentation')
+    parser.add_argument('--scaling', action='store_true', help='Apply scaling augmentation')
+    parser.add_argument('--rotation', action='store_true', help='Apply rotation augmentation')
+    parser.add_argument('--permutation', action='store_true', help='Apply permutation augmentation')
+    parser.add_argument('--randompermutation', action='store_true', help='Apply random permutation augmentation')
+    parser.add_argument('--timewarp', action='store_true', help='Apply time warp augmentation')
+    parser.add_argument('--windowslice', action='store_true', help='Apply window slice augmentation')
+    parser.add_argument('--windowwarp', action='store_true', help='Apply window warp augmentation')
+    parser.add_argument('--spawner', action='store_true', help='Apply spawner augmentation')
+    parser.add_argument('--dtwwarp', action='store_true', help='Apply DTW-based warp augmentation')
+    parser.add_argument('--shapedtwwarp', action='store_true', help='Apply shape DTW warp augmentation')
+    parser.add_argument('--wdba', action='store_true', help='Apply WDBA augmentation')
+    parser.add_argument('--discdtw', action='store_true', help='Apply discriminative DTW augmentation')
+    parser.add_argument('--discsdtw', action='store_true', help='Apply discriminative shape DTW augmentation')
+    parser.add_argument('--tps', action='store_true', help='Apply TPS augmentation')
+    
+    parser.add_argument('--stride', type=int, default=2, help='# of patches stride (default: 2)')
+    parser.add_argument('--patch_len', type=int, default=10, help='Patch length (default: 10)')
+    parser.add_argument('--shuffle_rate', type=float, default=0.3, help='Shuffle rate (default: 0.3)')
+
+    args = parser.parse_args()
+    
+    if args.list:
+        datasets = list_ucr_datasets()
+        if datasets:
+            print("Available datasets:")
+            for dataset in datasets:
+                print(f"  - {dataset}")
+        else:
+            print("No datasets found or UCR_PATH is incorrect.")
+        sys.exit(0)
+    
+    # Run on specific dataset
+    dataset_name = args.dataset if args.dataset else "FordB"
+    
+    print(f"Running hyperparameter tuning on {dataset_name} dataset")
+    print(f"Using {args.features} features and {args.iterations} iterations")
+    if args.use_augmentation:
+        print(f"Using data augmentation with ratio {args.augmentation_ratio}")
+    
+    run_minirocket_experiment(dataset_name, args)

time_series_classification/minirocket/src/main.py

@@ -0,0 +1,491 @@
+# Modified from MiniRocket (https://github.com/angus924/minirocket)
+# Original authors: Angus Dempster, Daniel F. Schmidt, Geoffrey I. Webb
+# Copyright (C) 2025 Jafar Bakhshaliyev
+# Licensed under GNU General Public License v3.0
+
+import numpy as np
+import pandas as pd
+import os
+import time
+import sys
+import argparse
+from sklearn.linear_model import RidgeClassifierCV
+from sklearn.metrics import accuracy_score
+import augmentation as aug
+from minirocket import fit, transform
+
+
+UCR_PATH = ""  # Update this path for your environment
+
+
+def run_augmentation(x, y, args):
+    """
+    Apply data augmentation to the input data based on args.
+    
+    Parameters:
+    -----------
+    x : numpy.ndarray
+        Original time series data
+    y : numpy.ndarray
+        Original labels
+    args : argparse.Namespace
+        Command line arguments containing augmentation options
+        
+    Returns:
+    --------
+    x_aug : numpy.ndarray
+        Augmented time series data
+    y_aug : numpy.ndarray
+        Augmented labels
+    augmentation_tags : str
+        String describing the applied augmentations
+    """
+    print("Augmenting data for dataset %s" % args.dataset)
+
+    np.random.seed(args.seed)
+    x_aug = x.copy()
+    y_aug = y.copy()
+    start_time = time.time()
+    
+    augmentation_tags = ""
+    
+    if args.augmentation_ratio > 0:
+        augmentation_tags = "%d" % args.augmentation_ratio
+        print(f"Original training size: {x.shape[0]} samples")
+        
+        for n in range(args.augmentation_ratio):
+            x_temp, current_tags = augment(x, y, args)
+            
+            x_temp = x_temp.astype(np.float32)
+            x_aug = np.vstack((x_aug, x_temp))
+            y_aug = np.append(y_aug, y)
+            
+            print(f"Round {n+1}: {current_tags} done - Added {x_temp.shape[0]} samples")
+            
+            if n == 0:
+                augmentation_tags += current_tags
+                
+        print(f"Augmented training size: {x_aug.shape[0]} samples")
+        print(f"Augmented data type: {x_aug.dtype}")
+        
+        if args.extra_tag:
+            augmentation_tags += "_" + args.extra_tag
+    else:
+        augmentation_tags = "none"
+        if args.extra_tag:
+            augmentation_tags = args.extra_tag
+    
+    x_aug = x_aug.astype(np.float32)
+
+    time_dif = time.time() - start_time
+    with open("augmentation_time.txt", "a") as f:
+        f.write(f"Dataset: {args.dataset}, Augmentation tags: {augmentation_tags}, Time taken: {time_dif:.2f} seconds\n")
+    print(f"Data augmentation completed in {time_dif:.2f} seconds")
+            
+    return x_aug, y_aug, augmentation_tags
+
+    
+def augment(x, y, args):
+    """
+    Apply specified augmentations to the data.
+    
+    Parameters:
+    -----------
+    x : numpy.ndarray
+        Original time series data
+    y : numpy.ndarray
+        Original labels
+    args : argparse.Namespace
+        Command line arguments containing augmentation options
+        
+    Returns:
+    --------
+    x : numpy.ndarray
+        Augmented time series data
+    augmentation_tags : str
+        String describing the applied augmentations
+    """
+    augmentation_tags = ""
+
+    x_aug = x.copy()
+    
+    needs_reshape = False
+    original_shape = x_aug.shape
+
+    if len(x_aug.shape) == 2:
+        # Reshape from (n_samples, timesteps) to (n_samples, timesteps, 1)
+        x_aug = x_aug.reshape(x_aug.shape[0], x_aug.shape[1], 1)
+        needs_reshape = True
+    
+    if args.jitter:
+        x_aug = aug.jitter(x_aug)
+        augmentation_tags += "_jitter"
+
+    if args.tps:
+        x_aug = aug.tps(x_aug, y, args.patch_len, args.stride, args.shuffle_rate)
+        augmentation_tags += "_tps"
+        
+    if args.scaling:
+        x_aug = aug.scaling(x_aug)
+        augmentation_tags += "_scaling"
+        
+    if args.rotation:
+        x_aug = aug.rotation(x_aug)
+        augmentation_tags += "_rotation"
+        
+    if args.permutation:
+        x_aug = aug.permutation(x_aug)
+        augmentation_tags += "_permutation"
+        
+    if args.randompermutation:
+        x_aug = aug.permutation(x_aug, seg_mode="random")
+        augmentation_tags += "_randomperm"
+        
+    if args.magwarp:
+        x_aug = aug.magnitude_warp(x_aug)
+        augmentation_tags += "_magwarp"
+        
+    if args.timewarp:
+        x_aug = aug.time_warp(x_aug)
+        augmentation_tags += "_timewarp"
+        
+    if args.windowslice:
+        x_aug = aug.window_slice(x_aug)
+        augmentation_tags += "_windowslice"
+        
+    if args.windowwarp:
+        x_aug = aug.window_warp(x_aug)
+        augmentation_tags += "_windowwarp"
+        
+    if args.spawner:
+        x_aug = aug.spawner(x_aug, y)
+        augmentation_tags += "_spawner"
+        
+    if args.dtwwarp:
+        x_aug = aug.random_guided_warp(x_aug, y)
+        augmentation_tags += "_rgw"
+        
+    if args.shapedtwwarp:
+        x_aug = aug.random_guided_warp_shape(x_aug, y)
+        augmentation_tags += "_rgws"
+        
+    if args.wdba:
+        x_aug = aug.wdba(x_aug, y)
+        augmentation_tags += "_wdba"
+        
+    if args.discdtw:
+        x_aug = aug.discriminative_guided_warp(x_aug, y)
+        augmentation_tags += "_dgw"
+        
+    if args.discsdtw:
+        x_aug = aug.discriminative_guided_warp_shape(x_aug, y)
+        augmentation_tags += "_dgws"
+        
+    if needs_reshape:
+        x_aug = x_aug.reshape(original_shape)
+        
+    if not augmentation_tags:
+        augmentation_tags = "_none"
+    
+    return x_aug.astype(np.float32), augmentation_tags
+
+
+def load_ucr_dataset(dataset_name):
+    """
+    Load a UCR dataset from TSV files.
+    
+    Parameters:
+    -----------
+    dataset_name : str
+        Name of the dataset (e.g., 'FordB')
+    
+    Returns:
+    --------
+    X_train : numpy.ndarray
+        Training data (time series)
+    y_train : numpy.ndarray
+        Training labels
+    X_test : numpy.ndarray
+        Test data (time series)
+    y_test : numpy.ndarray
+        Test labels
+    """
+    # training data
+    train_file = os.path.join(UCR_PATH, dataset_name, f"{dataset_name}_TRAIN.tsv")
+    if not os.path.exists(train_file):
+        train_file = os.path.join(UCR_PATH, dataset_name, f"{dataset_name}_Train.tsv")
+    
+    # testing data
+    test_file = os.path.join(UCR_PATH, dataset_name, f"{dataset_name}_TEST.tsv")
+    if not os.path.exists(test_file):
+        test_file = os.path.join(UCR_PATH, dataset_name, f"{dataset_name}_Test.tsv")
+    
+    if not os.path.exists(train_file) or not os.path.exists(test_file):
+        raise FileNotFoundError(f"Dataset files for {dataset_name} not found: {train_file}, {test_file}")
+    
+    print(f"Loading files: {train_file}, {test_file}")
+    
+    # Load data
+    train_df = pd.read_csv(train_file, sep='\t', header=None)
+    test_df = pd.read_csv(test_file, sep='\t', header=None)
+    
+    y_train = train_df.iloc[:, 0].values
+    X_train = train_df.iloc[:, 1:].values
+    
+    y_test = test_df.iloc[:, 0].values
+    X_test = test_df.iloc[:, 1:].values
+    
+    unique_train, counts_train = np.unique(y_train, return_counts=True)
+    unique_test, counts_test = np.unique(y_test, return_counts=True)
+    
+    print(f"Train class distribution: {dict(zip(unique_train, counts_train))}")
+    print(f"Test class distribution: {dict(zip(unique_test, counts_test))}")
+    
+    X_train = X_train.astype(np.float32)
+    X_test = X_test.astype(np.float32)
+    
+    print(f"Data loaded successfully. Train shape: {X_train.shape}, Test shape: {X_test.shape}")
+    print(f"Train data type: {X_train.dtype}, Test data type: {X_test.dtype}")
+    
+    return X_train, y_train, X_test, y_test
+
+def run_minirocket_experiment(dataset_name, args):
+    """
+    Run MiniRocket on a UCR dataset with multiple iterations and optional augmentation.
+    
+    Parameters:
+    -----------
+    dataset_name : str
+        Name of the dataset (e.g., 'FordB')
+    args : argparse.Namespace
+        Command line arguments containing options
+    
+    Returns:
+    --------
+    mean_accuracy : float
+        Mean accuracy across iterations
+    std_accuracy : float
+        Standard deviation of accuracy across iterations
+    """
+
+    print(f"Loading dataset: {dataset_name}")
+    X_train, y_train, X_test, y_test = load_ucr_dataset(dataset_name)
+    
+    # Apply augmentation 
+    if args.use_augmentation:
+        X_train, y_train, augmentation_tags = run_augmentation(X_train, y_train, args)
+    else:
+        augmentation_tags = "none"
+    
+    accuracies = []
+    runtimes = []
+    
+    for iteration in range(args.iterations):
+        print(f"Running iteration {iteration+1}/{args.iterations}")
+        
+        start_time = time.time()
+        
+        np.random.seed(args.seed + iteration)
+        
+        parameters = fit(X_train, num_features=args.features)
+        X_train_transform = transform(X_train, parameters)
+        classifier = RidgeClassifierCV(alphas=np.logspace(-3, 3, 10))
+        classifier.fit(X_train_transform, y_train)
+        X_test_transform = transform(X_test, parameters)
+        predictions = classifier.predict(X_test_transform)
+        
+        accuracy = accuracy_score(y_test, predictions)
+        runtime = time.time() - start_time
+        
+        accuracies.append(accuracy)
+        runtimes.append(runtime)
+        
+        print(f"Iteration {iteration+1} - Accuracy: {accuracy:.4f}, Runtime: {runtime:.2f} seconds")
+    
+
+    mean_accuracy = np.mean(accuracies)
+    std_accuracy = np.std(accuracies)
+    mean_runtime = np.mean(runtimes)
+    
+    print(f"\nResults for {dataset_name} with augmentation: {augmentation_tags}")
+    print(f"Train size: {X_train.shape[0]} samples")
+    print(f"Mean Accuracy: {mean_accuracy:.4f} ± {std_accuracy:.4f}")
+    print(f"Mean Runtime: {mean_runtime:.2f} seconds")
+    print(f"Individual Accuracies: {accuracies}")
+    
+    results_df = pd.DataFrame({
+        'Dataset': [dataset_name],
+        'Augmentation': [augmentation_tags],
+        'Train_Size': [X_train.shape[0]],
+        'Test_Size': [X_test.shape[0]],
+        'Mean_Accuracy': [mean_accuracy],
+        'Std_Dev': [std_accuracy],
+        'Mean_Runtime': [mean_runtime],
+        'Iterations': [args.iterations],
+        'Features': [args.features],
+        'Individual_Accuracies': [','.join(map(str, accuracies))],
+        'patch_len': [args.patch_len],
+        'stride': [args.stride],
+        'shuffle_rate': [args.shuffle_rate],
+    })
+    
+    results_filename = f"minirocket_results_{dataset_name}_{augmentation_tags}.csv"
+
+    if os.path.exists(results_filename):
+        existing_df = pd.read_csv(results_filename)
+        combined_df = pd.concat([existing_df, results_df], ignore_index=True)
+        combined_df.to_csv(results_filename, index=False)
+        print(f"Results appended to {results_filename}")
+    else:
+        results_df.to_csv(results_filename, index=False)
+        print(f"Results saved to new file {results_filename}")
+        
+    return mean_accuracy, std_accuracy, runtimes, augmentation_tags
+
+def run_all_datasets(args):
+    """
+    Run MiniRocket on all available datasets.
+    
+    Parameters:
+    -----------
+    args : argparse.Namespace
+        Command line arguments containing options
+    """
+    # list of available datasets
+    datasets = [d for d in os.listdir(UCR_PATH) if os.path.isdir(os.path.join(UCR_PATH, d))]
+    
+    if not datasets:
+        print(f"No datasets found in {UCR_PATH}")
+        return
+    
+    print(f"Found {len(datasets)} datasets: {', '.join(datasets)}")
+    
+    results = []
+    
+    for dataset_name in datasets:
+        print(f"\n{'='*50}")
+        print(f"Processing dataset: {dataset_name}")
+        print(f"{'='*50}")
+        
+        try:
+            mean_acc, std_acc, _, aug_tags = run_minirocket_experiment(dataset_name, args)
+            
+            results.append({
+                'Dataset': dataset_name,
+                'Augmentation': aug_tags,
+                'Mean_Accuracy': mean_acc,
+                'Std_Dev': std_acc,
+                'patch_len': args.patch_len,
+                'stride': args.stride,
+                'shuffle_rate': args.shuffle_rate
+            })
+        
+        except Exception as e:
+            print(f"Error processing dataset {dataset_name}: {e}")
+    
+
+    if results:
+        results_df = pd.DataFrame(results)
+        overall_mean = results_df['Mean_Accuracy'].mean()
+        
+        print("\n" + "="*70)
+        print("SUMMARY OF RESULTS")
+        print("="*70)
+        print(f"{'Dataset':<25} {'Augmentation':<25} {'Mean Accuracy':<15} {'Std Dev':<10}")
+        print("-"*70)
+        
+        for _, row in results_df.iterrows():
+            print(f"{row['Dataset']:<25} {row['Augmentation']:<25} {row['Mean_Accuracy']:.4f}{' '*8} {row['Std_Dev']:.4f}")
+        
+        print("-"*70)
+        print(f"{'OVERALL':<25} {'':<25} {overall_mean:.4f}")
+        print("="*70)
+        
+
+        aug_tag = "none" if not args.use_augmentation else "aug"
+        results_df.to_csv(f"minirocket_summary_results_{aug_tag}.csv", index=False)
+        print(f"\nSummary results saved to minirocket_summary_results_{aug_tag}.csv")
+
+def list_ucr_datasets():
+    """List all available UCR datasets in the UCR_PATH directory"""
+    try:
+        datasets = [d for d in os.listdir(UCR_PATH) if os.path.isdir(os.path.join(UCR_PATH, d))]
+        return sorted(datasets)
+    except Exception as e:
+        print(f"Error listing datasets: {e}")
+        return []
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description='Run MiniRocket on UCR datasets with optional augmentation')
+    
+    # Dataset selection
+    parser.add_argument('--dataset', type=str, help='Dataset name (default: FordB)')
+    parser.add_argument('--list', action='store_true', help='List available datasets')
+    parser.add_argument('--all', action='store_true', help='Run on all available datasets')
+    
+    # MiniRocket parameters
+    parser.add_argument('--features', type=int, default=10000, help='Number of features (default: 10000)')
+    parser.add_argument('--iterations', type=int, default=5, help='Number of iterations (default: 5)')
+    parser.add_argument('--seed', type=int, default=42, help='Random seed (default: 42)')
+    
+    # Augmentation control
+    parser.add_argument('--use-augmentation', action='store_true', help='Use data augmentation')
+    parser.add_argument('--augmentation-ratio', type=int, default=0, 
+                        help='Number of augmented copies to add (default: 0)')
+    parser.add_argument('--extra-tag', type=str, default='', 
+                        help='Extra tag to add to augmentation tags')
+    
+    # Augmentation methods
+    parser.add_argument('--jitter', action='store_true', help='Apply jitter augmentation')
+    parser.add_argument('--scaling', action='store_true', help='Apply scaling augmentation')
+    parser.add_argument('--rotation', action='store_true', help='Apply rotation augmentation')
+    parser.add_argument('--permutation', action='store_true', help='Apply permutation augmentation')
+    parser.add_argument('--randompermutation', action='store_true', help='Apply random permutation augmentation')
+    parser.add_argument('--magwarp', action='store_true', help='Apply magnitude warp augmentation')
+    parser.add_argument('--timewarp', action='store_true', help='Apply time warp augmentation')
+    parser.add_argument('--windowslice', action='store_true', help='Apply window slice augmentation')
+    parser.add_argument('--windowwarp', action='store_true', help='Apply window warp augmentation')
+    parser.add_argument('--spawner', action='store_true', help='Apply spawner augmentation')
+    parser.add_argument('--dtwwarp', action='store_true', help='Apply DTW-based warp augmentation')
+    parser.add_argument('--shapedtwwarp', action='store_true', help='Apply shape DTW warp augmentation')
+    parser.add_argument('--wdba', action='store_true', help='Apply WDBA augmentation')
+    parser.add_argument('--discdtw', action='store_true', help='Apply discriminative DTW augmentation')
+    parser.add_argument('--discsdtw', action='store_true', help='Apply discriminative shape DTW augmentation')
+    parser.add_argument('--tps', action='store_true', help='Apply TPS augmentation')
+
+    parser.add_argument('--stride', type=int, default=0, help='# of patches stride')
+    parser.add_argument('--patch_len', type=int, default=0, help='# of patches')
+    parser.add_argument('--shuffle_rate', type=float, default=0.0, help='shuffle rate')
+
+
+    
+    args = parser.parse_args()
+    
+
+    if args.list:
+        datasets = list_ucr_datasets()
+        if datasets:
+            print("Available datasets:")
+            for dataset in datasets:
+                print(f"  - {dataset}")
+        else:
+            print("No datasets found or UCR_PATH is incorrect.")
+        sys.exit(0)
+    
+    # Run on all datasets
+    if args.all:
+        print(f"Running MiniRocket on all available datasets")
+        print(f"Using {args.features} features and {args.iterations} iterations")
+        if args.use_augmentation:
+            print(f"Using data augmentation with ratio {args.augmentation_ratio}")
+        run_all_datasets(args)
+        sys.exit(0)
+    
+    # Run on specific dataset
+    dataset_name = args.dataset if args.dataset else "FordB"
+    print(f"Running MiniRocket on {dataset_name} dataset")
+    print(f"Using {args.features} features and {args.iterations} iterations")
+    if args.use_augmentation:
+        print(f"Using data augmentation with ratio {args.augmentation_ratio}")
+    
+    run_minirocket_experiment(dataset_name, args)

time_series_classification/minirocket/src/minirocket.py

@@ -0,0 +1,234 @@
+# Angus Dempster, Daniel F Schmidt, Geoffrey I Webb
+
+# MiniRocket: A Very Fast (Almost) Deterministic Transform for Time Series
+# Classification
+
+# https://arxiv.org/abs/2012.08791
+
+from numba import njit, prange, vectorize
+import numpy as np
+
+@njit("float32[:](float32[:,:],int32[:],int32[:],float32[:])", fastmath = True, parallel = False, cache = True)
+def _fit_biases(X, dilations, num_features_per_dilation, quantiles):
+
+    num_examples, input_length = X.shape
+
+    # equivalent to:
+    # >>> from itertools import combinations
+    # >>> indices = np.array([_ for _ in combinations(np.arange(9), 3)], dtype = np.int32)
+    indices = np.array((
+        0,1,2,0,1,3,0,1,4,0,1,5,0,1,6,0,1,7,0,1,8,
+        0,2,3,0,2,4,0,2,5,0,2,6,0,2,7,0,2,8,0,3,4,
+        0,3,5,0,3,6,0,3,7,0,3,8,0,4,5,0,4,6,0,4,7,
+        0,4,8,0,5,6,0,5,7,0,5,8,0,6,7,0,6,8,0,7,8,
+        1,2,3,1,2,4,1,2,5,1,2,6,1,2,7,1,2,8,1,3,4,
+        1,3,5,1,3,6,1,3,7,1,3,8,1,4,5,1,4,6,1,4,7,
+        1,4,8,1,5,6,1,5,7,1,5,8,1,6,7,1,6,8,1,7,8,
+        2,3,4,2,3,5,2,3,6,2,3,7,2,3,8,2,4,5,2,4,6,
+        2,4,7,2,4,8,2,5,6,2,5,7,2,5,8,2,6,7,2,6,8,
+        2,7,8,3,4,5,3,4,6,3,4,7,3,4,8,3,5,6,3,5,7,
+        3,5,8,3,6,7,3,6,8,3,7,8,4,5,6,4,5,7,4,5,8,
+        4,6,7,4,6,8,4,7,8,5,6,7,5,6,8,5,7,8,6,7,8
+    ), dtype = np.int32).reshape(84, 3)
+
+    num_kernels = len(indices)
+    num_dilations = len(dilations)
+
+    num_features = num_kernels * np.sum(num_features_per_dilation)
+
+    biases = np.zeros(num_features, dtype = np.float32)
+
+    feature_index_start = 0
+
+    for dilation_index in range(num_dilations):
+
+        dilation = dilations[dilation_index]
+        padding = ((9 - 1) * dilation) // 2
+
+        num_features_this_dilation = num_features_per_dilation[dilation_index]
+
+        for kernel_index in range(num_kernels):
+
+            feature_index_end = feature_index_start + num_features_this_dilation
+
+            _X = X[np.random.randint(num_examples)]
+
+            A = -_X          # A = alpha * X = -X
+            G = _X + _X + _X # G = gamma * X = 3X
+
+            C_alpha = np.zeros(input_length, dtype = np.float32)
+            C_alpha[:] = A
+
+            C_gamma = np.zeros((9, input_length), dtype = np.float32)
+            C_gamma[9 // 2] = G
+
+            start = dilation
+            end = input_length - padding
+
+            for gamma_index in range(9 // 2):
+
+                C_alpha[-end:] = C_alpha[-end:] + A[:end]
+                C_gamma[gamma_index, -end:] = G[:end]
+
+                end += dilation
+
+            for gamma_index in range(9 // 2 + 1, 9):
+
+                C_alpha[:-start] = C_alpha[:-start] + A[start:]
+                C_gamma[gamma_index, :-start] = G[start:]
+
+                start += dilation
+
+            index_0, index_1, index_2 = indices[kernel_index]
+
+            C = C_alpha + C_gamma[index_0] + C_gamma[index_1] + C_gamma[index_2]
+
+            biases[feature_index_start:feature_index_end] = np.quantile(C, quantiles[feature_index_start:feature_index_end])
+
+            feature_index_start = feature_index_end
+
+    return biases
+
+def _fit_dilations(input_length, num_features, max_dilations_per_kernel):
+
+    num_kernels = 84
+
+    num_features_per_kernel = num_features // num_kernels
+    true_max_dilations_per_kernel = min(num_features_per_kernel, max_dilations_per_kernel)
+    multiplier = num_features_per_kernel / true_max_dilations_per_kernel
+
+    max_exponent = np.log2((input_length - 1) / (9 - 1))
+    dilations, num_features_per_dilation = \
+    np.unique(np.logspace(0, max_exponent, true_max_dilations_per_kernel, base = 2).astype(np.int32), return_counts = True)
+    num_features_per_dilation = (num_features_per_dilation * multiplier).astype(np.int32) # this is a vector
+
+    remainder = num_features_per_kernel - np.sum(num_features_per_dilation)
+    i = 0
+    while remainder > 0:
+        num_features_per_dilation[i] += 1
+        remainder -= 1
+        i = (i + 1) % len(num_features_per_dilation)
+
+    return dilations, num_features_per_dilation
+
+# low-discrepancy sequence to assign quantiles to kernel/dilation combinations
+def _quantiles(n):
+    return np.array([(_ * ((np.sqrt(5) + 1) / 2)) % 1 for _ in range(1, n + 1)], dtype = np.float32)
+
+def fit(X, num_features = 10_000, max_dilations_per_kernel = 32):
+
+    _, input_length = X.shape
+
+    num_kernels = 84
+
+    dilations, num_features_per_dilation = _fit_dilations(input_length, num_features, max_dilations_per_kernel)
+
+    num_features_per_kernel = np.sum(num_features_per_dilation)
+
+    quantiles = _quantiles(num_kernels * num_features_per_kernel)
+
+    biases = _fit_biases(X, dilations, num_features_per_dilation, quantiles)
+
+    return dilations, num_features_per_dilation, biases
+
+# _PPV(C, b).mean() returns PPV for vector C (convolution output) and scalar b (bias)
+@vectorize("float32(float32,float32)", nopython = True, cache = True)
+def _PPV(a, b):
+    if a > b:
+        return 1
+    else:
+        return 0
+
+@njit("float32[:,:](float32[:,:],Tuple((int32[:],int32[:],float32[:])))", fastmath = True, parallel = True, cache = True)
+def transform(X, parameters):
+
+    num_examples, input_length = X.shape
+
+    dilations, num_features_per_dilation, biases = parameters
+
+    # equivalent to:
+    # >>> from itertools import combinations
+    # >>> indices = np.array([_ for _ in combinations(np.arange(9), 3)], dtype = np.int32)
+    indices = np.array((
+        0,1,2,0,1,3,0,1,4,0,1,5,0,1,6,0,1,7,0,1,8,
+        0,2,3,0,2,4,0,2,5,0,2,6,0,2,7,0,2,8,0,3,4,
+        0,3,5,0,3,6,0,3,7,0,3,8,0,4,5,0,4,6,0,4,7,
+        0,4,8,0,5,6,0,5,7,0,5,8,0,6,7,0,6,8,0,7,8,
+        1,2,3,1,2,4,1,2,5,1,2,6,1,2,7,1,2,8,1,3,4,
+        1,3,5,1,3,6,1,3,7,1,3,8,1,4,5,1,4,6,1,4,7,
+        1,4,8,1,5,6,1,5,7,1,5,8,1,6,7,1,6,8,1,7,8,
+        2,3,4,2,3,5,2,3,6,2,3,7,2,3,8,2,4,5,2,4,6,
+        2,4,7,2,4,8,2,5,6,2,5,7,2,5,8,2,6,7,2,6,8,
+        2,7,8,3,4,5,3,4,6,3,4,7,3,4,8,3,5,6,3,5,7,
+        3,5,8,3,6,7,3,6,8,3,7,8,4,5,6,4,5,7,4,5,8,
+        4,6,7,4,6,8,4,7,8,5,6,7,5,6,8,5,7,8,6,7,8
+    ), dtype = np.int32).reshape(84, 3)
+
+    num_kernels = len(indices)
+    num_dilations = len(dilations)
+
+    num_features = num_kernels * np.sum(num_features_per_dilation)
+
+    features = np.zeros((num_examples, num_features), dtype = np.float32)
+
+    for example_index in prange(num_examples):
+
+        _X = X[example_index]
+
+        A = -_X          # A = alpha * X = -X
+        G = _X + _X + _X # G = gamma * X = 3X
+
+        feature_index_start = 0
+
+        for dilation_index in range(num_dilations):
+
+            _padding0 = dilation_index % 2
+
+            dilation = dilations[dilation_index]
+            padding = ((9 - 1) * dilation) // 2
+
+            num_features_this_dilation = num_features_per_dilation[dilation_index]
+
+            C_alpha = np.zeros(input_length, dtype = np.float32)
+            C_alpha[:] = A
+
+            C_gamma = np.zeros((9, input_length), dtype = np.float32)
+            C_gamma[9 // 2] = G
+
+            start = dilation
+            end = input_length - padding
+
+            for gamma_index in range(9 // 2):
+
+                C_alpha[-end:] = C_alpha[-end:] + A[:end]
+                C_gamma[gamma_index, -end:] = G[:end]
+
+                end += dilation
+
+            for gamma_index in range(9 // 2 + 1, 9):
+
+                C_alpha[:-start] = C_alpha[:-start] + A[start:]
+                C_gamma[gamma_index, :-start] = G[start:]
+
+                start += dilation
+
+            for kernel_index in range(num_kernels):
+
+                feature_index_end = feature_index_start + num_features_this_dilation
+
+                _padding1 = (_padding0 + kernel_index) % 2
+
+                index_0, index_1, index_2 = indices[kernel_index]
+
+                C = C_alpha + C_gamma[index_0] + C_gamma[index_1] + C_gamma[index_2]
+
+                if _padding1 == 0:
+                    for feature_count in range(num_features_this_dilation):
+                        features[example_index, feature_index_start + feature_count] = _PPV(C, biases[feature_index_start + feature_count]).mean()
+                else:
+                    for feature_count in range(num_features_this_dilation):
+                        features[example_index, feature_index_start + feature_count] = _PPV(C[padding:-padding], biases[feature_index_start + feature_count]).mean()
+
+                feature_index_start = feature_index_end
+
+    return features

time_series_classification/minirocket/src/minirocket_dv.py

@@ -0,0 +1,126 @@
+# Angus Dempster, Daniel F Schmidt, Geoffrey I Webb
+
+# MiniRocket: A Very Fast (Almost) Deterministic Transform for Time Series
+# Classification
+
+# https://arxiv.org/abs/2012.08791
+
+from numba import njit
+import numpy as np
+
+from minirocket import _PPV, _fit_dilations, _quantiles
+
+@njit("Tuple((float32[:],float32[:,:]))(float32[:,:],int32[:],int32[:],float32[:])", fastmath = True, parallel = False, cache = True)
+def _fit_biases_transform(X, dilations, num_features_per_dilation, quantiles):
+
+    num_examples, input_length = X.shape
+
+    # equivalent to:
+    # >>> from itertools import combinations
+    # >>> indices = np.array([_ for _ in combinations(np.arange(9), 3)], dtype = np.int32)
+    indices = np.array((
+        0,1,2,0,1,3,0,1,4,0,1,5,0,1,6,0,1,7,0,1,8,
+        0,2,3,0,2,4,0,2,5,0,2,6,0,2,7,0,2,8,0,3,4,
+        0,3,5,0,3,6,0,3,7,0,3,8,0,4,5,0,4,6,0,4,7,
+        0,4,8,0,5,6,0,5,7,0,5,8,0,6,7,0,6,8,0,7,8,
+        1,2,3,1,2,4,1,2,5,1,2,6,1,2,7,1,2,8,1,3,4,
+        1,3,5,1,3,6,1,3,7,1,3,8,1,4,5,1,4,6,1,4,7,
+        1,4,8,1,5,6,1,5,7,1,5,8,1,6,7,1,6,8,1,7,8,
+        2,3,4,2,3,5,2,3,6,2,3,7,2,3,8,2,4,5,2,4,6,
+        2,4,7,2,4,8,2,5,6,2,5,7,2,5,8,2,6,7,2,6,8,
+        2,7,8,3,4,5,3,4,6,3,4,7,3,4,8,3,5,6,3,5,7,
+        3,5,8,3,6,7,3,6,8,3,7,8,4,5,6,4,5,7,4,5,8,
+        4,6,7,4,6,8,4,7,8,5,6,7,5,6,8,5,7,8,6,7,8
+    ), dtype = np.int32).reshape(84, 3)
+
+    num_kernels = len(indices)
+    num_dilations = len(dilations)
+
+    num_features = num_kernels * np.sum(num_features_per_dilation)
+
+    biases = np.zeros(num_features, dtype = np.float32)
+
+    features = np.zeros((num_examples, num_features), dtype = np.float32)
+
+    feature_index_start = 0
+
+    for dilation_index in range(num_dilations):
+
+        _padding0 = dilation_index % 2
+
+        dilation = dilations[dilation_index]
+        padding = ((9 - 1) * dilation) // 2
+
+        num_features_this_dilation = num_features_per_dilation[dilation_index]
+
+        for kernel_index in range(num_kernels):
+
+            feature_index_end = feature_index_start + num_features_this_dilation
+
+            _padding1 = (_padding0 + kernel_index) % 2
+
+            index_0, index_1, index_2 = indices[kernel_index]
+
+            C = np.zeros((num_examples, input_length), dtype = np.float32)
+
+            for example_index in range(num_examples):
+
+                _X = X[example_index]
+
+                A = -_X          # A = alpha * X = -X
+                G = _X + _X + _X # G = gamma * X = 3X
+
+                C_alpha = np.zeros(input_length, dtype = np.float32)
+                C_alpha[:] = A
+
+                C_gamma = np.zeros((9, input_length), dtype = np.float32)
+                C_gamma[9 // 2] = G
+
+                start = dilation
+                end = input_length - padding
+
+                for gamma_index in range(9 // 2):
+
+                    C_alpha[-end:] = C_alpha[-end:] + A[:end]
+                    C_gamma[gamma_index, -end:] = G[:end]
+
+                    end += dilation
+
+                for gamma_index in range(9 // 2 + 1, 9):
+
+                    C_alpha[:-start] = C_alpha[:-start] + A[start:]
+                    C_gamma[gamma_index, :-start] = G[start:]
+
+                    start += dilation
+
+                C[example_index] = C_alpha + C_gamma[index_0] + C_gamma[index_1] + C_gamma[index_2]
+
+            biases[feature_index_start:feature_index_end] = np.quantile(C, quantiles[feature_index_start:feature_index_end])
+
+            for example_index in range(num_examples):
+                if _padding1 == 0:
+                    for feature_count in range(num_features_this_dilation):
+                        features[example_index, feature_index_start + feature_count] = _PPV(C[example_index], biases[feature_index_start + feature_count]).mean()
+                else:
+                    for feature_count in range(num_features_this_dilation):
+                        features[example_index, feature_index_start + feature_count] = _PPV(C[example_index][padding:-padding], biases[feature_index_start + feature_count]).mean()
+
+            feature_index_start = feature_index_end
+
+    return biases, features
+
+def fit_transform(X, num_features = 10_000, max_dilations_per_kernel = 32):
+
+    _, input_length = X.shape
+
+    num_kernels = 84
+
+    dilations, num_features_per_dilation = _fit_dilations(input_length, num_features, max_dilations_per_kernel)
+
+    num_features_per_kernel = np.sum(num_features_per_dilation)
+
+    quantiles = _quantiles(num_kernels * num_features_per_kernel)
+
+    biases, features = _fit_biases_transform(X, dilations, num_features_per_dilation, quantiles)
+
+    return (dilations, num_features_per_dilation, biases), features

time_series_classification/minirocket/src/minirocket_multivariate.py

@@ -0,0 +1,283 @@
+# Angus Dempster, Daniel F Schmidt, Geoffrey I Webb
+
+# MiniRocket: A Very Fast (Almost) Deterministic Transform for Time Series
+# Classification
+
+# https://arxiv.org/abs/2012.08791
+
+# ** This is a naive extension of MiniRocket to multivariate time series. **
+
+from numba import njit, prange, vectorize
+import numpy as np
+
+@njit("float32[:](float32[:,:,:],int32[:],int32[:],int32[:],int32[:],float32[:])", fastmath = True, parallel = False, cache = True)
+def _fit_biases(X, num_channels_per_combination, channel_indices, dilations, num_features_per_dilation, quantiles):
+
+    num_examples, num_channels, input_length = X.shape
+
+    # equivalent to:
+    # >>> from itertools import combinations
+    # >>> indices = np.array([_ for _ in combinations(np.arange(9), 3)], dtype = np.int32)
+    indices = np.array((
+        0,1,2,0,1,3,0,1,4,0,1,5,0,1,6,0,1,7,0,1,8,
+        0,2,3,0,2,4,0,2,5,0,2,6,0,2,7,0,2,8,0,3,4,
+        0,3,5,0,3,6,0,3,7,0,3,8,0,4,5,0,4,6,0,4,7,
+        0,4,8,0,5,6,0,5,7,0,5,8,0,6,7,0,6,8,0,7,8,
+        1,2,3,1,2,4,1,2,5,1,2,6,1,2,7,1,2,8,1,3,4,
+        1,3,5,1,3,6,1,3,7,1,3,8,1,4,5,1,4,6,1,4,7,
+        1,4,8,1,5,6,1,5,7,1,5,8,1,6,7,1,6,8,1,7,8,
+        2,3,4,2,3,5,2,3,6,2,3,7,2,3,8,2,4,5,2,4,6,
+        2,4,7,2,4,8,2,5,6,2,5,7,2,5,8,2,6,7,2,6,8,
+        2,7,8,3,4,5,3,4,6,3,4,7,3,4,8,3,5,6,3,5,7,
+        3,5,8,3,6,7,3,6,8,3,7,8,4,5,6,4,5,7,4,5,8,
+        4,6,7,4,6,8,4,7,8,5,6,7,5,6,8,5,7,8,6,7,8
+    ), dtype = np.int32).reshape(84, 3)
+
+    num_kernels = len(indices)
+    num_dilations = len(dilations)
+
+    num_features = num_kernels * np.sum(num_features_per_dilation)
+
+    biases = np.zeros(num_features, dtype = np.float32)
+
+    feature_index_start = 0
+
+    combination_index = 0
+    num_channels_start = 0
+
+    for dilation_index in range(num_dilations):
+
+        dilation = dilations[dilation_index]
+        padding = ((9 - 1) * dilation) // 2
+
+        num_features_this_dilation = num_features_per_dilation[dilation_index]
+
+        for kernel_index in range(num_kernels):
+
+            feature_index_end = feature_index_start + num_features_this_dilation
+
+            num_channels_this_combination = num_channels_per_combination[combination_index]
+
+            num_channels_end = num_channels_start + num_channels_this_combination
+
+            channels_this_combination = channel_indices[num_channels_start:num_channels_end]
+
+            _X = X[np.random.randint(num_examples)][channels_this_combination]
+
+            A = -_X          # A = alpha * X = -X
+            G = _X + _X + _X # G = gamma * X = 3X
+
+            C_alpha = np.zeros((num_channels_this_combination, input_length), dtype = np.float32)
+            C_alpha[:] = A
+
+            C_gamma = np.zeros((9, num_channels_this_combination, input_length), dtype = np.float32)
+            C_gamma[9 // 2] = G
+
+            start = dilation
+            end = input_length - padding
+
+            for gamma_index in range(9 // 2):
+
+                C_alpha[:, -end:] = C_alpha[:, -end:] + A[:, :end]
+                C_gamma[gamma_index, :, -end:] = G[:, :end]
+
+                end += dilation
+
+            for gamma_index in range(9 // 2 + 1, 9):
+
+                C_alpha[:, :-start] = C_alpha[:, :-start] + A[:, start:]
+                C_gamma[gamma_index, :, :-start] = G[:, start:]
+
+                start += dilation
+
+            index_0, index_1, index_2 = indices[kernel_index]
+
+            C = C_alpha + C_gamma[index_0] + C_gamma[index_1] + C_gamma[index_2]
+            C = np.sum(C, axis = 0)
+
+            biases[feature_index_start:feature_index_end] = np.quantile(C, quantiles[feature_index_start:feature_index_end])
+
+            feature_index_start = feature_index_end
+
+            combination_index += 1
+            num_channels_start = num_channels_end
+
+    return biases
+
+def _fit_dilations(input_length, num_features, max_dilations_per_kernel):
+
+    num_kernels = 84
+
+    num_features_per_kernel = num_features // num_kernels
+    true_max_dilations_per_kernel = min(num_features_per_kernel, max_dilations_per_kernel)
+    multiplier = num_features_per_kernel / true_max_dilations_per_kernel
+
+    max_exponent = np.log2((input_length - 1) / (9 - 1))
+    dilations, num_features_per_dilation = \
+    np.unique(np.logspace(0, max_exponent, true_max_dilations_per_kernel, base = 2).astype(np.int32), return_counts = True)
+    num_features_per_dilation = (num_features_per_dilation * multiplier).astype(np.int32) # this is a vector
+
+    remainder = num_features_per_kernel - np.sum(num_features_per_dilation)
+    i = 0
+    while remainder > 0:
+        num_features_per_dilation[i] += 1
+        remainder -= 1
+        i = (i + 1) % len(num_features_per_dilation)
+
+    return dilations, num_features_per_dilation
+
+# low-discrepancy sequence to assign quantiles to kernel/dilation combinations
+def _quantiles(n):
+    return np.array([(_ * ((np.sqrt(5) + 1) / 2)) % 1 for _ in range(1, n + 1)], dtype = np.float32)
+
+def fit(X, num_features = 10_000, max_dilations_per_kernel = 32):
+
+    _, num_channels, input_length = X.shape
+
+    num_kernels = 84
+
+    dilations, num_features_per_dilation = _fit_dilations(input_length, num_features, max_dilations_per_kernel)
+
+    num_features_per_kernel = np.sum(num_features_per_dilation)
+
+    quantiles = _quantiles(num_kernels * num_features_per_kernel)
+
+    num_dilations = len(dilations)
+    num_combinations = num_kernels * num_dilations
+
+    max_num_channels = min(num_channels, 9)
+    max_exponent = np.log2(max_num_channels + 1)
+
+    num_channels_per_combination = (2 ** np.random.uniform(0, max_exponent, num_combinations)).astype(np.int32)
+
+    channel_indices = np.zeros(num_channels_per_combination.sum(), dtype = np.int32)
+
+    num_channels_start = 0
+    for combination_index in range(num_combinations):
+        num_channels_this_combination = num_channels_per_combination[combination_index]
+        num_channels_end = num_channels_start + num_channels_this_combination
+        channel_indices[num_channels_start:num_channels_end] = np.random.choice(num_channels, num_channels_this_combination, replace = False)
+
+        num_channels_start = num_channels_end
+
+    biases = _fit_biases(X, num_channels_per_combination, channel_indices, dilations, num_features_per_dilation, quantiles)
+
+    return num_channels_per_combination, channel_indices, dilations, num_features_per_dilation, biases
+
+# _PPV(C, b).mean() returns PPV for vector C (convolution output) and scalar b (bias)
+@vectorize("float32(float32,float32)", nopython = True, cache = True)
+def _PPV(a, b):
+    if a > b:
+        return 1
+    else:
+        return 0
+
+@njit("float32[:,:](float32[:,:,:],Tuple((int32[:],int32[:],int32[:],int32[:],float32[:])))", fastmath = True, parallel = True, cache = True)
+def transform(X, parameters):
+
+    num_examples, num_channels, input_length = X.shape
+
+    num_channels_per_combination, channel_indices, dilations, num_features_per_dilation, biases = parameters
+
+    # equivalent to:
+    # >>> from itertools import combinations
+    # >>> indices = np.array([_ for _ in combinations(np.arange(9), 3)], dtype = np.int32)
+    indices = np.array((
+        0,1,2,0,1,3,0,1,4,0,1,5,0,1,6,0,1,7,0,1,8,
+        0,2,3,0,2,4,0,2,5,0,2,6,0,2,7,0,2,8,0,3,4,
+        0,3,5,0,3,6,0,3,7,0,3,8,0,4,5,0,4,6,0,4,7,
+        0,4,8,0,5,6,0,5,7,0,5,8,0,6,7,0,6,8,0,7,8,
+        1,2,3,1,2,4,1,2,5,1,2,6,1,2,7,1,2,8,1,3,4,
+        1,3,5,1,3,6,1,3,7,1,3,8,1,4,5,1,4,6,1,4,7,
+        1,4,8,1,5,6,1,5,7,1,5,8,1,6,7,1,6,8,1,7,8,
+        2,3,4,2,3,5,2,3,6,2,3,7,2,3,8,2,4,5,2,4,6,
+        2,4,7,2,4,8,2,5,6,2,5,7,2,5,8,2,6,7,2,6,8,
+        2,7,8,3,4,5,3,4,6,3,4,7,3,4,8,3,5,6,3,5,7,
+        3,5,8,3,6,7,3,6,8,3,7,8,4,5,6,4,5,7,4,5,8,
+        4,6,7,4,6,8,4,7,8,5,6,7,5,6,8,5,7,8,6,7,8
+    ), dtype = np.int32).reshape(84, 3)
+
+    num_kernels = len(indices)
+    num_dilations = len(dilations)
+
+    num_features = num_kernels * np.sum(num_features_per_dilation)
+
+    features = np.zeros((num_examples, num_features), dtype = np.float32)
+
+    for example_index in prange(num_examples):
+
+        _X = X[example_index]
+
+        A = -_X          # A = alpha * X = -X
+        G = _X + _X + _X # G = gamma * X = 3X
+
+        feature_index_start = 0
+
+        combination_index = 0
+        num_channels_start = 0
+
+        for dilation_index in range(num_dilations):
+
+            _padding0 = dilation_index % 2
+
+            dilation = dilations[dilation_index]
+            padding = ((9 - 1) * dilation) // 2
+
+            num_features_this_dilation = num_features_per_dilation[dilation_index]
+
+            C_alpha = np.zeros((num_channels, input_length), dtype = np.float32)
+            C_alpha[:] = A
+
+            C_gamma = np.zeros((9, num_channels, input_length), dtype = np.float32)
+            C_gamma[9 // 2] = G
+
+            start = dilation
+            end = input_length - padding
+
+            for gamma_index in range(9 // 2):
+
+                C_alpha[:, -end:] = C_alpha[:, -end:] + A[:, :end]
+                C_gamma[gamma_index, :, -end:] = G[:, :end]
+
+                end += dilation
+
+            for gamma_index in range(9 // 2 + 1, 9):
+
+                C_alpha[:, :-start] = C_alpha[:, :-start] + A[:, start:]
+                C_gamma[gamma_index, :, :-start] = G[:, start:]
+
+                start += dilation
+
+            for kernel_index in range(num_kernels):
+
+                feature_index_end = feature_index_start + num_features_this_dilation
+
+                num_channels_this_combination = num_channels_per_combination[combination_index]
+
+                num_channels_end = num_channels_start + num_channels_this_combination
+
+                channels_this_combination = channel_indices[num_channels_start:num_channels_end]
+
+                _padding1 = (_padding0 + kernel_index) % 2
+
+                index_0, index_1, index_2 = indices[kernel_index]
+
+                C = C_alpha[channels_this_combination] + \
+                    C_gamma[index_0][channels_this_combination] + \
+                    C_gamma[index_1][channels_this_combination] + \
+                    C_gamma[index_2][channels_this_combination]
+                C = np.sum(C, axis = 0)
+
+                if _padding1 == 0:
+                    for feature_count in range(num_features_this_dilation):
+                        features[example_index, feature_index_start + feature_count] = _PPV(C, biases[feature_index_start + feature_count]).mean()
+                else:
+                    for feature_count in range(num_features_this_dilation):
+                        features[example_index, feature_index_start + feature_count] = _PPV(C[padding:-padding], biases[feature_index_start + feature_count]).mean()
+
+                feature_index_start = feature_index_end
+
+                combination_index += 1
+                num_channels_start = num_channels_end
+
+    return features

time_series_classification/minirocket/src/minirocket_multivariate_variable.py

@@ -0,0 +1,312 @@
+# Angus Dempster, Daniel F Schmidt, Geoffrey I Webb
+
+# MiniRocket: A Very Fast (Almost) Deterministic Transform for Time Series
+# Classification
+
+# https://arxiv.org/abs/2012.08791
+
+# ** This is an experimental extension of MiniRocket to variable-length,
+#    multivariate input.  It is untested, may contain errors, and may be
+#    inefficient in terms of both storage and computation. **
+
+from numba import njit, prange, vectorize
+import numpy as np
+
+@njit("float32[:](float32[:,:],int32[:],int32[:],int32[:],int32[:],int32[:],float32[:])", fastmath = True, parallel = False, cache = True)
+def _fit_biases(X, L, num_channels_per_combination, channel_indices, dilations, num_features_per_dilation, quantiles):
+
+    num_examples = len(L)
+
+    num_channels, _ = X.shape
+
+    # equivalent to:
+    # >>> from itertools import combinations
+    # >>> indices = np.array([_ for _ in combinations(np.arange(9), 3)], dtype = np.int32)
+    indices = np.array((
+        0,1,2,0,1,3,0,1,4,0,1,5,0,1,6,0,1,7,0,1,8,
+        0,2,3,0,2,4,0,2,5,0,2,6,0,2,7,0,2,8,0,3,4,
+        0,3,5,0,3,6,0,3,7,0,3,8,0,4,5,0,4,6,0,4,7,
+        0,4,8,0,5,6,0,5,7,0,5,8,0,6,7,0,6,8,0,7,8,
+        1,2,3,1,2,4,1,2,5,1,2,6,1,2,7,1,2,8,1,3,4,
+        1,3,5,1,3,6,1,3,7,1,3,8,1,4,5,1,4,6,1,4,7,
+        1,4,8,1,5,6,1,5,7,1,5,8,1,6,7,1,6,8,1,7,8,
+        2,3,4,2,3,5,2,3,6,2,3,7,2,3,8,2,4,5,2,4,6,
+        2,4,7,2,4,8,2,5,6,2,5,7,2,5,8,2,6,7,2,6,8,
+        2,7,8,3,4,5,3,4,6,3,4,7,3,4,8,3,5,6,3,5,7,
+        3,5,8,3,6,7,3,6,8,3,7,8,4,5,6,4,5,7,4,5,8,
+        4,6,7,4,6,8,4,7,8,5,6,7,5,6,8,5,7,8,6,7,8
+    ), dtype = np.int32).reshape(84, 3)
+
+    num_kernels = len(indices)
+    num_dilations = len(dilations)
+
+    num_features = num_kernels * np.sum(num_features_per_dilation)
+
+    biases = np.zeros(num_features, dtype = np.float32)
+
+    feature_index_start = 0
+
+    combination_index = 0
+    num_channels_start = 0
+
+    for dilation_index in range(num_dilations):
+
+        dilation = dilations[dilation_index]
+        padding = ((9 - 1) * dilation) // 2
+
+        num_features_this_dilation = num_features_per_dilation[dilation_index]
+
+        for kernel_index in range(num_kernels):
+
+            feature_index_end = feature_index_start + num_features_this_dilation
+
+            num_channels_this_combination = num_channels_per_combination[combination_index]
+
+            num_channels_end = num_channels_start + num_channels_this_combination
+
+            channels_this_combination = channel_indices[num_channels_start:num_channels_end]
+
+            example_index = np.random.randint(num_examples)
+
+            input_length = np.int64(L[example_index])
+
+            b = np.sum(L[0:example_index + 1])
+            a = b - input_length
+
+            _X = X[channels_this_combination, a:b]
+
+            A = -_X          # A = alpha * X = -X
+            G = _X + _X + _X # G = gamma * X = 3X
+
+            C_alpha = np.zeros((num_channels_this_combination, input_length), dtype = np.float32)
+            C_alpha[:] = A
+
+            C_gamma = np.zeros((9, num_channels_this_combination, input_length), dtype = np.float32)
+            C_gamma[9 // 2] = G
+
+            start = dilation
+            end = input_length - padding
+
+            for gamma_index in range(9 // 2):
+
+                # thanks to Murtaza Jafferji @murtazajafferji for suggesting this fix
+                if end > 0:
+
+                    C_alpha[:, -end:] = C_alpha[:, -end:] + A[:, :end]
+                    C_gamma[gamma_index, :, -end:] = G[:, :end]
+
+                end += dilation
+
+            for gamma_index in range(9 // 2 + 1, 9):
+
+                if start < input_length:
+
+                    C_alpha[:, :-start] = C_alpha[:, :-start] + A[:, start:]
+                    C_gamma[gamma_index, :, :-start] = G[:, start:]
+
+                start += dilation
+
+            index_0, index_1, index_2 = indices[kernel_index]
+
+            C = C_alpha + C_gamma[index_0] + C_gamma[index_1] + C_gamma[index_2]
+            C = np.sum(C, axis = 0)
+
+            biases[feature_index_start:feature_index_end] = np.quantile(C, quantiles[feature_index_start:feature_index_end])
+
+            feature_index_start = feature_index_end
+
+            combination_index += 1
+            num_channels_start = num_channels_end
+
+    return biases
+
+def _fit_dilations(reference_length, num_features, max_dilations_per_kernel):
+
+    num_kernels = 84
+
+    num_features_per_kernel = num_features // num_kernels
+    true_max_dilations_per_kernel = min(num_features_per_kernel, max_dilations_per_kernel)
+    multiplier = num_features_per_kernel / true_max_dilations_per_kernel
+
+    max_exponent = np.log2((reference_length - 1) / (9 - 1))
+    dilations, num_features_per_dilation = \
+    np.unique(np.logspace(0, max_exponent, true_max_dilations_per_kernel, base = 2).astype(np.int32), return_counts = True)
+    num_features_per_dilation = (num_features_per_dilation * multiplier).astype(np.int32) # this is a vector
+
+    remainder = num_features_per_kernel - np.sum(num_features_per_dilation)
+    i = 0
+    while remainder > 0:
+        num_features_per_dilation[i] += 1
+        remainder -= 1
+        i = (i + 1) % len(num_features_per_dilation)
+
+    return dilations, num_features_per_dilation
+
+# low-discrepancy sequence to assign quantiles to kernel/dilation combinations
+def _quantiles(n):
+    return np.array([(_ * ((np.sqrt(5) + 1) / 2)) % 1 for _ in range(1, n + 1)], dtype = np.float32)
+
+def fit(X, L, reference_length = None, num_features = 10_000, max_dilations_per_kernel = 32):
+
+    # note in relation to dilation:
+    # * change *reference_length* according to what is appropriate for your
+    #   application, e.g., L.max(), L.mean(), np.median(L)
+    # * use fit(...) with an appropriate subset of time series, e.g., for
+    #   reference_length = L.mean(), call fit(...) using only time series of at
+    #   least length L.mean() [see filter_by_length(...)]
+    if reference_length == None:
+        reference_length = L.max()
+
+    num_channels, _ = X.shape
+
+    num_kernels = 84
+
+    dilations, num_features_per_dilation = _fit_dilations(reference_length, num_features, max_dilations_per_kernel)
+
+    num_features_per_kernel = np.sum(num_features_per_dilation)
+
+    quantiles = _quantiles(num_kernels * num_features_per_kernel)
+
+    num_dilations = len(dilations)
+    num_combinations = num_kernels * num_dilations
+
+    max_num_channels = min(num_channels, 9)
+    max_exponent = np.log2(max_num_channels + 1)
+
+    num_channels_per_combination = (2 ** np.random.uniform(0, max_exponent, num_combinations)).astype(np.int32)
+
+    channel_indices = np.zeros(num_channels_per_combination.sum(), dtype = np.int32)
+
+    num_channels_start = 0
+    for combination_index in range(num_combinations):
+        num_channels_this_combination = num_channels_per_combination[combination_index]
+        num_channels_end = num_channels_start + num_channels_this_combination
+        channel_indices[num_channels_start:num_channels_end] = np.random.choice(num_channels, num_channels_this_combination, replace = False)
+
+        num_channels_start = num_channels_end
+
+    biases = _fit_biases(X, L, num_channels_per_combination, channel_indices, dilations, num_features_per_dilation, quantiles)
+
+    return num_channels_per_combination, channel_indices, dilations, num_features_per_dilation, biases
+
+# _PPV(C, b).mean() returns PPV for vector C (convolution output) and scalar b (bias)
+@vectorize("float32(float32,float32)", nopython = True, cache = True)
+def _PPV(a, b):
+    if a > b:
+        return 1
+    else:
+        return 0
+
+@njit("float32[:,:](float32[:,:],int32[:],Tuple((int32[:],int32[:],int32[:],int32[:],float32[:])))", fastmath = True, parallel = True, cache = True)
+def transform(X, L, parameters):
+
+    num_examples = len(L)
+
+    num_channels, _ = X.shape
+
+    num_channels_per_combination, channel_indices, dilations, num_features_per_dilation, biases = parameters
+
+    # equivalent to:
+    # >>> from itertools import combinations
+    # >>> indices = np.array([_ for _ in combinations(np.arange(9), 3)], dtype = np.int32)
+    indices = np.array((
+        0,1,2,0,1,3,0,1,4,0,1,5,0,1,6,0,1,7,0,1,8,
+        0,2,3,0,2,4,0,2,5,0,2,6,0,2,7,0,2,8,0,3,4,
+        0,3,5,0,3,6,0,3,7,0,3,8,0,4,5,0,4,6,0,4,7,
+        0,4,8,0,5,6,0,5,7,0,5,8,0,6,7,0,6,8,0,7,8,
+        1,2,3,1,2,4,1,2,5,1,2,6,1,2,7,1,2,8,1,3,4,
+        1,3,5,1,3,6,1,3,7,1,3,8,1,4,5,1,4,6,1,4,7,
+        1,4,8,1,5,6,1,5,7,1,5,8,1,6,7,1,6,8,1,7,8,
+        2,3,4,2,3,5,2,3,6,2,3,7,2,3,8,2,4,5,2,4,6,
+        2,4,7,2,4,8,2,5,6,2,5,7,2,5,8,2,6,7,2,6,8,
+        2,7,8,3,4,5,3,4,6,3,4,7,3,4,8,3,5,6,3,5,7,
+        3,5,8,3,6,7,3,6,8,3,7,8,4,5,6,4,5,7,4,5,8,
+        4,6,7,4,6,8,4,7,8,5,6,7,5,6,8,5,7,8,6,7,8
+    ), dtype = np.int32).reshape(84, 3)
+
+    num_kernels = len(indices)
+    num_dilations = len(dilations)
+
+    num_features = num_kernels * np.sum(num_features_per_dilation)
+
+    features = np.zeros((num_examples, num_features), dtype = np.float32)
+
+    for example_index in prange(num_examples):
+
+        input_length = np.int64(L[example_index])
+
+        b = np.sum(L[0:example_index + 1])
+        a = b - input_length
+
+        _X = X[:, a:b]
+
+        A = -_X          # A = alpha * X = -X
+        G = _X + _X + _X # G = gamma * X = 3X
+
+        feature_index_start = 0
+
+        combination_index = 0
+        num_channels_start = 0
+
+        for dilation_index in range(num_dilations):
+
+            dilation = dilations[dilation_index]
+            padding = ((9 - 1) * dilation) // 2
+
+            num_features_this_dilation = num_features_per_dilation[dilation_index]
+
+            C_alpha = np.zeros((num_channels, input_length), dtype = np.float32)
+            C_alpha[:] = A
+
+            C_gamma = np.zeros((9, num_channels, input_length), dtype = np.float32)
+            C_gamma[9 // 2] = G
+
+            start = dilation
+            end = input_length - padding
+
+            for gamma_index in range(9 // 2):
+
+                # thanks to Murtaza Jafferji @murtazajafferji for suggesting this fix
+                if end > 0:
+
+                    C_alpha[:, -end:] = C_alpha[:, -end:] + A[:, :end]
+                    C_gamma[gamma_index, :, -end:] = G[:, :end]
+
+                end += dilation
+
+            for gamma_index in range(9 // 2 + 1, 9):
+
+                if start < input_length:
+
+                    C_alpha[:, :-start] = C_alpha[:, :-start] + A[:, start:]
+                    C_gamma[gamma_index, :, :-start] = G[:, start:]
+
+                start += dilation
+
+            for kernel_index in range(num_kernels):
+
+                feature_index_end = feature_index_start + num_features_this_dilation
+
+                num_channels_this_combination = num_channels_per_combination[combination_index]
+
+                num_channels_end = num_channels_start + num_channels_this_combination
+
+                channels_this_combination = channel_indices[num_channels_start:num_channels_end]
+
+                index_0, index_1, index_2 = indices[kernel_index]
+
+                C = C_alpha[channels_this_combination] + \
+                    C_gamma[index_0][channels_this_combination] + \
+                    C_gamma[index_1][channels_this_combination] + \
+                    C_gamma[index_2][channels_this_combination]
+                C = np.sum(C, axis = 0)
+
+                for feature_count in range(num_features_this_dilation):
+                    features[example_index, feature_index_start + feature_count] = _PPV(C, biases[feature_index_start + feature_count]).mean()
+
+                feature_index_start = feature_index_end
+
+                combination_index += 1
+                num_channels_start = num_channels_end
+
+    return features

time_series_classification/minirocket/src/minirocket_variable.py

@@ -0,0 +1,296 @@
+# Angus Dempster, Daniel F Schmidt, Geoffrey I Webb
+
+# MiniRocket: A Very Fast (Almost) Deterministic Transform for Time Series
+# Classification
+
+# https://arxiv.org/abs/2012.08791
+
+# ** This is an experimental extension of MiniRocket to variable-length input.
+#    It is untested, may contain errors, and may be inefficient in terms of both
+#    storage and computation. **
+
+from numba import njit, prange, vectorize
+import numpy as np
+
+@njit("float32[:](float32[:],int32[:],int32[:],int32[:],float32[:])", fastmath = True, parallel = False, cache = True)
+def _fit_biases(X, L, dilations, num_features_per_dilation, quantiles):
+
+    num_examples = len(L)
+
+    # equivalent to:
+    # >>> from itertools import combinations
+    # >>> indices = np.array([_ for _ in combinations(np.arange(9), 3)], dtype = np.int32)
+    indices = np.array((
+        0,1,2,0,1,3,0,1,4,0,1,5,0,1,6,0,1,7,0,1,8,
+        0,2,3,0,2,4,0,2,5,0,2,6,0,2,7,0,2,8,0,3,4,
+        0,3,5,0,3,6,0,3,7,0,3,8,0,4,5,0,4,6,0,4,7,
+        0,4,8,0,5,6,0,5,7,0,5,8,0,6,7,0,6,8,0,7,8,
+        1,2,3,1,2,4,1,2,5,1,2,6,1,2,7,1,2,8,1,3,4,
+        1,3,5,1,3,6,1,3,7,1,3,8,1,4,5,1,4,6,1,4,7,
+        1,4,8,1,5,6,1,5,7,1,5,8,1,6,7,1,6,8,1,7,8,
+        2,3,4,2,3,5,2,3,6,2,3,7,2,3,8,2,4,5,2,4,6,
+        2,4,7,2,4,8,2,5,6,2,5,7,2,5,8,2,6,7,2,6,8,
+        2,7,8,3,4,5,3,4,6,3,4,7,3,4,8,3,5,6,3,5,7,
+        3,5,8,3,6,7,3,6,8,3,7,8,4,5,6,4,5,7,4,5,8,
+        4,6,7,4,6,8,4,7,8,5,6,7,5,6,8,5,7,8,6,7,8
+    ), dtype = np.int32).reshape(84, 3)
+
+    num_kernels = len(indices)
+    num_dilations = len(dilations)
+
+    num_features = num_kernels * np.sum(num_features_per_dilation)
+
+    biases = np.zeros(num_features, dtype = np.float32)
+
+    feature_index_start = 0
+
+    for dilation_index in range(num_dilations):
+
+        dilation = dilations[dilation_index]
+        padding = ((9 - 1) * dilation) // 2
+
+        num_features_this_dilation = num_features_per_dilation[dilation_index]
+
+        for kernel_index in range(num_kernels):
+
+            feature_index_end = feature_index_start + num_features_this_dilation
+
+            example_index = np.random.randint(num_examples)
+
+            input_length = np.int64(L[example_index])
+
+            b = np.sum(L[0:example_index + 1])
+            a = b - input_length
+
+            _X = X[a:b]
+
+            A = -_X          # A = alpha * X = -X
+            G = _X + _X + _X # G = gamma * X = 3X
+
+            C_alpha = np.zeros(input_length, dtype = np.float32)
+            C_alpha[:] = A
+
+            C_gamma = np.zeros((9, input_length), dtype = np.float32)
+            C_gamma[9 // 2] = G
+
+            start = dilation
+            end = input_length - padding
+
+            for gamma_index in range(9 // 2):
+
+                # thanks to Murtaza Jafferji @murtazajafferji for suggesting this fix
+                if end > 0:
+
+                    C_alpha[-end:] = C_alpha[-end:] + A[:end]
+                    C_gamma[gamma_index, -end:] = G[:end]
+
+                end += dilation
+
+            for gamma_index in range(9 // 2 + 1, 9):
+
+                if start < input_length:
+
+                    C_alpha[:-start] = C_alpha[:-start] + A[start:]
+                    C_gamma[gamma_index, :-start] = G[start:]
+
+                start += dilation
+
+            index_0, index_1, index_2 = indices[kernel_index]
+
+            C = C_alpha + C_gamma[index_0] + C_gamma[index_1] + C_gamma[index_2]
+
+            biases[feature_index_start:feature_index_end] = np.quantile(C, quantiles[feature_index_start:feature_index_end])
+
+            feature_index_start = feature_index_end
+
+    return biases
+
+def _fit_dilations(reference_length, num_features, max_dilations_per_kernel):
+
+    num_kernels = 84
+
+    num_features_per_kernel = num_features // num_kernels
+    true_max_dilations_per_kernel = min(num_features_per_kernel, max_dilations_per_kernel)
+    multiplier = num_features_per_kernel / true_max_dilations_per_kernel
+
+    max_exponent = np.log2((reference_length - 1) / (9 - 1))
+    dilations, num_features_per_dilation = \
+    np.unique(np.logspace(0, max_exponent, true_max_dilations_per_kernel, base = 2).astype(np.int32), return_counts = True)
+    num_features_per_dilation = (num_features_per_dilation * multiplier).astype(np.int32) # this is a vector
+
+    remainder = num_features_per_kernel - np.sum(num_features_per_dilation)
+    i = 0
+    while remainder > 0:
+        num_features_per_dilation[i] += 1
+        remainder -= 1
+        i = (i + 1) % len(num_features_per_dilation)
+
+    return dilations, num_features_per_dilation
+
+# low-discrepancy sequence to assign quantiles to kernel/dilation combinations
+def _quantiles(n):
+    return np.array([(_ * ((np.sqrt(5) + 1) / 2)) % 1 for _ in range(1, n + 1)], dtype = np.float32)
+
+def fit(X, L, reference_length = None, num_features = 10_000, max_dilations_per_kernel = 32):
+
+    # note in relation to dilation:
+    # * change *reference_length* according to what is appropriate for your
+    #   application, e.g., L.max(), L.mean(), np.median(L)
+    # * use fit(...) with an appropriate subset of time series, e.g., for
+    #   reference_length = L.mean(), call fit(...) using only time series of at
+    #   least length L.mean() [see filter_by_length(...)]
+    if reference_length == None:
+        reference_length = L.max()
+
+    num_kernels = 84
+
+    dilations, num_features_per_dilation = _fit_dilations(reference_length, num_features, max_dilations_per_kernel)
+
+    num_features_per_kernel = np.sum(num_features_per_dilation)
+
+    quantiles = _quantiles(num_kernels * num_features_per_kernel)
+
+    biases = _fit_biases(X, L, dilations, num_features_per_dilation, quantiles)
+
+    return dilations, num_features_per_dilation, biases
+
+# _PPV(C, b).mean() returns PPV for vector C (convolution output) and scalar b (bias)
+@vectorize("float32(float32,float32)", nopython = True, cache = True)
+def _PPV(a, b):
+    if a > b:
+        return 1
+    else:
+        return 0
+
+@njit("float32[:,:](float32[:],int32[:],Tuple((int32[:],int32[:],float32[:])))", fastmath = True, parallel = True, cache = True)
+def transform(X, L, parameters):
+
+    num_examples = len(L)
+
+    dilations, num_features_per_dilation, biases = parameters
+
+    # equivalent to:
+    # >>> from itertools import combinations
+    # >>> indices = np.array([_ for _ in combinations(np.arange(9), 3)], dtype = np.int32)
+    indices = np.array((
+        0,1,2,0,1,3,0,1,4,0,1,5,0,1,6,0,1,7,0,1,8,
+        0,2,3,0,2,4,0,2,5,0,2,6,0,2,7,0,2,8,0,3,4,
+        0,3,5,0,3,6,0,3,7,0,3,8,0,4,5,0,4,6,0,4,7,
+        0,4,8,0,5,6,0,5,7,0,5,8,0,6,7,0,6,8,0,7,8,
+        1,2,3,1,2,4,1,2,5,1,2,6,1,2,7,1,2,8,1,3,4,
+        1,3,5,1,3,6,1,3,7,1,3,8,1,4,5,1,4,6,1,4,7,
+        1,4,8,1,5,6,1,5,7,1,5,8,1,6,7,1,6,8,1,7,8,
+        2,3,4,2,3,5,2,3,6,2,3,7,2,3,8,2,4,5,2,4,6,
+        2,4,7,2,4,8,2,5,6,2,5,7,2,5,8,2,6,7,2,6,8,
+        2,7,8,3,4,5,3,4,6,3,4,7,3,4,8,3,5,6,3,5,7,
+        3,5,8,3,6,7,3,6,8,3,7,8,4,5,6,4,5,7,4,5,8,
+        4,6,7,4,6,8,4,7,8,5,6,7,5,6,8,5,7,8,6,7,8
+    ), dtype = np.int32).reshape(84, 3)
+
+    num_kernels = len(indices)
+    num_dilations = len(dilations)
+
+    num_features = num_kernels * np.sum(num_features_per_dilation)
+
+    features = np.zeros((num_examples, num_features), dtype = np.float32)
+
+    for example_index in prange(num_examples):
+
+        input_length = np.int64(L[example_index])
+
+        b = np.sum(L[0:example_index + 1])
+        a = b - input_length
+
+        _X = X[a:b]
+
+        A = -_X          # A = alpha * X = -X
+        G = _X + _X + _X # G = gamma * X = 3X
+
+        feature_index_start = 0
+
+        for dilation_index in range(num_dilations):
+
+            _padding0 = dilation_index % 2
+
+            dilation = dilations[dilation_index]
+            padding = ((9 - 1) * dilation) // 2
+
+            num_features_this_dilation = num_features_per_dilation[dilation_index]
+
+            C_alpha = np.zeros(input_length, dtype = np.float32)
+            C_alpha[:] = A
+
+            C_gamma = np.zeros((9, input_length), dtype = np.float32)
+            C_gamma[9 // 2] = G
+
+            start = dilation
+            end = input_length - padding
+
+            for gamma_index in range(9 // 2):
+
+                # thanks to Murtaza Jafferji @murtazajafferji for suggesting this fix
+                if end > 0:
+
+                    C_alpha[-end:] = C_alpha[-end:] + A[:end]
+                    C_gamma[gamma_index, -end:] = G[:end]
+
+                end += dilation
+
+            for gamma_index in range(9 // 2 + 1, 9):
+
+                if start < input_length:
+
+                    C_alpha[:-start] = C_alpha[:-start] + A[start:]
+                    C_gamma[gamma_index, :-start] = G[start:]
+
+                start += dilation
+
+            for kernel_index in range(num_kernels):
+
+                feature_index_end = feature_index_start + num_features_this_dilation
+
+                # force padding
+                # alternatively, pass output through np.nan_to_num(...)
+                # _padding1 = (_padding0 + kernel_index) % 2
+                _padding1 = 0
+
+                index_0, index_1, index_2 = indices[kernel_index]
+
+                C = C_alpha + C_gamma[index_0] + C_gamma[index_1] + C_gamma[index_2]
+
+                if _padding1 == 0:
+                    for feature_count in range(num_features_this_dilation):
+                        features[example_index, feature_index_start + feature_count] = _PPV(C, biases[feature_index_start + feature_count]).mean()
+                else:
+                    for feature_count in range(num_features_this_dilation):
+                        features[example_index, feature_index_start + feature_count] = _PPV(C[padding:-padding], biases[feature_index_start + feature_count]).mean()
+
+                feature_index_start = feature_index_end
+
+    return features
+
+# return only time series of at least *min_length*
+def filter_by_length(X, L, min_length = 0):
+
+    _L = L[L >= min_length]
+    _X = np.zeros(_L.sum(), dtype = np.float32)
+
+    count = 0
+
+    for example_index in range(len(L)):
+
+        if L[example_index] >= min_length:
+
+            # indices for L
+            b = L[0:example_index + 1].sum()
+            a = b - L[example_index]
+
+            # indices for _L
+            _b = _L[0:count + 1].sum()
+            _a = _b - _L[count]
+
+            _X[_a:_b] = X[a:b]
+
+            count += 1
+
+    return _X, _L

time_series_classification/minirocket/src/softmax.py

@@ -0,0 +1,241 @@
+# Angus Dempster, Daniel F Schmidt, Geoffrey I Webb
+
+# MiniRocket: A Very Fast (Almost) Deterministic Transform for Time Series
+# Classification
+
+# https://arxiv.org/abs/2012.08791
+
+import copy
+import numpy as np
+import pandas as pd
+import torch, torch.nn as nn, torch.optim as optim
+
+from minirocket import fit, transform
+
+def train(path, num_classes, training_size, **kwargs):
+
+    # -- init ------------------------------------------------------------------
+
+    # default hyperparameters are reusable for any dataset
+    args = \
+    {
+        "num_features"    : 10_000,
+        "validation_size" : 2 ** 11,
+        "chunk_size"      : 2 ** 12,
+        "minibatch_size"  : 256,
+        "lr"              : 1e-4,
+        "max_epochs"      : 50,
+        "patience_lr"     : 5,  #  50 minibatches
+        "patience"        : 10, # 100 minibatches
+        "cache_size"      : training_size # set to 0 to prevent caching
+    }
+    args = {**args, **kwargs}
+
+    _num_features = 84 * (args["num_features"] // 84)
+    num_chunks = np.int32(np.ceil(training_size / args["chunk_size"]))
+
+    def init(layer):
+        if isinstance(layer, nn.Linear):
+            nn.init.constant_(layer.weight.data, 0)
+            nn.init.constant_(layer.bias.data, 0)
+
+    # -- cache -----------------------------------------------------------------
+
+    # cache as much as possible to avoid unecessarily repeating the transform
+    # consider caching to disk if appropriate, along the lines of numpy.memmap
+
+    cache_X = torch.zeros((args["cache_size"], _num_features))
+    cache_Y = torch.zeros(args["cache_size"], dtype = torch.long)
+    cache_count = 0
+    fully_cached = False
+
+    # -- model -----------------------------------------------------------------
+
+    model = nn.Sequential(nn.Linear(_num_features, num_classes))
+    loss_function = nn.CrossEntropyLoss()
+    optimizer = optim.Adam(model.parameters(), lr = args["lr"])
+    scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, factor = 0.5, min_lr = 1e-8, patience = args["patience_lr"])
+    model.apply(init)
+
+    # -- validation data -------------------------------------------------------
+
+    # gotcha: copy() is essential to avoid competition for memory access with read_csv(...)
+    validation_data = pd.read_csv(path,
+                                  header = None,
+                                  sep = ",",
+                                  nrows = args["validation_size"],
+                                  engine = "c").values.copy()
+    Y_validation, X_validation = torch.LongTensor(validation_data[:, 0]), validation_data[:, 1:].astype(np.float32)
+
+    # -- run -------------------------------------------------------------------
+
+    minibatch_count = 0
+    best_validation_loss = np.inf
+    stall_count = 0
+    stop = False
+
+    print("Training... (faster once caching is finished)")
+
+    for epoch in range(args["max_epochs"]):
+
+        if epoch > 0 and stop:
+            break
+
+        if not fully_cached:
+            file = pd.read_csv(path,
+                               header = None,
+                               sep = ",",
+                               skiprows = args["validation_size"],
+                               chunksize = args["chunk_size"],
+                               engine = "c")
+
+        for chunk_index in range(num_chunks):
+
+            a = chunk_index * args["chunk_size"]
+            b = min(a + args["chunk_size"], training_size)
+            _b = b - a
+
+            if epoch > 0 and stop:
+                break
+
+            print(f"Epoch {epoch + 1}; Chunk = {chunk_index + 1}...".ljust(80, " "), end = "\r", flush = True)
+
+            # if not fully cached, read next file chunk
+            if not fully_cached:
+
+                # gotcha: copy() is essential to avoid competition for memory access with read_csv(...)
+                training_data = file.get_chunk().values[:_b].copy()
+                Y_training, X_training = torch.LongTensor(training_data[:, 0]), training_data[:, 1:].astype(np.float32)
+
+                if epoch == 0 and chunk_index == 0:
+
+                    parameters = fit(X_training, args["num_features"])
+
+                    # transform validation data
+                    X_validation_transform = transform(X_validation, parameters)
+
+            # if cached, retrieve from cache
+            if b <= cache_count:
+
+                X_training_transform = cache_X[a:b]
+                Y_training = cache_Y[a:b]
+
+            # else, transform and cache
+            else:
+
+                # transform training data
+                X_training_transform = transform(X_training, parameters)
+
+                if epoch == 0 and chunk_index == 0:
+
+                    # per-feature mean and standard deviation
+                    f_mean = X_training_transform.mean(0)
+                    f_std = X_training_transform.std(0) + 1e-8
+
+                    # normalise validation features
+                    X_validation_transform = (X_validation_transform - f_mean) / f_std
+                    X_validation_transform = torch.FloatTensor(X_validation_transform)
+
+                # normalise training features
+                X_training_transform = (X_training_transform - f_mean) / f_std
+                X_training_transform = torch.FloatTensor(X_training_transform)
+
+                # cache as much of the transform as possible
+                if b <= args["cache_size"]:
+                    cache_X[a:b] = X_training_transform
+                    cache_Y[a:b] = Y_training
+                    cache_count = b
+
+                    if cache_count >= training_size:
+                        fully_cached = True
+
+            minibatches = torch.randperm(len(X_training_transform)).split(args["minibatch_size"])
+
+            # train on transformed features
+            for minibatch_index, minibatch in enumerate(minibatches):
+
+                if epoch > 0 and stop:
+                    break
+
+                if minibatch_index > 0 and len(minibatch) < args["minibatch_size"]:
+                    break
+
+                # -- training --------------------------------------------------
+
+                optimizer.zero_grad()
+                _Y_training = model(X_training_transform[minibatch])
+                training_loss = loss_function(_Y_training, Y_training[minibatch])
+                training_loss.backward()
+                optimizer.step()
+
+                minibatch_count += 1
+
+                if minibatch_count % 10 == 0:
+
+                    _Y_validation = model(X_validation_transform)
+                    validation_loss = loss_function(_Y_validation, Y_validation)
+
+                    scheduler.step(validation_loss)
+
+                    if validation_loss.item() >= best_validation_loss:
+                        stall_count += 1
+                        if stall_count >= args["patience"]:
+                            stop = True
+                            print(f"\n<Stopped at Epoch {epoch + 1}>")
+                    else:
+                        best_validation_loss = validation_loss.item()
+                        best_model = copy.deepcopy(model)
+                        if not stop:
+                            stall_count = 0
+
+    return parameters, best_model, f_mean, f_std
+
+def predict(path,
+            parameters,
+            model,
+            f_mean,
+            f_std,
+            **kwargs):
+
+    args = \
+    {
+        "score"      : True,
+        "chunk_size" : 2 ** 12,
+        "test_size"  : None
+    }
+    args = {**args, **kwargs}
+
+    file = pd.read_csv(path,
+                       header = None,
+                       sep = ",",
+                       chunksize = args["chunk_size"],
+                       nrows = args["test_size"],
+                       engine = "c")
+
+    predictions = []
+
+    correct = 0
+    total = 0
+
+    for chunk_index, chunk in enumerate(file):
+
+        print(f"Chunk = {chunk_index + 1}...".ljust(80, " "), end = "\r")
+
+        # gotcha: copy() is essential to avoid competition for memory access with read_csv(...)
+        test_data = chunk.values.copy()
+        Y_test, X_test = test_data[:, 0], test_data[:, 1:].astype(np.float32)
+
+        X_test_transform = transform(X_test, parameters)
+        X_test_transform = (X_test_transform - f_mean) / f_std
+        X_test_transform = torch.FloatTensor(X_test_transform)
+
+        _predictions = model(X_test_transform).argmax(1).numpy()
+        predictions.append(_predictions)
+
+        total += len(test_data)
+        correct += (_predictions == Y_test).sum()
+
+    if args["score"]:
+        return np.concatenate(predictions), correct / total
+    else:
+        return np.concatenate(predictions)

time_series_forecasting/data_provider/data_factory.py

@@ -0,0 +1,120 @@
+# Adapted from dominant-shuffle (https://github.com/zuojie2024/dominant-shuffle) and FrAug implementations
+# Original works: Dominant Shuffle by Kai Zhao et al., FrAug by Muxi Chen et al.
+
+from data_provider.data_loader import Dataset_ETT_hour, Dataset_ETT_minute, Dataset_Custom, Dataset_Pred, Dataset_PEMS
+from torch.utils.data import DataLoader
+import torch
+#from data_provider.uea import collate_fn
+data_dict = {
+    'ETTh1': Dataset_ETT_hour,
+    'ETTh2': Dataset_ETT_hour,
+    'ETTm1': Dataset_ETT_minute,
+    'ETTm2': Dataset_ETT_minute,
+    'custom': Dataset_Custom,
+    'PEMS': Dataset_PEMS,
+}
+
+
+def data_provider(args, flag):
+    Data = data_dict[args.data]
+    timeenc = 0 if args.embed != 'timeF' else 1
+
+    if flag == 'test':
+        shuffle_flag = False
+        drop_last = True
+        batch_size = args.batch_size
+        #if args.task_name == 'anomaly_detection' or args.task_name == 'classification':
+        #    batch_size = args.batch_size
+        #else:
+        #    batch_size = 1  # bsz=1 for evaluation
+        freq = args.freq
+    elif flag == 'pred':
+        shuffle_flag = False
+        drop_last = False
+        batch_size = 1
+        freq = args.freq
+        Data = Dataset_Pred
+    else:
+        shuffle_flag = True
+        drop_last = True
+        batch_size = args.batch_size
+        freq = args.freq
+    
+    
+    if args.task_name == 'classification':
+        drop_last = False
+        data_set = Data(
+            root_path=args.root_path,
+            flag=flag,
+        )
+
+        data_loader = DataLoader(
+            data_set,
+            batch_size=batch_size,
+            shuffle=shuffle_flag,
+            num_workers=args.num_workers, #
+            drop_last=drop_last,
+            collate_fn=lambda x: collate_fn(x, max_len=args.seq_len)
+        )
+        return data_set, data_loader
+    elif args.task_name == 'short_term_forecast':
+        drop_last = False
+        data_set = Data(
+            root_path=args.root_path,
+            data_path=args.data_path,
+            flag=flag,
+            size=[args.seq_len, args.label_len, args.pred_len],
+            features=args.features,
+            target=args.target,
+            timeenc=timeenc,
+            freq=freq,
+            seasonal_patterns=args.seasonal_patterns
+        )
+        print(flag, len(data_set))
+        data_loader = DataLoader(
+            data_set,
+            batch_size=batch_size,
+            shuffle=shuffle_flag,
+            num_workers=args.num_workers,
+            drop_last=drop_last)
+        return data_set, data_loader
+    else:
+        data_set = Data(
+        config=args,
+        root_path=args.root_path,
+        data_path=args.data_path,
+        flag=flag,
+        size=[args.seq_len, args.label_len, args.pred_len],
+        features=args.features,
+        target=args.target,
+        timeenc=timeenc,
+        freq=freq,
+        cycle=args.cycle
+        )
+        print(flag, len(data_set))
+        data_loader = DataLoader(
+        data_set,
+        batch_size=batch_size,
+        shuffle=shuffle_flag,
+        num_workers=args.num_workers,
+        drop_last=drop_last)
+        return data_set, data_loader
+
+def collate_fn(batch, max_len):
+    # Assuming batch is a list of tuples (input, target) and input is a sequence
+    # Pad all sequences to the same length (max_len) with zeros
+    inputs, targets = zip(*batch)
+    
+    # Pad sequences to max_len
+    padded_inputs = torch.nn.utils.rnn.pad_sequence(
+        [torch.tensor(seq) for seq in inputs],
+        batch_first=True, padding_value=0
+    )[:, :max_len]  # Ensure padding to exactly max_len
+    
+    # Similarly, process targets (adjust as necessary)
+    padded_targets = torch.nn.utils.rnn.pad_sequence(
+        [torch.tensor(target) for target in targets],
+        batch_first=True, padding_value=0
+    )[:, :max_len]
+    
+    return padded_inputs, padded_targets

time_series_forecasting/data_provider/data_loader.py

@@ -0,0 +1,659 @@
+# Adapted from dominant-shuffle (https://github.com/zuojie2024/dominant-shuffle) and FrAug implementations
+# Original works: Dominant Shuffle by Kai Zhao et al., FrAug by Muxi Chen et al.
+# Modified and extended by Jafar Bakhshaliyev (2025)
+
+
+
+import os
+import numpy as np
+import pandas as pd
+import os
+import torch
+from torch.utils.data import Dataset
+from sklearn.preprocessing import StandardScaler
+from utils.timefeatures import time_features
+import warnings
+#from data_provider.m4 import M4Dataset, M4Meta
+from utils.augmentations import augmentation
+#from data_provider.uea import subsample, interpolate_missing, Normalizer
+#from sktime.datasets import load_from_tsfile_to_dataframe
+#import glob
+import re
+warnings.filterwarnings('ignore')
+
+class Dataset_ETT_hour(Dataset):
+    def __init__(self, config, root_path, flag='train', size=None,
+                 features='S', data_path='ETTh1.csv',
+                 target='OT', scale=True, timeenc=0, freq='h', cycle = None):
+        # size [seq_len, label_len, pred_len]
+        # info
+        self.args = config
+        if size == None:
+            self.seq_len = 24 * 4 * 4
+            self.label_len = 24 * 4
+            self.pred_len = 24 * 4
+        else:
+            self.seq_len = size[0]
+            self.label_len = size[1]
+            self.pred_len = size[2]
+        # init
+        assert flag in ['train', 'test', 'val']
+        type_map = {'train': 0, 'val': 1, 'test': 2}
+        self.set_type = type_map[flag]
+
+        self.features = features
+        self.target = target
+        self.scale = scale
+        self.timeenc = timeenc
+        self.freq = freq
+        self.cycle = cycle
+
+        self.root_path = root_path
+        self.data_path = data_path
+        self.__read_data__()
+        self.collect_all_data()
+        if self.args.in_dataset_augmentation and self.set_type==0:
+            self.data_augmentation()
+        
+    def __read_data__(self):
+        self.scaler = StandardScaler()
+        df_raw = pd.read_csv(os.path.join(self.root_path,
+                                          self.data_path))
+
+        border1s = [0, 12 * 30 * 24 - self.seq_len, 12 * 30 * 24 + 4 * 30 * 24 - self.seq_len]
+        border2s = [12 * 30 * 24, 12 * 30 * 24 + 4 * 30 * 24, 12 * 30 * 24 + 8 * 30 * 24]
+
+        if self.args.test_time_train:
+            border1s = [0, 18 * 30 * 24 - self.seq_len, 20 * 30 * 24]
+            border2s = [18 * 30 * 24, 20 * 30 * 24, 20 * 30 * 24]
+
+        border1 = border1s[self.set_type]
+        border2 = border2s[self.set_type]
+
+        if self.features == 'M' or self.features == 'MS':
+            cols_data = df_raw.columns[1:]
+            df_data = df_raw[cols_data]
+        elif self.features == 'S':
+            df_data = df_raw[[self.target]]
+
+        if self.scale:
+            train_data = df_data[border1s[0]:border2s[0]]
+            self.scaler.fit(train_data.values)
+            data = self.scaler.transform(df_data.values)
+        else:
+            data = df_data.values
+
+        df_stamp = df_raw[['date']][border1:border2]
+        df_stamp['date'] = pd.to_datetime(df_stamp.date)
+        if self.timeenc == 0:
+            df_stamp['month'] = df_stamp.date.apply(lambda row: row.month, 1)
+            df_stamp['day'] = df_stamp.date.apply(lambda row: row.day, 1)
+            df_stamp['weekday'] = df_stamp.date.apply(lambda row: row.weekday(), 1)
+            df_stamp['hour'] = df_stamp.date.apply(lambda row: row.hour, 1)
+            data_stamp = df_stamp.drop(['date'], 1).values
+        elif self.timeenc == 1:
+            data_stamp = time_features(pd.to_datetime(df_stamp['date'].values), freq=self.freq)
+            data_stamp = data_stamp.transpose(1, 0)
+
+        self.data_x = data[border1:border2]
+        self.data_y = data[border1:border2]
+        self.data_stamp = data_stamp
+        self.cycle_index = (np.arange(len(data)) % self.cycle)[border1:border2]
+
+    
+    def regenerate_augmentation_data(self):
+        self.collect_all_data()
+        self.data_augmentation()
+
+    def reload_data(self, x_data, y_data, x_time, y_time):
+        self.x_data = x_data
+        self.y_data = y_data
+        self.x_time = x_time
+        self.y_time = y_time
+
+    def collect_all_data(self):
+        self.x_data = []
+        self.y_data = []
+        self.x_time = []
+        self.y_time = []
+        self.s_end_list = []
+        data_len = len(self.data_x) - self.seq_len - self.pred_len + 1
+        mask_data_len = int((1-self.args.data_size) * data_len) if self.args.data_size < 1 else 0
+        for i in range(len(self.data_x) - self.seq_len - self.pred_len + 1):
+            if (self.set_type == 0 and i >= mask_data_len) or self.set_type != 0:
+                s_begin = i
+                s_end = s_begin + self.seq_len
+                r_begin = s_end - self.label_len
+                r_end = r_begin + self.label_len + self.pred_len
+                self.x_data.append(self.data_x[s_begin:s_end])
+                self.y_data.append(self.data_y[r_begin:r_end])
+                self.x_time.append(self.data_stamp[s_begin:s_end]) 
+                self.y_time.append(self.data_stamp[r_begin:r_end])
+                self.s_end_list.append(s_end)  # Save s_end
+
+    def data_augmentation(self):
+        origin_len = len(self.x_data)
+        if not self.args.closer_data_aug_more:
+            aug_size = [self.args.aug_data_size for i in range(origin_len)]
+        else:
+            aug_size = [int(self.args.aug_data_size * i/origin_len) + 1 for i in range(origin_len)]
+
+        for i in range(origin_len):
+            for _ in range(aug_size[i]):
+                aug = augmentation('dataset')
+                if self.args.aug_method == 'f_mask':
+                    x,y = aug.freq_dropout(self.x_data[i],self.y_data[i],dropout_rate=self.args.aug_rate)
+                elif self.args.aug_method == 'f_mix':
+                    rand = float(np.random.random(1))
+                    i2 = int(rand*len(self.x_data))
+                    x,y = aug.freq_mix(self.x_data[i],self.y_data[i],self.x_data[i2],self.y_data[i2],dropout_rate=self.args.aug_rate)
+                elif self.args.aug_method == 'sp':
+                    x,y = aug.seasonal_shuffle(self.x_data[i],self.y_data[i], patch_len=self.args.patch_len, stride=self.args.skip, rate=self.args.aug_rate, season=self.args.season, frequency=self.args.aug_freq)
+                else: 
+                    raise ValueError
+                self.x_data.append(x)
+                self.y_data.append(y)
+                self.x_time.append(self.x_time[i]) 
+                self.y_time.append(self.y_time[i])
+                self.s_end_list.append(self.s_end_list[i])
+
+    def __getitem__(self, index):
+        seq_x = self.x_data[index]
+        seq_y = self.y_data[index]
+        cycle_index = torch.tensor(self.cycle_index[self.s_end_list[index]])
+        return seq_x, seq_y, self.x_time[index], self.y_time[index], cycle_index
+
+    def __len__(self):
+        return len(self.x_data)
+
+    def inverse_transform(self, data):
+        return self.scaler.inverse_transform(data)
+
+
+class Dataset_ETT_minute(Dataset):
+    def __init__(self, config, root_path, flag='train', size=None,
+                 features='S', data_path='ETTm1.csv',
+                 target='OT', scale=True, timeenc=0, freq='t', cycle = None):
+        self.args = config
+        if size == None:
+            self.seq_len = 24 * 4 * 4
+            self.label_len = 24 * 4
+            self.pred_len = 24 * 4
+        else:
+            self.seq_len = size[0]
+            self.label_len = size[1]
+            self.pred_len = size[2]
+        # init
+        assert flag in ['train', 'test', 'val']
+        type_map = {'train': 0, 'val': 1, 'test': 2}
+        self.set_type = type_map[flag]
+
+        self.features = features
+        self.target = target
+        self.scale = scale
+        self.timeenc = timeenc
+        self.freq = freq
+        self.cycle = cycle
+
+        self.root_path = root_path
+        self.data_path = data_path
+        self.__read_data__()
+        self.collect_all_data()
+        if self.args.in_dataset_augmentation and self.set_type==0:
+            self.data_augmentation()
+
+    def __read_data__(self):
+        self.scaler = StandardScaler()
+        df_raw = pd.read_csv(os.path.join(self.root_path,
+                                          self.data_path))
+
+        border1s = [0, 12 * 30 * 24 * 4 - self.seq_len, 12 * 30 * 24 * 4 + 4 * 30 * 24 * 4 - self.seq_len]
+        border2s = [12 * 30 * 24 * 4, 12 * 30 * 24 * 4 + 4 * 30 * 24 * 4, 12 * 30 * 24 * 4 + 8 * 30 * 24 * 4]
+
+        if self.args.test_time_train:
+            border1s = [0, 18 * 30 * 24 * 4 - self.seq_len, 20 * 30 * 24 * 4]
+            border2s = [18 * 30 * 24 * 4, 20 * 30 * 24 * 4, 20 * 30 * 24 * 4]
+
+        border1 = border1s[self.set_type]
+        border2 = border2s[self.set_type]
+
+        if self.features == 'M' or self.features == 'MS':
+            cols_data = df_raw.columns[1:]
+            df_data = df_raw[cols_data]
+        elif self.features == 'S':
+            df_data = df_raw[[self.target]]
+
+        if self.scale:
+            train_data = df_data[border1s[0]:border2s[0]]
+            self.scaler.fit(train_data.values)
+            data = self.scaler.transform(df_data.values)
+        else:
+            data = df_data.values
+
+        df_stamp = df_raw[['date']][border1:border2]
+        df_stamp['date'] = pd.to_datetime(df_stamp.date)
+        if self.timeenc == 0:
+            df_stamp['month'] = df_stamp.date.apply(lambda row: row.month, 1)
+            df_stamp['day'] = df_stamp.date.apply(lambda row: row.day, 1)
+            df_stamp['weekday'] = df_stamp.date.apply(lambda row: row.weekday(), 1)
+            df_stamp['hour'] = df_stamp.date.apply(lambda row: row.hour, 1)
+            df_stamp['minute'] = df_stamp.date.apply(lambda row: row.minute, 1)
+            df_stamp['minute'] = df_stamp.minute.map(lambda x: x // 15)
+            data_stamp = df_stamp.drop(['date'], 1).values
+        elif self.timeenc == 1:
+            data_stamp = time_features(pd.to_datetime(df_stamp['date'].values), freq=self.freq)
+            data_stamp = data_stamp.transpose(1, 0)
+
+        self.data_x = data[border1:border2]
+        self.data_y = data[border1:border2]
+        self.data_stamp = data_stamp
+        self.cycle_index = (np.arange(len(data)) % self.cycle)[border1:border2]
+
+    def regenerate_augmentation_data(self):
+        self.collect_all_data()
+        self.data_augmentation()
+
+    def reload_data(self, x_data, y_data, x_time, y_time):
+        self.x_data = x_data
+        self.y_data = y_data
+        self.x_time = x_time
+        self.y_time = y_time
+
+    def collect_all_data(self):
+        self.x_data = []
+        self.y_data = []
+        self.x_time = []
+        self.y_time = []
+        self.s_end_list = []
+        data_len = len(self.data_x) - self.seq_len - self.pred_len + 1
+        mask_data_len = int((1-self.args.data_size) * data_len) if self.args.data_size < 1 else 0
+        for i in range(len(self.data_x) - self.seq_len - self.pred_len + 1):
+            if (self.set_type == 0 and i >= mask_data_len) or self.set_type != 0:
+                s_begin = i
+                s_end = s_begin + self.seq_len
+                r_begin = s_end - self.label_len
+                r_end = r_begin + self.label_len + self.pred_len
+                self.x_data.append(self.data_x[s_begin:s_end])
+                self.y_data.append(self.data_y[r_begin:r_end])
+                self.x_time.append(self.data_stamp[s_begin:s_end]) 
+                self.y_time.append(self.data_stamp[r_begin:r_end])
+                self.s_end_list.append(s_end)  # 💡 Save s_end
+
+    def data_augmentation(self):
+        origin_len = len(self.x_data)
+        if not self.args.closer_data_aug_more:
+            aug_size = [self.args.aug_data_size for i in range(origin_len)]
+        else:
+            aug_size = [int(self.args.aug_data_size * i/origin_len) + 1 for i in range(origin_len)]
+
+        for i in range(origin_len):
+            for _ in range(aug_size[i]):
+                aug = augmentation('dataset')
+                if self.args.aug_method == 'f_mask':
+                    x,y = aug.freq_dropout(self.x_data[i],self.y_data[i],dropout_rate=self.args.aug_rate)
+                elif self.args.aug_method == 'f_mix':
+                    rand = float(np.random.random(1))
+                    i2 = int(rand*len(self.x_data))
+                    x,y = aug.freq_mix(self.x_data[i],self.y_data[i],self.x_data[i2],self.y_data[i2],dropout_rate=self.args.aug_rate)
+                elif self.args.aug_method == 'sp':
+                    x,y = aug.seasonal_shuffle(self.x_data[i],self.y_data[i], patch_len=self.args.patch_len, stride=self.args.skip, rate=self.args.aug_rate, season=self.args.season, frequency=self.args.aug_freq)
+                else: 
+                    raise ValueError
+                self.x_data.append(x)
+                self.y_data.append(y)
+                self.x_time.append(self.x_time[i]) 
+                self.y_time.append(self.y_time[i])
+                self.s_end_list.append(self.s_end_list[i])
+
+    def __getitem__(self, index):
+        seq_x = self.x_data[index]
+        seq_y = self.y_data[index]
+        cycle_index = torch.tensor(self.cycle_index[self.s_end_list[index]])
+        return seq_x, seq_y, self.x_time[index], self.y_time[index], cycle_index
+
+    def __len__(self):
+        return len(self.x_data)
+
+    def inverse_transform(self, data):
+        return self.scaler.inverse_transform(data)
+
+
+class Dataset_Custom(Dataset):
+    def __init__(self, config, root_path, flag='train', size=None,
+                 features='S', data_path='ETTh1.csv',
+                 target='OT', scale=True, timeenc=0, freq='h', cycle = None):
+        self.args = config
+        # info
+        if size == None:
+            self.seq_len = 24 * 4 * 4
+            self.label_len = 24 * 4
+            self.pred_len = 24 * 4
+        else:
+            self.seq_len = size[0]
+            self.label_len = size[1]
+            self.pred_len = size[2]
+        # init
+        assert flag in ['train', 'test', 'val']
+        type_map = {'train': 0, 'val': 1, 'test': 2}
+        self.set_type = type_map[flag]
+
+        self.features = features
+        self.target = target
+        self.scale = scale
+        self.timeenc = timeenc
+        self.freq = freq
+        self.cycle = cycle
+
+        self.root_path = root_path
+        self.data_path = data_path
+        self.__read_data__()
+        self.collect_all_data()
+        if self.args.in_dataset_augmentation and self.set_type==0:
+            self.data_augmentation()
+
+    def __read_data__(self):
+        self.scaler = StandardScaler()
+        df_raw = pd.read_csv(os.path.join(self.root_path,
+                                          self.data_path))
+        '''
+        df_raw.columns: ['date', ...(other features), target feature]
+        '''
+        cols = list(df_raw.columns)
+        cols.remove(self.target)
+        cols.remove('date')
+        df_raw = df_raw[['date'] + cols + [self.target]]
+        # print(cols)
+        num_train = int(len(df_raw) * 0.7)
+        num_test = int(len(df_raw) * 0.2)
+        num_vali = len(df_raw) - num_train - num_test
+        border1s = [0, num_train - self.seq_len, len(df_raw) - num_test - self.seq_len]
+        border2s = [num_train, num_train + num_vali, len(df_raw)]
+
+        if self.args.test_time_train:
+            num_train = int(len(df_raw) * 0.9)
+            border1s = [0, num_train - self.seq_len, len(df_raw)]
+            border2s = [num_train, len(df_raw), len(df_raw)]
+        
+        border1 = border1s[self.set_type]
+        border2 = border2s[self.set_type]
+
+        if self.features == 'M' or self.features == 'MS':
+            cols_data = df_raw.columns[1:]
+            df_data = df_raw[cols_data]
+        elif self.features == 'S':
+            df_data = df_raw[[self.target]]
+
+        if self.scale:
+            train_data = df_data[border1s[0]:border2s[0]]
+            self.scaler.fit(train_data.values)
+            # print(self.scaler.mean_)
+            # exit()
+            data = self.scaler.transform(df_data.values)
+        else:
+            data = df_data.values
+
+        df_stamp = df_raw[['date']][border1:border2]
+        df_stamp['date'] = pd.to_datetime(df_stamp.date)
+        if self.timeenc == 0:
+            df_stamp['month'] = df_stamp.date.apply(lambda row: row.month, 1)
+            df_stamp['day'] = df_stamp.date.apply(lambda row: row.day, 1)
+            df_stamp['weekday'] = df_stamp.date.apply(lambda row: row.weekday(), 1)
+            df_stamp['hour'] = df_stamp.date.apply(lambda row: row.hour, 1)
+            data_stamp = df_stamp.drop(['date'], 1).values
+        elif self.timeenc == 1:
+            data_stamp = time_features(pd.to_datetime(df_stamp['date'].values), freq=self.freq)
+            data_stamp = data_stamp.transpose(1, 0)
+
+        self.data_x = data[border1:border2]
+        self.data_y = data[border1:border2]
+        self.data_stamp = data_stamp
+        self.cycle_index = (np.arange(len(data)) % self.cycle)[border1:border2]
+
+    def regenerate_augmentation_data(self):
+        self.collect_all_data()
+        self.data_augmentation()
+
+    def reload_data(self, x_data, y_data, x_time, y_time):
+        self.x_data = x_data
+        self.y_data = y_data
+        self.x_time = x_time
+        self.y_time = y_time
+        
+    def collect_all_data(self):
+        self.x_data = []
+        self.y_data = []
+        self.x_time = []
+        self.y_time = []
+        self.s_end_list = []
+        data_len = len(self.data_x) - self.seq_len - self.pred_len + 1
+        mask_data_len = int((1-self.args.data_size) * data_len) if self.args.data_size < 1 else 0
+        for i in range(len(self.data_x) - self.seq_len - self.pred_len + 1):
+            if (self.set_type == 0 and i >= mask_data_len) or self.set_type != 0:
+                s_begin = i
+                s_end = s_begin + self.seq_len
+                r_begin = s_end - self.label_len
+                r_end = r_begin + self.label_len + self.pred_len
+                self.x_data.append(self.data_x[s_begin:s_end])
+                self.y_data.append(self.data_y[r_begin:r_end])
+                self.x_time.append(self.data_stamp[s_begin:s_end]) 
+                self.y_time.append(self.data_stamp[r_begin:r_end])
+                self.s_end_list.append(s_end)  # 💡 Save s_end
+        
+    def data_augmentation(self):
+        origin_len = len(self.x_data)
+        if not self.args.closer_data_aug_more:
+            aug_size = [self.args.aug_data_size for i in range(origin_len)]
+        else:
+            aug_size = [int(self.args.aug_data_size * i/origin_len) + 1 for i in range(origin_len)]
+
+        for i in range(origin_len):
+            for _ in range(aug_size[i]):
+                aug = augmentation('dataset')
+                if self.args.aug_method == 'f_mask':
+                    x,y = aug.freq_dropout(self.x_data[i],self.y_data[i],dropout_rate=self.args.aug_rate)
+                elif self.args.aug_method == 'f_mix':
+                    rand = float(np.random.random(1))
+                    i2 = int(rand*len(self.x_data))
+                    x,y = aug.freq_mix(self.x_data[i],self.y_data[i],self.x_data[i2],self.y_data[i2],dropout_rate=self.args.aug_rate)
+                elif self.args.aug_method == 'sp':
+                    x,y = aug.seasonal_shuffle(self.x_data[i],self.y_data[i], patch_len=self.args.patch_len, stride=self.args.skip, rate=self.args.aug_rate, season=self.args.season, frequency=self.args.aug_freq)
+                else: 
+                    raise ValueError
+                self.x_data.append(x)
+                self.y_data.append(y)
+                self.x_time.append(self.x_time[i]) 
+                self.y_time.append(self.y_time[i])
+                self.s_end_list.append(self.s_end_list[i])
+
+    def __getitem__(self, index):
+        seq_x = self.x_data[index]
+        seq_y = self.y_data[index]
+        cycle_index = torch.tensor(self.cycle_index[self.s_end_list[index]])
+        return seq_x, seq_y, self.x_time[index], self.y_time[index], cycle_index
+
+    def __len__(self):
+        return len(self.x_data)
+
+    def inverse_transform(self, data):
+        return self.scaler.inverse_transform(data)
+    
+
+class Dataset_Pred(Dataset):
+    def __init__(self, args, root_path, flag='pred', size=None,
+                 features='S', data_path='ETTh1.csv',
+                 target='OT', scale=True, inverse=False, timeenc=0, freq='15min', cols=None):
+        # size [seq_len, label_len, pred_len]
+        # info
+        if size == None:
+            self.seq_len = 24 * 4 * 4
+            self.label_len = 24 * 4
+            self.pred_len = 24 * 4
+        else:
+            self.seq_len = size[0]
+            self.label_len = size[1]
+            self.pred_len = size[2]
+        # init
+        assert flag in ['pred']
+
+        self.features = features
+        self.target = target
+        self.scale = scale
+        self.inverse = inverse
+        self.timeenc = timeenc
+        self.freq = freq
+        self.cols = cols
+        self.root_path = root_path
+        self.data_path = data_path
+        self.__read_data__()
+
+    def __read_data__(self):
+        self.scaler = StandardScaler()
+        df_raw = pd.read_csv(os.path.join(self.root_path,
+                                          self.data_path))
+        '''
+        df_raw.columns: ['date', ...(other features), target feature]
+        '''
+        if self.cols:
+            cols = self.cols.copy()
+            cols.remove(self.target)
+        else:
+            cols = list(df_raw.columns)
+            cols.remove(self.target)
+            cols.remove('date')
+        df_raw = df_raw[['date'] + cols + [self.target]]
+        border1 = len(df_raw) - self.seq_len
+        border2 = len(df_raw)
+
+        if self.features == 'M' or self.features == 'MS':
+            cols_data = df_raw.columns[1:]
+            df_data = df_raw[cols_data]
+        elif self.features == 'S':
+            df_data = df_raw[[self.target]]
+
+        if self.scale:
+            self.scaler.fit(df_data.values)
+            data = self.scaler.transform(df_data.values)
+        else:
+            data = df_data.values
+
+        tmp_stamp = df_raw[['date']][border1:border2]
+        tmp_stamp['date'] = pd.to_datetime(tmp_stamp.date)
+        pred_dates = pd.date_range(tmp_stamp.date.values[-1], periods=self.pred_len + 1, freq=self.freq)
+
+        df_stamp = pd.DataFrame(columns=['date'])
+        df_stamp.date = list(tmp_stamp.date.values) + list(pred_dates[1:])
+        if self.timeenc == 0:
+            df_stamp['month'] = df_stamp.date.apply(lambda row: row.month, 1)
+            df_stamp['day'] = df_stamp.date.apply(lambda row: row.day, 1)
+            df_stamp['weekday'] = df_stamp.date.apply(lambda row: row.weekday(), 1)
+            df_stamp['hour'] = df_stamp.date.apply(lambda row: row.hour, 1)
+            df_stamp['minute'] = df_stamp.date.apply(lambda row: row.minute, 1)
+            df_stamp['minute'] = df_stamp.minute.map(lambda x: x // 15)
+            data_stamp = df_stamp.drop(['date'], 1).values
+        elif self.timeenc == 1:
+            data_stamp = time_features(pd.to_datetime(df_stamp['date'].values), freq=self.freq)
+            data_stamp = data_stamp.transpose(1, 0)
+
+        self.data_x = data[border1:border2]
+        if self.inverse:
+            self.data_y = df_data.values[border1:border2]
+        else:
+            self.data_y = data[border1:border2]
+        self.data_stamp = data_stamp
+
+    def __getitem__(self, index):
+        s_begin = index
+        s_end = s_begin + self.seq_len
+        r_begin = s_end - self.label_len
+        r_end = r_begin + self.label_len + self.pred_len
+
+        seq_x = self.data_x[s_begin:s_end]
+        if self.inverse:
+            seq_y = self.data_x[r_begin:r_begin + self.label_len]
+        else:
+            seq_y = self.data_y[r_begin:r_begin + self.label_len]
+        seq_x_mark = self.data_stamp[s_begin:s_end]
+        seq_y_mark = self.data_stamp[r_begin:r_end]
+
+        return seq_x, seq_y, seq_x_mark, seq_y_mark
+
+    def __len__(self):
+        return len(self.data_x) - self.seq_len + 1
+
+    def inverse_transform(self, data):
+        return self.scaler.inverse_transform(data)
+
+
+
+class Dataset_PEMS(Dataset):
+    def __init__(self, config, root_path, flag='train', size=None,
+                 features='S', data_path='ETTh1.csv',
+                 target='OT', scale=True, timeenc=0, freq='h', cycle = None):
+        # size [seq_len, label_len, pred_len]
+        # info
+        self.seq_len = size[0]
+        self.label_len = size[1]
+        self.pred_len = size[2]
+        # init
+        assert flag in ['train', 'test', 'val']
+        type_map = {'train': 0, 'val': 1, 'test': 2}
+        self.set_type = type_map[flag]
+
+        self.features = features
+        self.target = target
+        self.scale = scale
+        self.timeenc = timeenc
+        self.freq = freq
+        self.cycle = cycle
+
+
+        self.root_path = root_path
+        self.data_path = data_path
+        
+        self.__read_data__()
+
+    def __read_data__(self):
+        self.scaler = StandardScaler()
+        data_file = os.path.join(self.root_path, self.data_path)
+        data = np.load(data_file, allow_pickle=True)
+        data = data['data'][:, :, 0]
+
+        train_ratio = 0.6
+        valid_ratio = 0.2
+        train_data = data[:int(train_ratio * len(data))]
+        valid_data = data[int(train_ratio * len(data)): int((train_ratio + valid_ratio) * len(data))]
+        test_data = data[int((train_ratio + valid_ratio) * len(data)):]
+        total_data = [train_data, valid_data, test_data]
+        data = total_data[self.set_type]
+
+        if self.scale:
+            self.scaler.fit(train_data)
+            data = self.scaler.transform(data)
+
+        df = pd.DataFrame(data)
+        df = df.fillna(method='ffill', limit=len(df)).fillna(method='bfill', limit=len(df)).values
+
+        self.data_x = df
+        self.data_y = df
+        self.cycle_index = (np.arange(len(data)) % self.cycle)
+
+    def __getitem__(self, index):
+        s_begin = index
+        s_end = s_begin + self.seq_len
+        r_begin = s_end - self.label_len
+        r_end = r_begin + self.label_len + self.pred_len
+
+        seq_x = self.data_x[s_begin:s_end]
+        seq_y = self.data_y[r_begin:r_end]
+        seq_x_mark = torch.zeros((seq_x.shape[0], 1))
+        seq_y_mark = torch.zeros((seq_x.shape[0], 1))
+        cycle_index = 24
+
+        return seq_x, seq_y, seq_x_mark, seq_y_mark, cycle_index
+
+    def __len__(self):
+        return len(self.data_x) - self.seq_len - self.pred_len + 1
+
+    def inverse_transform(self, data):
+        return self.scaler.inverse_transform(data)
+

time_series_forecasting/exp/exp_basic.py

@@ -0,0 +1,40 @@
+# Adapted from dominant-shuffle (https://github.com/zuojie2024/dominant-shuffle) and FrAug implementations
+# Original works: Dominant Shuffle by Kai Zhao et al., FrAug by Muxi Chen et al.
+
+import os
+import torch
+import numpy as np
+
+
+class Exp_Basic(object):
+    def __init__(self, args):
+        self.args = args
+        self.device = self._acquire_device()
+        self.model = self._build_model().to(self.device)
+
+    def _build_model(self):
+        raise NotImplementedError
+        return None
+
+    def _acquire_device(self):
+        if self.args.use_gpu:
+            os.environ["CUDA_VISIBLE_DEVICES"] = str(
+                self.args.gpu) if not self.args.use_multi_gpu else self.args.devices
+            device = torch.device('cuda:{}'.format(self.args.gpu))
+            print('Use GPU: cuda:{}'.format(self.args.gpu))
+        else:
+            device = torch.device('cpu')
+            print('Use CPU')
+        return device
+
+    def _get_data(self):
+        pass
+
+    def vali(self):
+        pass
+
+    def train(self):
+        pass
+
+    def test(self):
+        pass

time_series_forecasting/exp/exp_main.py

@@ -0,0 +1,493 @@
+# Adapted from dominant-shuffle (https://github.com/zuojie2024/dominant-shuffle) and FrAug implementations
+# Original works: Dominant Shuffle by Kai Zhao et al., FrAug by Muxi Chen et al.
+# Modified and extended by Jafar Bakhshaliyev (2025)
+
+from data_provider.data_factory import data_provider
+from exp.exp_basic import Exp_Basic
+from models import Autoformer, LightTS, MICN , TiDE, iTransformer, SCINet, Informer, Transformer, DLinear, Linear, NLinear, PatchTST, TSMixer, CycleNet
+from utils.tools import EarlyStopping, adjust_learning_rate, visual, test_params_flop
+from utils.metrics import metric
+import numpy as np
+import torch
+import torch.nn as nn
+from torch import optim
+from utils.augmentations import augmentation
+import os
+import time
+from torch.optim import lr_scheduler 
+import warnings
+import matplotlib.pyplot as plt
+from torch.optim.lr_scheduler import CosineAnnealingLR
+
+warnings.filterwarnings('ignore')
+
+class Exp_Main(Exp_Basic):
+    def __init__(self, args):
+        super(Exp_Main, self).__init__(args)
+
+    def _build_model(self):
+        model_dict = {
+            'Autoformer': Autoformer,
+            'LightTS': LightTS,
+            'MICN':MICN,
+            'TiDE': TiDE,
+            'iTransformer': iTransformer,
+            'SCINet': SCINet,
+            'Transformer': Transformer,
+            'Informer': Informer,
+            'DLinear': DLinear,
+            'NLinear': NLinear,
+            'Linear': Linear,
+            'TSMixer': TSMixer,
+            'PatchTST': PatchTST,
+            'CycleNet': CycleNet
+        }
+        model = model_dict[self.args.model].Model(self.args).float()
+
+        if self.args.use_multi_gpu and self.args.use_gpu:
+            model = nn.DataParallel(model, device_ids=self.args.device_ids)
+        return model
+
+    def _get_data(self, flag):
+        data_set, data_loader = data_provider(self.args, flag)
+        return data_set, data_loader
+
+    def _select_optimizer(self):
+        if self.args.model == 'needed':
+            model_optim = optim.Adam(self.model.parameters(), lr=self.args.learning_rate, betas=(0.9, 0.999), eps=1e-08)
+            scheduler = CosineAnnealingLR(optimizer=model_optim, T_max=30 * 8640, eta_min=0)
+        else:
+            model_optim = optim.Adam(self.model.parameters(), lr=self.args.learning_rate)
+            scheduler = None
+        return model_optim, scheduler
+
+
+    def _select_criterion(self):
+        criterion = nn.MSELoss()
+        return criterion
+
+    def vali(self, vali_data, vali_loader, criterion):
+        total_loss = []
+        self.model.eval()
+        with torch.no_grad():
+            for i, (batch_x, batch_y, batch_x_mark, batch_y_mark, batch_cycle) in enumerate(vali_loader):
+                batch_x = batch_x.float().to(self.device)
+                batch_y = batch_y.float()
+                batch_cycle = batch_cycle.int().to(self.device)
+
+                if 'PEMS' in self.args.data or 'Solar' in self.args.data:
+                    batch_x_mark = None
+                    batch_y_mark = None
+                else:
+                    batch_x_mark = batch_x_mark.float().to(self.device)
+                    batch_y_mark = batch_y_mark.float().to(self.device)
+
+                # decoder input
+                dec_inp = torch.zeros_like(batch_y[:, -self.args.pred_len:, :]).float()
+                dec_inp = torch.cat([batch_y[:, :self.args.label_len, :], dec_inp], dim=1).float().to(self.device)
+
+                # encoder - decoder
+                if self.args.use_amp:
+                    with torch.cuda.amp.autocast():
+                        if not self.args.use_former:
+                            if any(substr in self.args.model for substr in {'Cycle'}):
+                                outputs = self.model(batch_x, batch_cycle)
+                            else:
+                                outputs = self.model(batch_x)
+                        else:
+                            if self.args.output_attention:
+                                outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)[0]
+                            else:
+                                outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
+                else:
+                    if not self.args.use_former:
+                        if any(substr in self.args.model for substr in {'Cycle'}):
+                            outputs = self.model(batch_x, batch_cycle)
+                        else:
+                            outputs = self.model(batch_x)
+                    else:
+                        if self.args.output_attention:
+                            outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)[0]
+                        else:
+                            outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
+
+                f_dim = -1 if self.args.features == 'MS' else 0
+                outputs = outputs[:, -self.args.pred_len:, f_dim:]
+                batch_y = batch_y[:, -self.args.pred_len:, f_dim:].to(self.device)
+
+                pred = outputs.detach().cpu()
+                true = batch_y.detach().cpu()
+
+                loss = criterion(pred, true)
+
+                total_loss.append(loss)
+        total_loss = np.average(total_loss)
+        self.model.train()
+        return total_loss
+
+    def train(self, setting):
+        train_data, train_loader = self._get_data(flag='train')
+        vali_data, vali_loader = self._get_data(flag='val')
+        test_data, test_loader = self._get_data(flag='test')
+
+        path = os.path.join(self.args.checkpoints, setting)
+        if not os.path.exists(path):
+            os.makedirs(path)
+
+        time_now = time.time()
+
+        train_steps = len(train_loader)
+        early_stopping = EarlyStopping(patience=self.args.patience, verbose=True)
+
+        model_optim, scheduler = self._select_optimizer()
+        criterion = self._select_criterion()
+
+        if self.args.model == 'PatchTST':
+            scheduler = lr_scheduler.OneCycleLR(optimizer = model_optim,
+                                            steps_per_epoch = train_steps,
+                                            pct_start = self.args.pct_start,
+                                            epochs = self.args.train_epochs,
+                                            max_lr = self.args.learning_rate)
+
+        if self.args.use_amp:
+            scaler = torch.cuda.amp.GradScaler()
+
+        for epoch in range(self.args.train_epochs):
+            iter_count = 0
+            train_loss = []
+
+            self.model.train()
+            epoch_time = time.time()
+            
+            if self.args.in_dataset_augmentation:
+                train_loader.dataset.regenerate_augmentation_data()
+
+            for i, (batch_x, batch_y, batch_x_mark, batch_y_mark, batch_cycle) in enumerate(train_loader):
+                iter_count += 1
+                model_optim.zero_grad()
+
+                batch_x = batch_x.float().to(self.device)
+                batch_y = batch_y.float().to(self.device)
+                batch_cycle = batch_cycle.int().to(self.device)
+                if self.args.in_batch_augmentation and not self.args.wo_original_set:
+                    batch_cycle = torch.cat([batch_cycle, batch_cycle], dim=0)
+
+                if 'PEMS' in self.args.data or 'Solar' in self.args.data:
+                    batch_x_mark = None
+                    batch_y_mark = None
+                else:
+                    batch_x_mark = batch_x_mark.float().to(self.device)
+                    batch_y_mark = batch_y_mark.float().to(self.device)
+
+                if self.args.in_batch_augmentation:
+                    aug = augmentation('batch')
+                    methods = { 
+                                'f_mask':aug.freq_mask, 
+                                'f_mix': aug.freq_mix, 
+                                'noise': aug.noise,    
+                                'warp': aug.warping, 
+                                'flip': aug.flipping, 
+                                'mask': aug.masking, 
+                                'mask_seg': aug.masking_seg, 
+                                'noise_input':aug.noise_input, 
+                                'dom_shuffle':aug.dom_shuffle,
+                                'w_mask': aug.wave_mask,
+                                'w_mix': aug.wave_mix,
+                                'f_add': aug.freq_add,
+                                'f_pool': aug.freq_pool,
+                                'tps': aug.temporal_patch_shuffle,
+                                'robust_m': aug.robusttad_m,
+                                'robust_p': aug.robusttad_p,
+                                'upsample': aug.upsample,
+                                'asd': aug.asd,
+                                'mbb': aug.mbb,
+                                }
+                        
+                    if self.args.wo_original_set:
+                        xy = methods[self.args.aug_method](batch_x, batch_y[:, -self.args.pred_len:, :], rate=self.args.aug_rate)
+                        batch_x, batch_y = xy[:, :self.args.seq_len, :], xy[:, -self.args.label_len-self.args.pred_len:, :]
+                    else:
+                        for step in range(self.args.aug_data_size):
+
+                            if self.args.aug_method == 'w_mask' or self.args.aug_method == 'w_mix':
+                                xy2, indices = methods[self.args.aug_method](batch_x, batch_y[:, -self.args.pred_len:, :], self.args.rates, self.args.wavelet, self.args.level, self.args.uniform, self.args.sampling_rate)
+                                if not ('PEMS' in self.args.data or 'Solar' in self.args.data):
+                                    batch_x2_mark, batch_y2_mark = batch_x_mark[indices], batch_y_mark[indices]
+                            elif self.args.aug_method.startswith('tps'):
+                                xy2 = methods[self.args.aug_method](batch_x, batch_y[:, -self.args.pred_len:, :], self.args.aug_patch_len, self.args.aug_stride, self.args.aug_rate)
+                                batch_x2_mark, batch_y2_mark = batch_x_mark, batch_y_mark
+                            elif self.args.aug_method.startswith('robust'):
+                                xy2 = methods[self.args.aug_method](batch_x, batch_y[:, -self.args.pred_len:, :], self.args.aug_rate, self.args.K_num, self.args.seg_ratio)
+                                batch_x2_mark, batch_y2_mark = batch_x_mark, batch_y_mark
+                            elif self.args.aug_method.startswith('mbb'):
+                                xy2 = methods[self.args.aug_method](batch_x, batch_y[:, -self.args.pred_len:, :], self.args.block_size)
+                                batch_x2_mark, batch_y2_mark = batch_x_mark, batch_y_mark
+                            else:
+                                xy2 = methods[self.args.aug_method](batch_x, batch_y[:, -self.args.pred_len:, :], rate=self.args.aug_rate)
+                                batch_x2_mark, batch_y2_mark = batch_x_mark, batch_y_mark
+                            
+                            batch_x2, batch_y2 = xy2[:, :self.args.seq_len, :], xy2[:, -self.args.label_len-self.args.pred_len:, :]        
+                            
+                            batch_x = torch.cat([batch_x,batch_x2],dim=0)
+                            batch_y = torch.cat([batch_y,batch_y2],dim=0)
+
+                            if not ('PEMS' in self.args.data or 'Solar' in self.args.data):
+                                batch_x_mark = torch.cat([batch_x_mark,batch_x2_mark],dim=0)
+                                batch_y_mark = torch.cat([batch_y_mark,batch_y2_mark],dim=0)
+
+                        
+                # decoder input
+                dec_inp = torch.zeros_like(batch_y[:, -self.args.pred_len:, :]).float()
+                dec_inp = torch.cat([batch_y[:, :self.args.label_len, :], dec_inp], dim=1).float().to(self.device)
+                # encoder - decoder
+                if self.args.use_amp:
+                    with torch.cuda.amp.autocast():
+                        if not self.args.use_former:
+                            if any(substr in self.args.model for substr in {'Cycle'}):
+                                outputs = self.model(batch_x, batch_cycle)
+                            else:
+                                outputs = self.model(batch_x)
+                        else:
+                            if self.args.output_attention:
+                                outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)[0]
+                            else:
+                                outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
+                    
+                else:
+                    if not self.args.use_former:
+                        if any(substr in self.args.model for substr in {'Cycle'}):
+                            outputs = self.model(batch_x, batch_cycle)
+                        else:
+                            outputs = self.model(batch_x)
+                    else:
+                        if self.args.output_attention:
+                            outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)[0]
+                            
+                        else:
+                            outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
+
+                f_dim = -1 if self.args.features == 'MS' else 0
+                outputs = outputs[:, -self.args.pred_len:, f_dim:]
+                batch_y = batch_y[:, -self.args.pred_len:, f_dim:].to(self.device)
+                loss = criterion(outputs, batch_y)
+                train_loss.append(loss.item())
+
+                if (i + 1) % 100 == 0:
+                    print("\titers: {0}, epoch: {1} | loss: {2:.7f}".format(i + 1, epoch + 1, loss.item()))
+                    speed = (time.time() - time_now) / iter_count
+                    left_time = speed * ((self.args.train_epochs - epoch) * train_steps - i)
+                    print('\tspeed: {:.4f}s/iter; left time: {:.4f}s'.format(speed, left_time))
+                    iter_count = 0
+                    time_now = time.time()
+
+                if self.args.use_amp:
+                    scaler.scale(loss).backward()
+                    scaler.step(model_optim)
+                    scaler.update()
+                else:
+                    loss.backward()
+                    model_optim.step()
+
+                if self.args.lradj == 'TST':
+                   adjust_learning_rate(model_optim, scheduler, epoch + 1, self.args, printout=False)
+                   scheduler.step()
+
+            print("Epoch: {} cost time: {}".format(epoch + 1, time.time() - epoch_time))
+            train_loss = np.average(train_loss)
+            vali_loss = self.vali(vali_data, vali_loader, criterion)
+            test_loss = self.vali(test_data, test_loader, criterion)
+
+            print("Epoch: {0}, Steps: {1} | Train Loss: {2:.7f} Vali Loss: {3:.7f} Test Loss: {4:.7f}".format(
+                epoch + 1, train_steps, train_loss, vali_loss, test_loss))
+            early_stopping(vali_loss, self.model, path)
+            
+            if early_stopping.early_stop:
+                print("Early stopping")
+                break
+
+            if self.args.lradj != 'TST':
+                adjust_learning_rate(model_optim, scheduler, epoch + 1, self.args)
+            else:
+                print('Updating learning rate to {}'.format(scheduler.get_last_lr()[0]))
+
+
+            #if self.args.model == 'needed':
+                #scheduler.step()
+           # else:
+               # adjust_learning_rate(model_optim, epoch + 1, self.args)
+                
+
+
+        best_model_path = path + '/' + 'checkpoint.pth'
+        self.model.load_state_dict(torch.load(best_model_path))
+        min_val_loss = early_stopping.get_val_loss_min() # delete after search everywhere in train
+
+        return self.model, min_val_loss
+
+    def test(self, setting, test=0):
+        test_data, test_loader = self._get_data(flag='test')
+        
+        if test:
+            print('loading model')
+            self.model.load_state_dict(torch.load(os.path.join('./checkpoints/' + setting, 'checkpoint.pth')))
+
+        preds = []
+        trues = []
+        inputx = []
+        folder_path = './test_results/' + setting + '/'
+        if not os.path.exists(folder_path):
+            os.makedirs(folder_path)
+
+        self.model.eval()
+        with torch.no_grad():
+            for i, (batch_x, batch_y, batch_x_mark, batch_y_mark,batch_cycle ) in enumerate(test_loader):
+                batch_x = batch_x.float().to(self.device)
+                batch_y = batch_y.float().to(self.device)
+                batch_cycle = batch_cycle.int().to(self.device)
+                
+                if 'PEMS' in self.args.data or 'Solar' in self.args.data:
+                    batch_x_mark = None
+                    batch_y_mark = None
+                else:
+                    batch_x_mark = batch_x_mark.float().to(self.device)
+                    batch_y_mark = batch_y_mark.float().to(self.device)
+
+                # decoder input
+                dec_inp = torch.zeros_like(batch_y[:, -self.args.pred_len:, :]).float()
+                dec_inp = torch.cat([batch_y[:, :self.args.label_len, :], dec_inp], dim=1).float().to(self.device)
+                
+
+                # encoder - decoder
+                if self.args.use_amp:
+                    with torch.cuda.amp.autocast():
+                        if not self.args.use_former:
+                            if any(substr in self.args.model for substr in {'Cycle'}):
+                                outputs = self.model(batch_x, batch_cycle)
+                            else:
+                                outputs = self.model(batch_x)
+                        else:
+                            if self.args.output_attention:
+                                outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)[0]
+                            else:
+                                outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
+                else:
+                    if not self.args.use_former:
+                            if any(substr in self.args.model for substr in {'Cycle'}):
+                                outputs = self.model(batch_x, batch_cycle)
+                            else:
+                                outputs = self.model(batch_x)
+                    else:
+                        if self.args.output_attention:
+                            outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)[0]
+
+                        else:
+                            outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
+
+                f_dim = -1 if self.args.features == 'MS' else 0
+                outputs = outputs[:, -self.args.pred_len:, f_dim:]
+                batch_y = batch_y[:, -self.args.pred_len:, f_dim:].to(self.device)
+                outputs = outputs.detach().cpu().numpy()
+                batch_y = batch_y.detach().cpu().numpy()
+
+                pred = outputs  # outputs.detach().cpu().numpy()  # .squeeze()
+                true = batch_y  # batch_y.detach().cpu().numpy()  # .squeeze()
+
+                preds.append(pred)
+                trues.append(true)
+                inputx.append(batch_x.detach().cpu().numpy())
+               # if i % 20 == 0:
+                   # input = batch_x.detach().cpu().numpy()
+                   # gt = np.concatenate((input[0, :, -1], true[0, :, -1]), axis=0)
+                    #pd = np.concatenate((input[0, :, -1], pred[0, :, -1]), axis=0)
+                   # visual(gt, pd, os.path.join(folder_path, str(i) + '.pdf'), os.path.join(folder_path, str(i) + '.npy'))
+
+        if self.args.test_flop:
+            test_params_flop((batch_x.shape[1],batch_x.shape[2]))
+            exit()
+        preds = np.array(preds)
+        trues = np.array(trues)
+        inputx = np.array(inputx)
+
+        preds = preds.reshape(-1, preds.shape[-2], preds.shape[-1])
+        trues = trues.reshape(-1, trues.shape[-2], trues.shape[-1])
+        inputx = inputx.reshape(-1, inputx.shape[-2], inputx.shape[-1])
+
+        # result save
+        folder_path = './results/' + setting + '/'
+        if not os.path.exists(folder_path):
+            os.makedirs(folder_path)
+
+        mae, mse, rmse, mape, mspe, rse, corr = metric(preds, trues)
+        if self.args.use_PEMSmetric:
+            print('mae:{}, mape:{}, rmse:{}'.format(mae, mape, rmse))
+        else:
+            print('mse:{}, mae:{}, rse:{}'.format(mse, mae, rse))
+
+        f = open("result.txt", 'a')
+        f.write(setting + "  \n")
+        f.write('{} --- Pred {} -> mse:{}, mae:{}, rse:{}'.format(self.args.aug_method, self.args.pred_len, mse, mae, rse))
+        f.write('\n')
+        f.close()
+
+        # np.save(folder_path + 'metrics.npy', np.array([mae, mse, rmse, mape, mspe,rse, corr]))
+        # np.save(folder_path + 'pred.npy', preds)
+        # np.save(folder_path + 'true.npy', trues)
+        # np.save(folder_path + 'x.npy', inputx)
+        return mse, mae, rse
+
+    def predict(self, setting, load=False):
+        pred_data, pred_loader = self._get_data(flag='pred')
+
+        if load:
+            path = os.path.join(self.args.checkpoints, setting)
+            best_model_path = path + '/' + 'checkpoint.pth'
+            self.model.load_state_dict(torch.load(best_model_path))
+
+        preds = []
+
+        self.model.eval()
+        with torch.no_grad():
+            for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(pred_loader):
+                batch_x = batch_x.float().to(self.device)
+                batch_y = batch_y.float()
+                batch_x_mark = batch_x_mark.float().to(self.device)
+                batch_y_mark = batch_y_mark.float().to(self.device)
+
+                # decoder input
+                dec_inp = torch.zeros([batch_y.shape[0], self.args.pred_len, batch_y.shape[2]]).float().to(batch_y.device)
+                dec_inp = torch.cat([batch_y[:, :self.args.label_len, :], dec_inp], dim=1).float().to(self.device)
+
+                # encoder - decoder
+                if self.args.use_amp:
+                    with torch.cuda.amp.autocast():
+                        if not self.args.use_former:
+                            outputs = self.model(batch_x)
+                        else:
+                            if self.args.output_attention:
+                                outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)[0]
+                            else:
+                                outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
+                else:
+                    if not self.args.use_former:
+                        outputs = self.model(batch_x)
+                    else:
+                        if self.args.output_attention:
+                            outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)[0]
+                        else:
+                            outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
+
+                pred = outputs.detach().cpu().numpy()  # .squeeze()
+                preds.append(pred)
+
+        preds = np.array(preds)
+        preds = preds.reshape(-1, preds.shape[-2], preds.shape[-1])
+
+        # result save
+        folder_path = './results/' + setting + '/'
+        if not os.path.exists(folder_path):
+            os.makedirs(folder_path)
+
+        #np.save(folder_path + 'real_prediction.npy', preds)
+
+        return
+

time_series_forecasting/layers/AutoCorrelation.py

@@ -0,0 +1,164 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import matplotlib.pyplot as plt
+import numpy as np
+import math
+from math import sqrt
+import os
+
+
+class AutoCorrelation(nn.Module):
+    """
+    AutoCorrelation Mechanism with the following two phases:
+    (1) period-based dependencies discovery
+    (2) time delay aggregation
+    This block can replace the self-attention family mechanism seamlessly.
+    """
+    def __init__(self, mask_flag=True, factor=1, scale=None, attention_dropout=0.1, output_attention=False):
+        super(AutoCorrelation, self).__init__()
+        self.factor = factor
+        self.scale = scale
+        self.mask_flag = mask_flag
+        self.output_attention = output_attention
+        self.dropout = nn.Dropout(attention_dropout)
+
+    def time_delay_agg_training(self, values, corr):
+        """
+        SpeedUp version of Autocorrelation (a batch-normalization style design)
+        This is for the training phase.
+        """
+        head = values.shape[1]
+        channel = values.shape[2]
+        length = values.shape[3]
+        # find top k
+        top_k = int(self.factor * math.log(length))
+        mean_value = torch.mean(torch.mean(corr, dim=1), dim=1)
+        index = torch.topk(torch.mean(mean_value, dim=0), top_k, dim=-1)[1]
+        weights = torch.stack([mean_value[:, index[i]] for i in range(top_k)], dim=-1)
+        # update corr
+        tmp_corr = torch.softmax(weights, dim=-1)
+        # aggregation
+        tmp_values = values
+        delays_agg = torch.zeros_like(values).float()
+        for i in range(top_k):
+            pattern = torch.roll(tmp_values, -int(index[i]), -1)
+            delays_agg = delays_agg + pattern * \
+                         (tmp_corr[:, i].unsqueeze(1).unsqueeze(1).unsqueeze(1).repeat(1, head, channel, length))
+        return delays_agg
+
+    def time_delay_agg_inference(self, values, corr):
+        """
+        SpeedUp version of Autocorrelation (a batch-normalization style design)
+        This is for the inference phase.
+        """
+        batch = values.shape[0]
+        head = values.shape[1]
+        channel = values.shape[2]
+        length = values.shape[3]
+        # index init
+        init_index = torch.arange(length).unsqueeze(0).unsqueeze(0).unsqueeze(0).repeat(batch, head, channel, 1).cuda()
+        # find top k
+        top_k = int(self.factor * math.log(length))
+        mean_value = torch.mean(torch.mean(corr, dim=1), dim=1)
+        weights = torch.topk(mean_value, top_k, dim=-1)[0]
+        delay = torch.topk(mean_value, top_k, dim=-1)[1]
+        # update corr
+        tmp_corr = torch.softmax(weights, dim=-1)
+        # aggregation
+        tmp_values = values.repeat(1, 1, 1, 2)
+        delays_agg = torch.zeros_like(values).float()
+        for i in range(top_k):
+            tmp_delay = init_index + delay[:, i].unsqueeze(1).unsqueeze(1).unsqueeze(1).repeat(1, head, channel, length)
+            pattern = torch.gather(tmp_values, dim=-1, index=tmp_delay)
+            delays_agg = delays_agg + pattern * \
+                         (tmp_corr[:, i].unsqueeze(1).unsqueeze(1).unsqueeze(1).repeat(1, head, channel, length))
+        return delays_agg
+
+    def time_delay_agg_full(self, values, corr):
+        """
+        Standard version of Autocorrelation
+        """
+        batch = values.shape[0]
+        head = values.shape[1]
+        channel = values.shape[2]
+        length = values.shape[3]
+        # index init
+        init_index = torch.arange(length).unsqueeze(0).unsqueeze(0).unsqueeze(0).repeat(batch, head, channel, 1).cuda()
+        # find top k
+        top_k = int(self.factor * math.log(length))
+        weights = torch.topk(corr, top_k, dim=-1)[0]
+        delay = torch.topk(corr, top_k, dim=-1)[1]
+        # update corr
+        tmp_corr = torch.softmax(weights, dim=-1)
+        # aggregation
+        tmp_values = values.repeat(1, 1, 1, 2)
+        delays_agg = torch.zeros_like(values).float()
+        for i in range(top_k):
+            tmp_delay = init_index + delay[..., i].unsqueeze(-1)
+            pattern = torch.gather(tmp_values, dim=-1, index=tmp_delay)
+            delays_agg = delays_agg + pattern * (tmp_corr[..., i].unsqueeze(-1))
+        return delays_agg
+
+    def forward(self, queries, keys, values, attn_mask):
+        B, L, H, E = queries.shape
+        _, S, _, D = values.shape
+        if L > S:
+            zeros = torch.zeros_like(queries[:, :(L - S), :]).float()
+            values = torch.cat([values, zeros], dim=1)
+            keys = torch.cat([keys, zeros], dim=1)
+        else:
+            values = values[:, :L, :, :]
+            keys = keys[:, :L, :, :]
+
+        # period-based dependencies
+        q_fft = torch.fft.rfft(queries.permute(0, 2, 3, 1).contiguous(), dim=-1)
+        k_fft = torch.fft.rfft(keys.permute(0, 2, 3, 1).contiguous(), dim=-1)
+        res = q_fft * torch.conj(k_fft)
+        corr = torch.fft.irfft(res, dim=-1)
+
+        # time delay agg
+        if self.training:
+            V = self.time_delay_agg_training(values.permute(0, 2, 3, 1).contiguous(), corr).permute(0, 3, 1, 2)
+        else:
+            V = self.time_delay_agg_inference(values.permute(0, 2, 3, 1).contiguous(), corr).permute(0, 3, 1, 2)
+
+        if self.output_attention:
+            return (V.contiguous(), corr.permute(0, 3, 1, 2))
+        else:
+            return (V.contiguous(), None)
+
+
+class AutoCorrelationLayer(nn.Module):
+    def __init__(self, correlation, d_model, n_heads, d_keys=None,
+                 d_values=None):
+        super(AutoCorrelationLayer, self).__init__()
+
+        d_keys = d_keys or (d_model // n_heads)
+        d_values = d_values or (d_model // n_heads)
+
+        self.inner_correlation = correlation
+        self.query_projection = nn.Linear(d_model, d_keys * n_heads)
+        self.key_projection = nn.Linear(d_model, d_keys * n_heads)
+        self.value_projection = nn.Linear(d_model, d_values * n_heads)
+        self.out_projection = nn.Linear(d_values * n_heads, d_model)
+        self.n_heads = n_heads
+
+    def forward(self, queries, keys, values, attn_mask):
+        B, L, _ = queries.shape
+        _, S, _ = keys.shape
+        H = self.n_heads
+
+        queries = self.query_projection(queries).view(B, L, H, -1)
+        keys = self.key_projection(keys).view(B, S, H, -1)
+        values = self.value_projection(values).view(B, S, H, -1)
+
+        out, attn = self.inner_correlation(
+            queries,
+            keys,
+            values,
+            attn_mask
+        )
+        out = out.view(B, L, -1)
+
+        return self.out_projection(out), attn

time_series_forecasting/layers/Autoformer_EncDec.py

@@ -0,0 +1,191 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+class series_decomp_multi(nn.Module):
+    """
+    Series decomposition block
+    """
+    def __init__(self,kernel_size):
+        super(series_decomp_multi, self).__init__()
+        self.moving_avg = [moving_avg(kernel, stride=1) for kernel in kernel_size]
+        self.layer = torch.nn.Linear(1,len(kernel_size))
+
+    def forward(self, x):
+        moving_mean=[]
+        for func in self.moving_avg:
+            moving_avg = func(x)
+            moving_mean.append(moving_avg.unsqueeze(-1))
+        moving_mean=torch.cat(moving_mean,dim=-1)
+        moving_mean = torch.sum(moving_mean*nn.Softmax(-1)(self.layer(x.unsqueeze(-1))),dim=-1)
+        res = x - moving_mean
+        return res, moving_mean 
+        
+class my_Layernorm(nn.Module):
+    """
+    Special designed layernorm for the seasonal part
+    """
+    def __init__(self, channels):
+        super(my_Layernorm, self).__init__()
+        self.layernorm = nn.LayerNorm(channels)
+
+    def forward(self, x):
+        x_hat = self.layernorm(x)
+        bias = torch.mean(x_hat, dim=1).unsqueeze(1).repeat(1, x.shape[1], 1)
+        return x_hat - bias
+
+
+class moving_avg(nn.Module):
+    """
+    Moving average block to highlight the trend of time series
+    """
+    def __init__(self, kernel_size, stride):
+        super(moving_avg, self).__init__()
+        self.kernel_size = kernel_size
+        self.avg = nn.AvgPool1d(kernel_size=kernel_size, stride=stride, padding=0)
+
+    def forward(self, x):
+        # padding on the both ends of time series
+        front = x[:, 0:1, :].repeat(1, (self.kernel_size - 1) // 2, 1)
+        end = x[:, -1:, :].repeat(1, (self.kernel_size - 1) // 2, 1)
+        x = torch.cat([front, x, end], dim=1)
+        x = self.avg(x.permute(0, 2, 1))
+        x = x.permute(0, 2, 1)
+        return x
+
+
+class series_decomp(nn.Module):
+    """
+    Series decomposition block
+    """
+    def __init__(self, kernel_size):
+        super(series_decomp, self).__init__()
+        self.moving_avg = moving_avg(kernel_size, stride=1)
+
+    def forward(self, x):
+        moving_mean = self.moving_avg(x)
+        res = x - moving_mean
+        return res, moving_mean
+
+
+class EncoderLayer(nn.Module):
+    """
+    Autoformer encoder layer with the progressive decomposition architecture
+    """
+    def __init__(self, attention, d_model, d_ff=None, moving_avg=25, dropout=0.1, activation="relu"):
+        super(EncoderLayer, self).__init__()
+        d_ff = d_ff or 4 * d_model
+        self.attention = attention
+        self.conv1 = nn.Conv1d(in_channels=d_model, out_channels=d_ff, kernel_size=1, bias=False)
+        self.conv2 = nn.Conv1d(in_channels=d_ff, out_channels=d_model, kernel_size=1, bias=False)
+        self.decomp1 = series_decomp(moving_avg)
+        self.decomp2 = series_decomp(moving_avg)
+        self.dropout = nn.Dropout(dropout)
+        self.activation = F.relu if activation == "relu" else F.gelu
+
+    def forward(self, x, attn_mask=None):
+        new_x, attn = self.attention(
+            x, x, x,
+            attn_mask=attn_mask
+        )
+        x = x + self.dropout(new_x)
+        x, _ = self.decomp1(x)
+        y = x
+        y = self.dropout(self.activation(self.conv1(y.transpose(-1, 1))))
+        y = self.dropout(self.conv2(y).transpose(-1, 1))
+        res, _ = self.decomp2(x + y)
+        return res, attn
+
+
+class Encoder(nn.Module):
+    """
+    Autoformer encoder
+    """
+    def __init__(self, attn_layers, conv_layers=None, norm_layer=None):
+        super(Encoder, self).__init__()
+        self.attn_layers = nn.ModuleList(attn_layers)
+        self.conv_layers = nn.ModuleList(conv_layers) if conv_layers is not None else None
+        self.norm = norm_layer
+
+    def forward(self, x, attn_mask=None):
+        attns = []
+        if self.conv_layers is not None:
+            for attn_layer, conv_layer in zip(self.attn_layers, self.conv_layers):
+                x, attn = attn_layer(x, attn_mask=attn_mask)
+                x = conv_layer(x)
+                attns.append(attn)
+            x, attn = self.attn_layers[-1](x)
+            attns.append(attn)
+        else:
+            for attn_layer in self.attn_layers:
+                x, attn = attn_layer(x, attn_mask=attn_mask)
+                attns.append(attn)
+
+        if self.norm is not None:
+            x = self.norm(x)
+
+        return x, attns
+
+
+class DecoderLayer(nn.Module):
+    """
+    Autoformer decoder layer with the progressive decomposition architecture
+    """
+    def __init__(self, self_attention, cross_attention, d_model, c_out, d_ff=None,
+                 moving_avg=25, dropout=0.1, activation="relu"):
+        super(DecoderLayer, self).__init__()
+        d_ff = d_ff or 4 * d_model
+        self.self_attention = self_attention
+        self.cross_attention = cross_attention
+        self.conv1 = nn.Conv1d(in_channels=d_model, out_channels=d_ff, kernel_size=1, bias=False)
+        self.conv2 = nn.Conv1d(in_channels=d_ff, out_channels=d_model, kernel_size=1, bias=False)
+        self.decomp1 = series_decomp(moving_avg)
+        self.decomp2 = series_decomp(moving_avg)
+        self.decomp3 = series_decomp(moving_avg)
+        self.dropout = nn.Dropout(dropout)
+        self.projection = nn.Conv1d(in_channels=d_model, out_channels=c_out, kernel_size=3, stride=1, padding=1,
+                                    padding_mode='circular', bias=False)
+        self.activation = F.relu if activation == "relu" else F.gelu
+
+    def forward(self, x, cross, x_mask=None, cross_mask=None):
+        x = x + self.dropout(self.self_attention(
+            x, x, x,
+            attn_mask=x_mask
+        )[0])
+        x, trend1 = self.decomp1(x)
+        x = x + self.dropout(self.cross_attention(
+            x, cross, cross,
+            attn_mask=cross_mask
+        )[0])
+        x, trend2 = self.decomp2(x)
+        y = x
+        y = self.dropout(self.activation(self.conv1(y.transpose(-1, 1))))
+        y = self.dropout(self.conv2(y).transpose(-1, 1))
+        x, trend3 = self.decomp3(x + y)
+
+        residual_trend = trend1 + trend2 + trend3
+        residual_trend = self.projection(residual_trend.permute(0, 2, 1)).transpose(1, 2)
+        return x, residual_trend
+
+
+class Decoder(nn.Module):
+    """
+    Autoformer encoder
+    """
+    def __init__(self, layers, norm_layer=None, projection=None):
+        super(Decoder, self).__init__()
+        self.layers = nn.ModuleList(layers)
+        self.norm = norm_layer
+        self.projection = projection
+
+    def forward(self, x, cross, x_mask=None, cross_mask=None, trend=None):
+        for layer in self.layers:
+            x, residual_trend = layer(x, cross, x_mask=x_mask, cross_mask=cross_mask)
+            trend = trend + residual_trend
+
+        if self.norm is not None:
+            x = self.norm(x)
+
+        if self.projection is not None:
+            x = self.projection(x)
+        return x, trend

time_series_forecasting/layers/Conv_Blocks.py

@@ -0,0 +1,60 @@
+import torch
+import torch.nn as nn
+
+
+class Inception_Block_V1(nn.Module):
+    def __init__(self, in_channels, out_channels, num_kernels=6, init_weight=True):
+        super(Inception_Block_V1, self).__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.num_kernels = num_kernels
+        kernels = []
+        for i in range(self.num_kernels):
+            kernels.append(nn.Conv2d(in_channels, out_channels, kernel_size=2 * i + 1, padding=i))
+        self.kernels = nn.ModuleList(kernels)
+        if init_weight:
+            self._initialize_weights()
+
+    def _initialize_weights(self):
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+                if m.bias is not None:
+                    nn.init.constant_(m.bias, 0)
+
+    def forward(self, x):
+        res_list = []
+        for i in range(self.num_kernels):
+            res_list.append(self.kernels[i](x))
+        res = torch.stack(res_list, dim=-1).mean(-1)
+        return res
+
+
+class Inception_Block_V2(nn.Module):
+    def __init__(self, in_channels, out_channels, num_kernels=6, init_weight=True):
+        super(Inception_Block_V2, self).__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.num_kernels = num_kernels
+        kernels = []
+        for i in range(self.num_kernels // 2):
+            kernels.append(nn.Conv2d(in_channels, out_channels, kernel_size=[1, 2 * i + 3], padding=[0, i + 1]))
+            kernels.append(nn.Conv2d(in_channels, out_channels, kernel_size=[2 * i + 3, 1], padding=[i + 1, 0]))
+        kernels.append(nn.Conv2d(in_channels, out_channels, kernel_size=1))
+        self.kernels = nn.ModuleList(kernels)
+        if init_weight:
+            self._initialize_weights()
+
+    def _initialize_weights(self):
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+                if m.bias is not None:
+                    nn.init.constant_(m.bias, 0)
+
+    def forward(self, x):
+        res_list = []
+        for i in range(self.num_kernels + 1):
+            res_list.append(self.kernels[i](x))
+        res = torch.stack(res_list, dim=-1).mean(-1)
+        return res

time_series_forecasting/layers/Embed.py

@@ -0,0 +1,217 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from torch.nn.utils import weight_norm
+import math
+
+
+class PositionalEmbedding(nn.Module):
+    def __init__(self, d_model, max_len=5000):
+        super(PositionalEmbedding, self).__init__()
+        # Compute the positional encodings once in log space.
+        pe = torch.zeros(max_len, d_model).float()
+        pe.require_grad = False
+
+        position = torch.arange(0, max_len).float().unsqueeze(1)
+        div_term = (torch.arange(0, d_model, 2).float() * -(math.log(10000.0) / d_model)).exp()
+
+        pe[:, 0::2] = torch.sin(position * div_term)
+        pe[:, 1::2] = torch.cos(position * div_term)
+
+        pe = pe.unsqueeze(0)
+        self.register_buffer('pe', pe)
+
+    def forward(self, x):
+        return self.pe[:, :x.size(1)]
+
+
+class TokenEmbedding(nn.Module):
+    def __init__(self, c_in, d_model):
+        super(TokenEmbedding, self).__init__()
+        padding = 1 if torch.__version__ >= '1.5.0' else 2
+        self.tokenConv = nn.Conv1d(in_channels=c_in, out_channels=d_model,
+                                   kernel_size=3, padding=padding, padding_mode='circular', bias=False)
+        for m in self.modules():
+            if isinstance(m, nn.Conv1d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_in', nonlinearity='leaky_relu')
+
+    def forward(self, x):
+        x = self.tokenConv(x.permute(0, 2, 1)).transpose(1, 2)
+        return x
+
+
+class FixedEmbedding(nn.Module):
+    def __init__(self, c_in, d_model):
+        super(FixedEmbedding, self).__init__()
+
+        w = torch.zeros(c_in, d_model).float()
+        w.require_grad = False
+
+        position = torch.arange(0, c_in).float().unsqueeze(1)
+        div_term = (torch.arange(0, d_model, 2).float() * -(math.log(10000.0) / d_model)).exp()
+
+        w[:, 0::2] = torch.sin(position * div_term)
+        w[:, 1::2] = torch.cos(position * div_term)
+
+        self.emb = nn.Embedding(c_in, d_model)
+        self.emb.weight = nn.Parameter(w, requires_grad=False)
+
+    def forward(self, x):
+        return self.emb(x).detach()
+
+
+class TemporalEmbedding(nn.Module):
+    def __init__(self, d_model, embed_type='fixed', freq='h'):
+        super(TemporalEmbedding, self).__init__()
+
+        minute_size = 4
+        hour_size = 24
+        weekday_size = 7
+        day_size = 32
+        month_size = 13
+
+        Embed = FixedEmbedding if embed_type == 'fixed' else nn.Embedding
+        if freq == 't':
+            self.minute_embed = Embed(minute_size, d_model)
+        self.hour_embed = Embed(hour_size, d_model)
+        self.weekday_embed = Embed(weekday_size, d_model)
+        self.day_embed = Embed(day_size, d_model)
+        self.month_embed = Embed(month_size, d_model)
+
+    def forward(self, x):
+        x = x.long()
+
+        minute_x = self.minute_embed(x[:, :, 4]) if hasattr(self, 'minute_embed') else 0.
+        hour_x = self.hour_embed(x[:, :, 3])
+        weekday_x = self.weekday_embed(x[:, :, 2])
+        day_x = self.day_embed(x[:, :, 1])
+        month_x = self.month_embed(x[:, :, 0])
+
+        return hour_x + weekday_x + day_x + month_x + minute_x
+
+
+class TimeFeatureEmbedding(nn.Module):
+    def __init__(self, d_model, embed_type='timeF', freq='h'):
+        super(TimeFeatureEmbedding, self).__init__()
+
+        freq_map = {'h': 4, 't': 5, 's': 6, 'm': 1, 'a': 1, 'w': 2, 'd': 3, 'b': 3}
+        d_inp = freq_map[freq]
+        self.embed = nn.Linear(d_inp, d_model, bias=False)
+
+    def forward(self, x):
+        return self.embed(x)
+
+
+class DataEmbedding(nn.Module):
+    def __init__(self, c_in, d_model, embed_type='fixed', freq='h', dropout=0.1):
+        super(DataEmbedding, self).__init__()
+
+        self.value_embedding = TokenEmbedding(c_in=c_in, d_model=d_model)
+        self.position_embedding = PositionalEmbedding(d_model=d_model)
+        self.temporal_embedding = TemporalEmbedding(d_model=d_model, embed_type=embed_type,
+                                                    freq=freq) if embed_type != 'timeF' else TimeFeatureEmbedding(
+            d_model=d_model, embed_type=embed_type, freq=freq)
+        self.dropout = nn.Dropout(p=dropout)
+
+    def forward(self, x, x_mark):
+        if x_mark is None:
+            x = self.value_embedding(x) + self.position_embedding(x)
+        else:
+            x = self.value_embedding(
+                x) + self.temporal_embedding(x_mark) + self.position_embedding(x)
+        return self.dropout(x)
+
+
+class DataEmbedding_wo_pos(nn.Module):
+    def __init__(self, c_in, d_model, embed_type='fixed', freq='h', dropout=0.1):
+        super(DataEmbedding_wo_pos, self).__init__()
+
+        self.value_embedding = TokenEmbedding(c_in=c_in, d_model=d_model)
+        self.position_embedding = PositionalEmbedding(d_model=d_model)
+        self.temporal_embedding = TemporalEmbedding(d_model=d_model, embed_type=embed_type,
+                                                    freq=freq) if embed_type != 'timeF' else TimeFeatureEmbedding(
+            d_model=d_model, embed_type=embed_type, freq=freq)
+        self.dropout = nn.Dropout(p=dropout)
+
+    def forward(self, x, x_mark):
+        #x = self.value_embedding(x) + self.temporal_embedding(x_mark)
+        if x_mark is None:
+            x = self.value_embedding(x)
+        else:
+            x = self.value_embedding(x) + self.temporal_embedding(x_mark)
+        return self.dropout(x)
+
+class DataEmbedding_wo_pos_temp(nn.Module):
+    def __init__(self, c_in, d_model, embed_type='fixed', freq='h', dropout=0.1):
+        super(DataEmbedding_wo_pos_temp, self).__init__()
+
+        self.value_embedding = TokenEmbedding(c_in=c_in, d_model=d_model)
+        self.position_embedding = PositionalEmbedding(d_model=d_model)
+        self.temporal_embedding = TemporalEmbedding(d_model=d_model, embed_type=embed_type,
+                                                    freq=freq) if embed_type != 'timeF' else TimeFeatureEmbedding(
+            d_model=d_model, embed_type=embed_type, freq=freq)
+        self.dropout = nn.Dropout(p=dropout)
+
+    def forward(self, x, x_mark):
+        x = self.value_embedding(x)
+        return self.dropout(x)
+
+class DataEmbedding_wo_temp(nn.Module):
+    def __init__(self, c_in, d_model, embed_type='fixed', freq='h', dropout=0.1):
+        super(DataEmbedding_wo_temp, self).__init__()
+
+        self.value_embedding = TokenEmbedding(c_in=c_in, d_model=d_model)
+        self.position_embedding = PositionalEmbedding(d_model=d_model)
+        self.temporal_embedding = TemporalEmbedding(d_model=d_model, embed_type=embed_type,
+                                                    freq=freq) if embed_type != 'timeF' else TimeFeatureEmbedding(
+            d_model=d_model, embed_type=embed_type, freq=freq)
+        self.dropout = nn.Dropout(p=dropout)
+
+    def forward(self, x, x_mark):
+        x = self.value_embedding(x) + self.position_embedding(x)
+        return self.dropout(x)
+
+
+class PatchEmbedding(nn.Module):
+    def __init__(self, d_model, patch_len, stride, padding, dropout):
+        super(PatchEmbedding, self).__init__()
+        # Patching
+        self.patch_len = patch_len
+        self.stride = stride
+        self.padding_patch_layer = nn.ReplicationPad1d((0, padding))
+
+        # Backbone, Input encoding: projection of feature vectors onto a d-dim vector space
+        self.value_embedding = nn.Linear(patch_len, d_model, bias=False)
+
+        # Positional embedding
+        self.position_embedding = PositionalEmbedding(d_model)
+
+        # Residual dropout
+        self.dropout = nn.Dropout(dropout)
+
+    def forward(self, x):
+        # do patching
+        n_vars = x.shape[1]
+        x = self.padding_patch_layer(x)
+        x = x.unfold(dimension=-1, size=self.patch_len, step=self.stride)
+        x = torch.reshape(x, (x.shape[0] * x.shape[1], x.shape[2], x.shape[3]))
+        # Input encoding
+        x = self.value_embedding(x) + self.position_embedding(x)
+        return self.dropout(x), n_vars
+
+
+class DataEmbedding_inverted(nn.Module):
+    def __init__(self, c_in, d_model, embed_type='fixed', freq='h', dropout=0.1):
+        super(DataEmbedding_inverted, self).__init__()
+        self.value_embedding = nn.Linear(c_in, d_model)
+        self.dropout = nn.Dropout(p=dropout)
+
+    def forward(self, x, x_mark):
+        x = x.permute(0, 2, 1)
+        # x: [Batch Variate Time]
+        if x_mark is None:
+            x = self.value_embedding(x)
+        else:
+            x = self.value_embedding(torch.cat([x, x_mark.permute(0, 2, 1)], 1))
+        # x: [Batch Variate d_model]
+        return self.dropout(x)

time_series_forecasting/layers/PatchTST_backbone.py

@@ -0,0 +1,379 @@
+__all__ = ['PatchTST_backbone']
+
+# Cell
+from typing import Callable, Optional
+import torch
+from torch import nn
+from torch import Tensor
+import torch.nn.functional as F
+import numpy as np
+
+#from collections import OrderedDict
+from layers.PatchTST_layers import *
+from layers.RevIN import RevIN
+
+# Cell
+class PatchTST_backbone(nn.Module):
+    def __init__(self, c_in:int, context_window:int, target_window:int, patch_len:int, stride:int, max_seq_len:Optional[int]=1024, 
+                 n_layers:int=3, d_model=128, n_heads=16, d_k:Optional[int]=None, d_v:Optional[int]=None,
+                 d_ff:int=256, norm:str='BatchNorm', attn_dropout:float=0., dropout:float=0., act:str="gelu", key_padding_mask:bool='auto',
+                 padding_var:Optional[int]=None, attn_mask:Optional[Tensor]=None, res_attention:bool=True, pre_norm:bool=False, store_attn:bool=False,
+                 pe:str='zeros', learn_pe:bool=True, fc_dropout:float=0., head_dropout = 0, padding_patch = None,
+                 pretrain_head:bool=False, head_type = 'flatten', individual = False, revin = True, affine = True, subtract_last = False,
+                 verbose:bool=False, **kwargs):
+        
+        super().__init__()
+        
+        # RevIn
+        self.revin = revin
+        if self.revin: self.revin_layer = RevIN(c_in, affine=affine, subtract_last=subtract_last)
+        
+        # Patching
+        self.patch_len = patch_len
+        self.stride = stride
+        self.padding_patch = padding_patch
+        patch_num = int((context_window - patch_len)/stride + 1)
+        if padding_patch == 'end': # can be modified to general case
+            self.padding_patch_layer = nn.ReplicationPad1d((0, stride)) 
+            patch_num += 1
+        
+        # Backbone 
+        self.backbone = TSTiEncoder(c_in, patch_num=patch_num, patch_len=patch_len, max_seq_len=max_seq_len,
+                                n_layers=n_layers, d_model=d_model, n_heads=n_heads, d_k=d_k, d_v=d_v, d_ff=d_ff,
+                                attn_dropout=attn_dropout, dropout=dropout, act=act, key_padding_mask=key_padding_mask, padding_var=padding_var,
+                                attn_mask=attn_mask, res_attention=res_attention, pre_norm=pre_norm, store_attn=store_attn,
+                                pe=pe, learn_pe=learn_pe, verbose=verbose, **kwargs)
+
+        # Head
+        self.head_nf = d_model * patch_num
+        self.n_vars = c_in
+        self.pretrain_head = pretrain_head
+        self.head_type = head_type
+        self.individual = individual
+
+        if self.pretrain_head: 
+            self.head = self.create_pretrain_head(self.head_nf, c_in, fc_dropout) # custom head passed as a partial func with all its kwargs
+        elif head_type == 'flatten': 
+            self.head = Flatten_Head(self.individual, self.n_vars, self.head_nf, target_window, head_dropout=head_dropout)
+        
+    
+    def forward(self, z):                                                                   # z: [bs x nvars x seq_len]
+        # norm
+        if self.revin: 
+            z = z.permute(0,2,1)
+            z = self.revin_layer(z, 'norm')
+            z = z.permute(0,2,1)
+            
+        # do patching
+        if self.padding_patch == 'end':
+            z = self.padding_patch_layer(z)
+        z = z.unfold(dimension=-1, size=self.patch_len, step=self.stride)                   # z: [bs x nvars x patch_num x patch_len]
+        z = z.permute(0,1,3,2)                                                              # z: [bs x nvars x patch_len x patch_num]
+        
+        # model
+        z = self.backbone(z)                                                                # z: [bs x nvars x d_model x patch_num]
+        z = self.head(z)                                                                    # z: [bs x nvars x target_window] 
+        
+        # denorm
+        if self.revin: 
+            z = z.permute(0,2,1)
+            z = self.revin_layer(z, 'denorm')
+            z = z.permute(0,2,1)
+        return z
+    
+    def create_pretrain_head(self, head_nf, vars, dropout):
+        return nn.Sequential(nn.Dropout(dropout),
+                    nn.Conv1d(head_nf, vars, 1)
+                    )
+
+
+class Flatten_Head(nn.Module):
+    def __init__(self, individual, n_vars, nf, target_window, head_dropout=0):
+        super().__init__()
+        
+        self.individual = individual
+        self.n_vars = n_vars
+        
+        if self.individual:
+            self.linears = nn.ModuleList()
+            self.dropouts = nn.ModuleList()
+            self.flattens = nn.ModuleList()
+            for i in range(self.n_vars):
+                self.flattens.append(nn.Flatten(start_dim=-2))
+                self.linears.append(nn.Linear(nf, target_window))
+                self.dropouts.append(nn.Dropout(head_dropout))
+        else:
+            self.flatten = nn.Flatten(start_dim=-2)
+            self.linear = nn.Linear(nf, target_window)
+            self.dropout = nn.Dropout(head_dropout)
+            
+    def forward(self, x):                                 # x: [bs x nvars x d_model x patch_num]
+        if self.individual:
+            x_out = []
+            for i in range(self.n_vars):
+                z = self.flattens[i](x[:,i,:,:])          # z: [bs x d_model * patch_num]
+                z = self.linears[i](z)                    # z: [bs x target_window]
+                z = self.dropouts[i](z)
+                x_out.append(z)
+            x = torch.stack(x_out, dim=1)                 # x: [bs x nvars x target_window]
+        else:
+            x = self.flatten(x)
+            x = self.linear(x)
+            x = self.dropout(x)
+        return x
+        
+        
+    
+    
+class TSTiEncoder(nn.Module):  #i means channel-independent
+    def __init__(self, c_in, patch_num, patch_len, max_seq_len=1024,
+                 n_layers=3, d_model=128, n_heads=16, d_k=None, d_v=None,
+                 d_ff=256, norm='BatchNorm', attn_dropout=0., dropout=0., act="gelu", store_attn=False,
+                 key_padding_mask='auto', padding_var=None, attn_mask=None, res_attention=True, pre_norm=False,
+                 pe='zeros', learn_pe=True, verbose=False, **kwargs):
+        
+        
+        super().__init__()
+        
+        self.patch_num = patch_num
+        self.patch_len = patch_len
+        
+        # Input encoding
+        q_len = patch_num
+        self.W_P = nn.Linear(patch_len, d_model)        # Eq 1: projection of feature vectors onto a d-dim vector space
+        self.seq_len = q_len
+
+        # Positional encoding
+        self.W_pos = positional_encoding(pe, learn_pe, q_len, d_model)
+
+        # Residual dropout
+        self.dropout = nn.Dropout(dropout)
+
+        # Encoder
+        self.encoder = TSTEncoder(q_len, d_model, n_heads, d_k=d_k, d_v=d_v, d_ff=d_ff, norm=norm, attn_dropout=attn_dropout, dropout=dropout,
+                                   pre_norm=pre_norm, activation=act, res_attention=res_attention, n_layers=n_layers, store_attn=store_attn)
+
+        
+    def forward(self, x) -> Tensor:                                              # x: [bs x nvars x patch_len x patch_num]
+        
+        n_vars = x.shape[1]
+        # Input encoding
+        x = x.permute(0,1,3,2)                                                   # x: [bs x nvars x patch_num x patch_len]
+        x = self.W_P(x)                                                          # x: [bs x nvars x patch_num x d_model]
+
+        u = torch.reshape(x, (x.shape[0]*x.shape[1],x.shape[2],x.shape[3]))      # u: [bs * nvars x patch_num x d_model]
+        u = self.dropout(u + self.W_pos)                                         # u: [bs * nvars x patch_num x d_model]
+
+        # Encoder
+        z = self.encoder(u)                                                      # z: [bs * nvars x patch_num x d_model]
+        z = torch.reshape(z, (-1,n_vars,z.shape[-2],z.shape[-1]))                # z: [bs x nvars x patch_num x d_model]
+        z = z.permute(0,1,3,2)                                                   # z: [bs x nvars x d_model x patch_num]
+        
+        return z    
+            
+            
+    
+# Cell
+class TSTEncoder(nn.Module):
+    def __init__(self, q_len, d_model, n_heads, d_k=None, d_v=None, d_ff=None, 
+                        norm='BatchNorm', attn_dropout=0., dropout=0., activation='gelu',
+                        res_attention=False, n_layers=1, pre_norm=False, store_attn=False):
+        super().__init__()
+
+        self.layers = nn.ModuleList([TSTEncoderLayer(q_len, d_model, n_heads=n_heads, d_k=d_k, d_v=d_v, d_ff=d_ff, norm=norm,
+                                                      attn_dropout=attn_dropout, dropout=dropout,
+                                                      activation=activation, res_attention=res_attention,
+                                                      pre_norm=pre_norm, store_attn=store_attn) for i in range(n_layers)])
+        self.res_attention = res_attention
+
+    def forward(self, src:Tensor, key_padding_mask:Optional[Tensor]=None, attn_mask:Optional[Tensor]=None):
+        output = src
+        scores = None
+        if self.res_attention:
+            for mod in self.layers: output, scores = mod(output, prev=scores, key_padding_mask=key_padding_mask, attn_mask=attn_mask)
+            return output
+        else:
+            for mod in self.layers: output = mod(output, key_padding_mask=key_padding_mask, attn_mask=attn_mask)
+            return output
+
+
+
+class TSTEncoderLayer(nn.Module):
+    def __init__(self, q_len, d_model, n_heads, d_k=None, d_v=None, d_ff=256, store_attn=False,
+                 norm='BatchNorm', attn_dropout=0, dropout=0., bias=True, activation="gelu", res_attention=False, pre_norm=False):
+        super().__init__()
+        assert not d_model%n_heads, f"d_model ({d_model}) must be divisible by n_heads ({n_heads})"
+        d_k = d_model // n_heads if d_k is None else d_k
+        d_v = d_model // n_heads if d_v is None else d_v
+
+        # Multi-Head attention
+        self.res_attention = res_attention
+        self.self_attn = _MultiheadAttention(d_model, n_heads, d_k, d_v, attn_dropout=attn_dropout, proj_dropout=dropout, res_attention=res_attention)
+
+        # Add & Norm
+        self.dropout_attn = nn.Dropout(dropout)
+        if "batch" in norm.lower():
+            self.norm_attn = nn.Sequential(Transpose(1,2), nn.BatchNorm1d(d_model), Transpose(1,2))
+        else:
+            self.norm_attn = nn.LayerNorm(d_model)
+
+        # Position-wise Feed-Forward
+        self.ff = nn.Sequential(nn.Linear(d_model, d_ff, bias=bias),
+                                get_activation_fn(activation),
+                                nn.Dropout(dropout),
+                                nn.Linear(d_ff, d_model, bias=bias))
+
+        # Add & Norm
+        self.dropout_ffn = nn.Dropout(dropout)
+        if "batch" in norm.lower():
+            self.norm_ffn = nn.Sequential(Transpose(1,2), nn.BatchNorm1d(d_model), Transpose(1,2))
+        else:
+            self.norm_ffn = nn.LayerNorm(d_model)
+
+        self.pre_norm = pre_norm
+        self.store_attn = store_attn
+
+
+    def forward(self, src:Tensor, prev:Optional[Tensor]=None, key_padding_mask:Optional[Tensor]=None, attn_mask:Optional[Tensor]=None) -> Tensor:
+
+        # Multi-Head attention sublayer
+        if self.pre_norm:
+            src = self.norm_attn(src)
+        ## Multi-Head attention
+        if self.res_attention:
+            src2, attn, scores = self.self_attn(src, src, src, prev, key_padding_mask=key_padding_mask, attn_mask=attn_mask)
+        else:
+            src2, attn = self.self_attn(src, src, src, key_padding_mask=key_padding_mask, attn_mask=attn_mask)
+        if self.store_attn:
+            self.attn = attn
+        ## Add & Norm
+        src = src + self.dropout_attn(src2) # Add: residual connection with residual dropout
+        if not self.pre_norm:
+            src = self.norm_attn(src)
+
+        # Feed-forward sublayer
+        if self.pre_norm:
+            src = self.norm_ffn(src)
+        ## Position-wise Feed-Forward
+        src2 = self.ff(src)
+        ## Add & Norm
+        src = src + self.dropout_ffn(src2) # Add: residual connection with residual dropout
+        if not self.pre_norm:
+            src = self.norm_ffn(src)
+
+        if self.res_attention:
+            return src, scores
+        else:
+            return src
+
+
+
+
+class _MultiheadAttention(nn.Module):
+    def __init__(self, d_model, n_heads, d_k=None, d_v=None, res_attention=False, attn_dropout=0., proj_dropout=0., qkv_bias=True, lsa=False):
+        """Multi Head Attention Layer
+        Input shape:
+            Q:       [batch_size (bs) x max_q_len x d_model]
+            K, V:    [batch_size (bs) x q_len x d_model]
+            mask:    [q_len x q_len]
+        """
+        super().__init__()
+        d_k = d_model // n_heads if d_k is None else d_k
+        d_v = d_model // n_heads if d_v is None else d_v
+
+        self.n_heads, self.d_k, self.d_v = n_heads, d_k, d_v
+
+        self.W_Q = nn.Linear(d_model, d_k * n_heads, bias=qkv_bias)
+        self.W_K = nn.Linear(d_model, d_k * n_heads, bias=qkv_bias)
+        self.W_V = nn.Linear(d_model, d_v * n_heads, bias=qkv_bias)
+
+        # Scaled Dot-Product Attention (multiple heads)
+        self.res_attention = res_attention
+        self.sdp_attn = _ScaledDotProductAttention(d_model, n_heads, attn_dropout=attn_dropout, res_attention=self.res_attention, lsa=lsa)
+
+        # Poject output
+        self.to_out = nn.Sequential(nn.Linear(n_heads * d_v, d_model), nn.Dropout(proj_dropout))
+
+
+    def forward(self, Q:Tensor, K:Optional[Tensor]=None, V:Optional[Tensor]=None, prev:Optional[Tensor]=None,
+                key_padding_mask:Optional[Tensor]=None, attn_mask:Optional[Tensor]=None):
+
+        bs = Q.size(0)
+        if K is None: K = Q
+        if V is None: V = Q
+
+        # Linear (+ split in multiple heads)
+        q_s = self.W_Q(Q).view(bs, -1, self.n_heads, self.d_k).transpose(1,2)       # q_s    : [bs x n_heads x max_q_len x d_k]
+        k_s = self.W_K(K).view(bs, -1, self.n_heads, self.d_k).permute(0,2,3,1)     # k_s    : [bs x n_heads x d_k x q_len] - transpose(1,2) + transpose(2,3)
+        v_s = self.W_V(V).view(bs, -1, self.n_heads, self.d_v).transpose(1,2)       # v_s    : [bs x n_heads x q_len x d_v]
+
+        # Apply Scaled Dot-Product Attention (multiple heads)
+        if self.res_attention:
+            output, attn_weights, attn_scores = self.sdp_attn(q_s, k_s, v_s, prev=prev, key_padding_mask=key_padding_mask, attn_mask=attn_mask)
+        else:
+            output, attn_weights = self.sdp_attn(q_s, k_s, v_s, key_padding_mask=key_padding_mask, attn_mask=attn_mask)
+        # output: [bs x n_heads x q_len x d_v], attn: [bs x n_heads x q_len x q_len], scores: [bs x n_heads x max_q_len x q_len]
+
+        # back to the original inputs dimensions
+        output = output.transpose(1, 2).contiguous().view(bs, -1, self.n_heads * self.d_v) # output: [bs x q_len x n_heads * d_v]
+        output = self.to_out(output)
+
+        if self.res_attention: return output, attn_weights, attn_scores
+        else: return output, attn_weights
+
+
+class _ScaledDotProductAttention(nn.Module):
+    r"""Scaled Dot-Product Attention module (Attention is all you need by Vaswani et al., 2017) with optional residual attention from previous layer
+    (Realformer: Transformer likes residual attention by He et al, 2020) and locality self sttention (Vision Transformer for Small-Size Datasets
+    by Lee et al, 2021)"""
+
+    def __init__(self, d_model, n_heads, attn_dropout=0., res_attention=False, lsa=False):
+        super().__init__()
+        self.attn_dropout = nn.Dropout(attn_dropout)
+        self.res_attention = res_attention
+        head_dim = d_model // n_heads
+        self.scale = nn.Parameter(torch.tensor(head_dim ** -0.5), requires_grad=lsa)
+        self.lsa = lsa
+
+    def forward(self, q:Tensor, k:Tensor, v:Tensor, prev:Optional[Tensor]=None, key_padding_mask:Optional[Tensor]=None, attn_mask:Optional[Tensor]=None):
+        '''
+        Input shape:
+            q               : [bs x n_heads x max_q_len x d_k]
+            k               : [bs x n_heads x d_k x seq_len]
+            v               : [bs x n_heads x seq_len x d_v]
+            prev            : [bs x n_heads x q_len x seq_len]
+            key_padding_mask: [bs x seq_len]
+            attn_mask       : [1 x seq_len x seq_len]
+        Output shape:
+            output:  [bs x n_heads x q_len x d_v]
+            attn   : [bs x n_heads x q_len x seq_len]
+            scores : [bs x n_heads x q_len x seq_len]
+        '''
+
+        # Scaled MatMul (q, k) - similarity scores for all pairs of positions in an input sequence
+        attn_scores = torch.matmul(q, k) * self.scale      # attn_scores : [bs x n_heads x max_q_len x q_len]
+
+        # Add pre-softmax attention scores from the previous layer (optional)
+        if prev is not None: attn_scores = attn_scores + prev
+
+        # Attention mask (optional)
+        if attn_mask is not None:                                     # attn_mask with shape [q_len x seq_len] - only used when q_len == seq_len
+            if attn_mask.dtype == torch.bool:
+                attn_scores.masked_fill_(attn_mask, -np.inf)
+            else:
+                attn_scores += attn_mask
+
+        # Key padding mask (optional)
+        if key_padding_mask is not None:                              # mask with shape [bs x q_len] (only when max_w_len == q_len)
+            attn_scores.masked_fill_(key_padding_mask.unsqueeze(1).unsqueeze(2), -np.inf)
+
+        # normalize the attention weights
+        attn_weights = F.softmax(attn_scores, dim=-1)                 # attn_weights   : [bs x n_heads x max_q_len x q_len]
+        attn_weights = self.attn_dropout(attn_weights)
+
+        # compute the new values given the attention weights
+        output = torch.matmul(attn_weights, v)                        # output: [bs x n_heads x max_q_len x d_v]
+
+        if self.res_attention: return output, attn_weights, attn_scores
+        else: return output, attn_weights
+

time_series_forecasting/layers/PatchTST_layers.py

@@ -0,0 +1,121 @@
+__all__ = ['Transpose', 'get_activation_fn', 'moving_avg', 'series_decomp', 'PositionalEncoding', 'SinCosPosEncoding', 'Coord2dPosEncoding', 'Coord1dPosEncoding', 'positional_encoding']           
+
+import torch
+from torch import nn
+import math
+
+class Transpose(nn.Module):
+    def __init__(self, *dims, contiguous=False): 
+        super().__init__()
+        self.dims, self.contiguous = dims, contiguous
+    def forward(self, x):
+        if self.contiguous: return x.transpose(*self.dims).contiguous()
+        else: return x.transpose(*self.dims)
+
+    
+def get_activation_fn(activation):
+    if callable(activation): return activation()
+    elif activation.lower() == "relu": return nn.ReLU()
+    elif activation.lower() == "gelu": return nn.GELU()
+    raise ValueError(f'{activation} is not available. You can use "relu", "gelu", or a callable') 
+    
+    
+# decomposition
+
+class moving_avg(nn.Module):
+    """
+    Moving average block to highlight the trend of time series
+    """
+    def __init__(self, kernel_size, stride):
+        super(moving_avg, self).__init__()
+        self.kernel_size = kernel_size
+        self.avg = nn.AvgPool1d(kernel_size=kernel_size, stride=stride, padding=0)
+
+    def forward(self, x):
+        # padding on the both ends of time series
+        front = x[:, 0:1, :].repeat(1, (self.kernel_size - 1) // 2, 1)
+        end = x[:, -1:, :].repeat(1, (self.kernel_size - 1) // 2, 1)
+        x = torch.cat([front, x, end], dim=1)
+        x = self.avg(x.permute(0, 2, 1))
+        x = x.permute(0, 2, 1)
+        return x
+
+
+class series_decomp(nn.Module):
+    """
+    Series decomposition block
+    """
+    def __init__(self, kernel_size):
+        super(series_decomp, self).__init__()
+        self.moving_avg = moving_avg(kernel_size, stride=1)
+
+    def forward(self, x):
+        moving_mean = self.moving_avg(x)
+        res = x - moving_mean
+        return res, moving_mean
+    
+    
+    
+# pos_encoding
+
+def PositionalEncoding(q_len, d_model, normalize=True):
+    pe = torch.zeros(q_len, d_model)
+    position = torch.arange(0, q_len).unsqueeze(1)
+    div_term = torch.exp(torch.arange(0, d_model, 2) * -(math.log(10000.0) / d_model))
+    pe[:, 0::2] = torch.sin(position * div_term)
+    pe[:, 1::2] = torch.cos(position * div_term)
+    if normalize:
+        pe = pe - pe.mean()
+        pe = pe / (pe.std() * 10)
+    return pe
+
+SinCosPosEncoding = PositionalEncoding
+
+def Coord2dPosEncoding(q_len, d_model, exponential=False, normalize=True, eps=1e-3, verbose=False):
+    x = .5 if exponential else 1
+    i = 0
+    for i in range(100):
+        cpe = 2 * (torch.linspace(0, 1, q_len).reshape(-1, 1) ** x) * (torch.linspace(0, 1, d_model).reshape(1, -1) ** x) - 1
+        pv(f'{i:4.0f}  {x:5.3f}  {cpe.mean():+6.3f}', verbose)
+        if abs(cpe.mean()) <= eps: break
+        elif cpe.mean() > eps: x += .001
+        else: x -= .001
+        i += 1
+    if normalize:
+        cpe = cpe - cpe.mean()
+        cpe = cpe / (cpe.std() * 10)
+    return cpe
+
+def Coord1dPosEncoding(q_len, exponential=False, normalize=True):
+    cpe = (2 * (torch.linspace(0, 1, q_len).reshape(-1, 1)**(.5 if exponential else 1)) - 1)
+    if normalize:
+        cpe = cpe - cpe.mean()
+        cpe = cpe / (cpe.std() * 10)
+    return cpe
+
+def positional_encoding(pe, learn_pe, q_len, d_model):
+    # Positional encoding
+    if pe == None:
+        W_pos = torch.empty((q_len, d_model)) # pe = None and learn_pe = False can be used to measure impact of pe
+        nn.init.uniform_(W_pos, -0.02, 0.02)
+        learn_pe = False
+    elif pe == 'zero':
+        W_pos = torch.empty((q_len, 1))
+        nn.init.uniform_(W_pos, -0.02, 0.02)
+    elif pe == 'zeros':
+        W_pos = torch.empty((q_len, d_model))
+        nn.init.uniform_(W_pos, -0.02, 0.02)
+    elif pe == 'normal' or pe == 'gauss':
+        W_pos = torch.zeros((q_len, 1))
+        torch.nn.init.normal_(W_pos, mean=0.0, std=0.1)
+    elif pe == 'uniform':
+        W_pos = torch.zeros((q_len, 1))
+        nn.init.uniform_(W_pos, a=0.0, b=0.1)
+    elif pe == 'lin1d': W_pos = Coord1dPosEncoding(q_len, exponential=False, normalize=True)
+    elif pe == 'exp1d': W_pos = Coord1dPosEncoding(q_len, exponential=True, normalize=True)
+    elif pe == 'lin2d': W_pos = Coord2dPosEncoding(q_len, d_model, exponential=False, normalize=True)
+    elif pe == 'exp2d': W_pos = Coord2dPosEncoding(q_len, d_model, exponential=True, normalize=True)
+    elif pe == 'sincos': W_pos = PositionalEncoding(q_len, d_model, normalize=True)
+    else: raise ValueError(f"{pe} is not a valid pe (positional encoder. Available types: 'gauss'=='normal', \
+        'zeros', 'zero', uniform', 'lin1d', 'exp1d', 'lin2d', 'exp2d', 'sincos', None.)")
+    return nn.Parameter(W_pos, requires_grad=learn_pe)

time_series_forecasting/layers/RevIN.py

@@ -0,0 +1,63 @@
+# code from https://github.com/ts-kim/RevIN, with minor modifications
+
+import torch
+import torch.nn as nn
+
+class RevIN(nn.Module):
+    def __init__(self, num_features: int, eps=1e-5, affine=True, subtract_last=False):
+        """
+        :param num_features: the number of features or channels
+        :param eps: a value added for numerical stability
+        :param affine: if True, RevIN has learnable affine parameters
+        """
+        super(RevIN, self).__init__()
+        self.num_features = num_features
+        self.eps = eps
+        self.affine = affine
+        self.subtract_last = subtract_last
+        if self.affine:
+            self._init_params()
+
+    def forward(self, x, mode:str):
+        if mode == 'norm':
+            self._get_statistics(x)
+            x = self._normalize(x)
+        elif mode == 'denorm':
+            x = self._denormalize(x)
+        else: raise NotImplementedError
+        return x
+
+    def _init_params(self):
+        # initialize RevIN params: (C,)
+        self.affine_weight = nn.Parameter(torch.ones(self.num_features))
+        self.affine_bias = nn.Parameter(torch.zeros(self.num_features))
+
+    def _get_statistics(self, x):
+        dim2reduce = tuple(range(1, x.ndim-1))
+        if self.subtract_last:
+            self.last = x[:,-1,:].unsqueeze(1)
+        else:
+            self.mean = torch.mean(x, dim=dim2reduce, keepdim=True).detach()
+        self.stdev = torch.sqrt(torch.var(x, dim=dim2reduce, keepdim=True, unbiased=False) + self.eps).detach()
+
+    def _normalize(self, x):
+        if self.subtract_last:
+            x = x - self.last
+        else:
+            x = x - self.mean
+        x = x / self.stdev
+        if self.affine:
+            x = x * self.affine_weight
+            x = x + self.affine_bias
+        return x
+
+    def _denormalize(self, x):
+        if self.affine:
+            x = x - self.affine_bias
+            x = x / (self.affine_weight + self.eps*self.eps)
+        x = x * self.stdev
+        if self.subtract_last:
+            x = x + self.last
+        else:
+            x = x + self.mean
+        return x

time_series_forecasting/layers/SelfAttention_Family.py

@@ -0,0 +1,224 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+import matplotlib.pyplot as plt
+
+import numpy as np
+import math
+from math import sqrt
+from utils.masking import TriangularCausalMask, ProbMask
+import os
+from einops import rearrange, repeat
+
+class FullAttention(nn.Module):
+    def __init__(self, mask_flag=True, factor=5, scale=None, attention_dropout=0.1, output_attention=False):
+        super(FullAttention, self).__init__()
+        self.scale = scale
+        self.mask_flag = mask_flag
+        self.output_attention = output_attention
+        self.dropout = nn.Dropout(attention_dropout)
+
+    def forward(self, queries, keys, values, attn_mask, tau=None, delta=None):
+        B, L, H, E = queries.shape
+        _, S, _, D = values.shape
+        scale = self.scale or 1. / sqrt(E)
+
+        scores = torch.einsum("blhe,bshe->bhls", queries, keys)
+
+        if self.mask_flag:
+            if attn_mask is None:
+                attn_mask = TriangularCausalMask(B, L, device=queries.device)
+
+            scores.masked_fill_(attn_mask.mask, -np.inf)
+
+        A = self.dropout(torch.softmax(scale * scores, dim=-1))
+        V = torch.einsum("bhls,bshd->blhd", A, values)
+
+        if self.output_attention:
+            return (V.contiguous(), A)
+        else:
+            return (V.contiguous(), None)
+
+
+class ProbAttention(nn.Module):
+    def __init__(self, mask_flag=True, factor=5, scale=None, attention_dropout=0.1, output_attention=False):
+        super(ProbAttention, self).__init__()
+        self.factor = factor
+        self.scale = scale
+        self.mask_flag = mask_flag
+        self.output_attention = output_attention
+        self.dropout = nn.Dropout(attention_dropout)
+
+    def _prob_QK(self, Q, K, sample_k, n_top):  # n_top: c*ln(L_q)
+        # Q [B, H, L, D]
+        B, H, L_K, E = K.shape
+        _, _, L_Q, _ = Q.shape
+
+        # calculate the sampled Q_K
+        K_expand = K.unsqueeze(-3).expand(B, H, L_Q, L_K, E)
+        index_sample = torch.randint(L_K, (L_Q, sample_k))  # real U = U_part(factor*ln(L_k))*L_q
+        K_sample = K_expand[:, :, torch.arange(L_Q).unsqueeze(1), index_sample, :]
+        Q_K_sample = torch.matmul(Q.unsqueeze(-2), K_sample.transpose(-2, -1)).squeeze()
+
+        # find the Top_k query with sparisty measurement
+        M = Q_K_sample.max(-1)[0] - torch.div(Q_K_sample.sum(-1), L_K)
+        M_top = M.topk(n_top, sorted=False)[1]
+
+        # use the reduced Q to calculate Q_K
+        Q_reduce = Q[torch.arange(B)[:, None, None],
+                   torch.arange(H)[None, :, None],
+                   M_top, :]  # factor*ln(L_q)
+        Q_K = torch.matmul(Q_reduce, K.transpose(-2, -1))  # factor*ln(L_q)*L_k
+
+        return Q_K, M_top
+
+    def _get_initial_context(self, V, L_Q):
+        B, H, L_V, D = V.shape
+        if not self.mask_flag:
+            # V_sum = V.sum(dim=-2)
+            V_sum = V.mean(dim=-2)
+            contex = V_sum.unsqueeze(-2).expand(B, H, L_Q, V_sum.shape[-1]).clone()
+        else:  # use mask
+            assert (L_Q == L_V)  # requires that L_Q == L_V, i.e. for self-attention only
+            contex = V.cumsum(dim=-2)
+        return contex
+
+    def _update_context(self, context_in, V, scores, index, L_Q, attn_mask):
+        B, H, L_V, D = V.shape
+
+        if self.mask_flag:
+            attn_mask = ProbMask(B, H, L_Q, index, scores, device=V.device)
+            scores.masked_fill_(attn_mask.mask, -np.inf)
+
+        attn = torch.softmax(scores, dim=-1)  # nn.Softmax(dim=-1)(scores)
+
+        context_in[torch.arange(B)[:, None, None],
+        torch.arange(H)[None, :, None],
+        index, :] = torch.matmul(attn, V).type_as(context_in)
+        if self.output_attention:
+            attns = (torch.ones([B, H, L_V, L_V]) / L_V).type_as(attn).to(attn.device)
+            attns[torch.arange(B)[:, None, None], torch.arange(H)[None, :, None], index, :] = attn
+            return (context_in, attns)
+        else:
+            return (context_in, None)
+
+    def forward(self, queries, keys, values, attn_mask):
+        B, L_Q, H, D = queries.shape
+        _, L_K, _, _ = keys.shape
+
+        queries = queries.transpose(2, 1)
+        keys = keys.transpose(2, 1)
+        values = values.transpose(2, 1)
+
+        U_part = self.factor * np.ceil(np.log(L_K)).astype('int').item()  # c*ln(L_k)
+        u = self.factor * np.ceil(np.log(L_Q)).astype('int').item()  # c*ln(L_q)
+
+        U_part = U_part if U_part < L_K else L_K
+        u = u if u < L_Q else L_Q
+
+        scores_top, index = self._prob_QK(queries, keys, sample_k=U_part, n_top=u)
+
+        # add scale factor
+        scale = self.scale or 1. / sqrt(D)
+        if scale is not None:
+            scores_top = scores_top * scale
+        # get the context
+        context = self._get_initial_context(values, L_Q)
+        # update the context with selected top_k queries
+        context, attn = self._update_context(context, values, scores_top, index, L_Q, attn_mask)
+
+        return context.contiguous(), attn
+
+
+class AttentionLayer(nn.Module):
+    def __init__(self, attention, d_model, n_heads, d_keys=None,
+                 d_values=None):
+        super(AttentionLayer, self).__init__()
+
+        d_keys = d_keys or (d_model // n_heads)
+        d_values = d_values or (d_model // n_heads)
+
+        self.inner_attention = attention
+        self.query_projection = nn.Linear(d_model, d_keys * n_heads)
+        self.key_projection = nn.Linear(d_model, d_keys * n_heads)
+        self.value_projection = nn.Linear(d_model, d_values * n_heads)
+        self.out_projection = nn.Linear(d_values * n_heads, d_model)
+        self.n_heads = n_heads
+
+    def forward(self, queries, keys, values, attn_mask, tau=None, delta=None):
+        B, L, _ = queries.shape
+        _, S, _ = keys.shape
+        H = self.n_heads
+
+        queries = self.query_projection(queries).view(B, L, H, -1)
+        keys = self.key_projection(keys).view(B, S, H, -1)
+        values = self.value_projection(values).view(B, S, H, -1)
+
+        out, attn = self.inner_attention(
+            queries,
+            keys,
+            values,
+            attn_mask
+        )
+        out = out.view(B, L, -1)
+
+        return self.out_projection(out), attn
+
+class TwoStageAttentionLayer(nn.Module):
+    '''
+    The Two Stage Attention (TSA) Layer
+    input/output shape: [batch_size, Data_dim(D), Seg_num(L), d_model]
+    '''
+
+    def __init__(self, configs,
+                 seg_num, factor, d_model, n_heads, d_ff=None, dropout=0.1):
+        super(TwoStageAttentionLayer, self).__init__()
+        d_ff = d_ff or 4 * d_model
+        self.time_attention = AttentionLayer(FullAttention(False, configs.factor, attention_dropout=configs.dropout,
+                                                           output_attention=configs.output_attention), d_model, n_heads)
+        self.dim_sender = AttentionLayer(FullAttention(False, configs.factor, attention_dropout=configs.dropout,
+                                                       output_attention=configs.output_attention), d_model, n_heads)
+        self.dim_receiver = AttentionLayer(FullAttention(False, configs.factor, attention_dropout=configs.dropout,
+                                                         output_attention=configs.output_attention), d_model, n_heads)
+        self.router = nn.Parameter(torch.randn(seg_num, factor, d_model))
+
+        self.dropout = nn.Dropout(dropout)
+
+        self.norm1 = nn.LayerNorm(d_model)
+        self.norm2 = nn.LayerNorm(d_model)
+        self.norm3 = nn.LayerNorm(d_model)
+        self.norm4 = nn.LayerNorm(d_model)
+
+        self.MLP1 = nn.Sequential(nn.Linear(d_model, d_ff),
+                                  nn.GELU(),
+                                  nn.Linear(d_ff, d_model))
+        self.MLP2 = nn.Sequential(nn.Linear(d_model, d_ff),
+                                  nn.GELU(),
+                                  nn.Linear(d_ff, d_model))
+
+    def forward(self, x, attn_mask=None, tau=None, delta=None):
+        # Cross Time Stage: Directly apply MSA to each dimension
+        batch = x.shape[0]
+        time_in = rearrange(x, 'b ts_d seg_num d_model -> (b ts_d) seg_num d_model')
+        time_enc, attn = self.time_attention(
+            time_in, time_in, time_in, attn_mask=None, tau=None, delta=None
+        )
+        dim_in = time_in + self.dropout(time_enc)
+        dim_in = self.norm1(dim_in)
+        dim_in = dim_in + self.dropout(self.MLP1(dim_in))
+        dim_in = self.norm2(dim_in)
+
+        # Cross Dimension Stage: use a small set of learnable vectors to aggregate and distribute messages to build the D-to-D connection
+        dim_send = rearrange(dim_in, '(b ts_d) seg_num d_model -> (b seg_num) ts_d d_model', b=batch)
+        batch_router = repeat(self.router, 'seg_num factor d_model -> (repeat seg_num) factor d_model', repeat=batch)
+        dim_buffer, attn = self.dim_sender(batch_router, dim_send, dim_send, attn_mask=None, tau=None, delta=None)
+        dim_receive, attn = self.dim_receiver(dim_send, dim_buffer, dim_buffer, attn_mask=None, tau=None, delta=None)
+        dim_enc = dim_send + self.dropout(dim_receive)
+        dim_enc = self.norm3(dim_enc)
+        dim_enc = dim_enc + self.dropout(self.MLP2(dim_enc))
+        dim_enc = self.norm4(dim_enc)
+
+        final_out = rearrange(dim_enc, '(b seg_num) ts_d d_model -> b ts_d seg_num d_model', b=batch)
+
+        return final_out

time_series_forecasting/layers/Transformer_EncDec.py

@@ -0,0 +1,131 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+class ConvLayer(nn.Module):
+    def __init__(self, c_in):
+        super(ConvLayer, self).__init__()
+        self.downConv = nn.Conv1d(in_channels=c_in,
+                                  out_channels=c_in,
+                                  kernel_size=3,
+                                  padding=2,
+                                  padding_mode='circular')
+        self.norm = nn.BatchNorm1d(c_in)
+        self.activation = nn.ELU()
+        self.maxPool = nn.MaxPool1d(kernel_size=3, stride=2, padding=1)
+
+    def forward(self, x):
+        x = self.downConv(x.permute(0, 2, 1))
+        x = self.norm(x)
+        x = self.activation(x)
+        x = self.maxPool(x)
+        x = x.transpose(1, 2)
+        return x
+
+
+class EncoderLayer(nn.Module):
+    def __init__(self, attention, d_model, d_ff=None, dropout=0.1, activation="relu"):
+        super(EncoderLayer, self).__init__()
+        d_ff = d_ff or 4 * d_model
+        self.attention = attention
+        self.conv1 = nn.Conv1d(in_channels=d_model, out_channels=d_ff, kernel_size=1)
+        self.conv2 = nn.Conv1d(in_channels=d_ff, out_channels=d_model, kernel_size=1)
+        self.norm1 = nn.LayerNorm(d_model)
+        self.norm2 = nn.LayerNorm(d_model)
+        self.dropout = nn.Dropout(dropout)
+        self.activation = F.relu if activation == "relu" else F.gelu
+
+    def forward(self, x, attn_mask=None):
+        new_x, attn = self.attention(
+            x, x, x,
+            attn_mask=attn_mask
+        )
+        x = x + self.dropout(new_x)
+
+        y = x = self.norm1(x)
+        y = self.dropout(self.activation(self.conv1(y.transpose(-1, 1))))
+        y = self.dropout(self.conv2(y).transpose(-1, 1))
+
+        return self.norm2(x + y), attn
+
+
+class Encoder(nn.Module):
+    def __init__(self, attn_layers, conv_layers=None, norm_layer=None):
+        super(Encoder, self).__init__()
+        self.attn_layers = nn.ModuleList(attn_layers)
+        self.conv_layers = nn.ModuleList(conv_layers) if conv_layers is not None else None
+        self.norm = norm_layer
+
+    def forward(self, x, attn_mask=None):
+        # x [B, L, D]
+        attns = []
+        if self.conv_layers is not None:
+            for attn_layer, conv_layer in zip(self.attn_layers, self.conv_layers):
+                x, attn = attn_layer(x, attn_mask=attn_mask)
+                x = conv_layer(x)
+                attns.append(attn)
+            x, attn = self.attn_layers[-1](x)
+            attns.append(attn)
+        else:
+            for attn_layer in self.attn_layers:
+                x, attn = attn_layer(x, attn_mask=attn_mask)
+                attns.append(attn)
+
+        if self.norm is not None:
+            x = self.norm(x)
+
+        return x, attns
+
+
+class DecoderLayer(nn.Module):
+    def __init__(self, self_attention, cross_attention, d_model, d_ff=None,
+                 dropout=0.1, activation="relu"):
+        super(DecoderLayer, self).__init__()
+        d_ff = d_ff or 4 * d_model
+        self.self_attention = self_attention
+        self.cross_attention = cross_attention
+        self.conv1 = nn.Conv1d(in_channels=d_model, out_channels=d_ff, kernel_size=1)
+        self.conv2 = nn.Conv1d(in_channels=d_ff, out_channels=d_model, kernel_size=1)
+        self.norm1 = nn.LayerNorm(d_model)
+        self.norm2 = nn.LayerNorm(d_model)
+        self.norm3 = nn.LayerNorm(d_model)
+        self.dropout = nn.Dropout(dropout)
+        self.activation = F.relu if activation == "relu" else F.gelu
+
+    def forward(self, x, cross, x_mask=None, cross_mask=None):
+        x = x + self.dropout(self.self_attention(
+            x, x, x,
+            attn_mask=x_mask
+        )[0])
+        x = self.norm1(x)
+
+        x = x + self.dropout(self.cross_attention(
+            x, cross, cross,
+            attn_mask=cross_mask
+        )[0])
+
+        y = x = self.norm2(x)
+        y = self.dropout(self.activation(self.conv1(y.transpose(-1, 1))))
+        y = self.dropout(self.conv2(y).transpose(-1, 1))
+
+        return self.norm3(x + y)
+
+
+class Decoder(nn.Module):
+    def __init__(self, layers, norm_layer=None, projection=None):
+        super(Decoder, self).__init__()
+        self.layers = nn.ModuleList(layers)
+        self.norm = norm_layer
+        self.projection = projection
+
+    def forward(self, x, cross, x_mask=None, cross_mask=None):
+        for layer in self.layers:
+            x = layer(x, cross, x_mask=x_mask, cross_mask=cross_mask)
+
+        if self.norm is not None:
+            x = self.norm(x)
+
+        if self.projection is not None:
+            x = self.projection(x)
+        return x

time_series_forecasting/models/Autoformer.py

@@ -0,0 +1,123 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from layers.Embed import DataEmbedding, DataEmbedding_wo_pos,DataEmbedding_wo_pos_temp,DataEmbedding_wo_temp
+from layers.AutoCorrelation import AutoCorrelation, AutoCorrelationLayer
+from layers.Autoformer_EncDec import Encoder, Decoder, EncoderLayer, DecoderLayer, my_Layernorm, series_decomp
+import math
+import numpy as np
+
+
+class Model(nn.Module):
+    """
+    Autoformer is the first method to achieve the series-wise connection,
+    with inherent O(LlogL) complexity
+    """
+    def __init__(self, configs):
+        super(Model, self).__init__()
+        self.seq_len = configs.seq_len
+        self.label_len = configs.label_len
+        self.pred_len = configs.pred_len
+        self.output_attention = configs.output_attention
+
+        # Decomp
+        kernel_size = configs.moving_avg
+        self.decomp = series_decomp(kernel_size)
+
+        # Embedding
+        # The series-wise connection inherently contains the sequential information.
+        # Thus, we can discard the position embedding of transformers.
+        if configs.embed_type == 0:
+            self.enc_embedding = DataEmbedding_wo_pos(configs.enc_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+            self.dec_embedding = DataEmbedding_wo_pos(configs.dec_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+        elif configs.embed_type == 1:
+            self.enc_embedding = DataEmbedding(configs.enc_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+            self.dec_embedding = DataEmbedding(configs.dec_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+        elif configs.embed_type == 2:
+            self.enc_embedding = DataEmbedding_wo_pos(configs.enc_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+            self.dec_embedding = DataEmbedding_wo_pos(configs.dec_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+
+        elif configs.embed_type == 3:
+            self.enc_embedding = DataEmbedding_wo_temp(configs.enc_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+            self.dec_embedding = DataEmbedding_wo_temp(configs.dec_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+        elif configs.embed_type == 4:
+            self.enc_embedding = DataEmbedding_wo_pos_temp(configs.enc_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+            self.dec_embedding = DataEmbedding_wo_pos_temp(configs.dec_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+        
+        # Encoder
+        self.encoder = Encoder(
+            [
+                EncoderLayer(
+                    AutoCorrelationLayer(
+                        AutoCorrelation(False, configs.factor, attention_dropout=configs.dropout,
+                                        output_attention=configs.output_attention),
+                        configs.d_model, configs.n_heads),
+                    configs.d_model,
+                    configs.d_ff,
+                    moving_avg=configs.moving_avg,
+                    dropout=configs.dropout,
+                    activation=configs.activation
+                ) for l in range(configs.e_layers)
+            ],
+            norm_layer=my_Layernorm(configs.d_model)
+        )
+        # Decoder
+        self.decoder = Decoder(
+            [
+                DecoderLayer(
+                    AutoCorrelationLayer(
+                        AutoCorrelation(True, configs.factor, attention_dropout=configs.dropout,
+                                        output_attention=False),
+                        configs.d_model, configs.n_heads),
+                    AutoCorrelationLayer(
+                        AutoCorrelation(False, configs.factor, attention_dropout=configs.dropout,
+                                        output_attention=False),
+                        configs.d_model, configs.n_heads),
+                    configs.d_model,
+                    configs.c_out,
+                    configs.d_ff,
+                    moving_avg=configs.moving_avg,
+                    dropout=configs.dropout,
+                    activation=configs.activation,
+                )
+                for l in range(configs.d_layers)
+            ],
+            norm_layer=my_Layernorm(configs.d_model),
+            projection=nn.Linear(configs.d_model, configs.c_out, bias=True)
+        )
+
+
+            
+    def forward(self, x_enc, x_mark_enc, x_dec, x_mark_dec,
+                enc_self_mask=None, dec_self_mask=None, dec_enc_mask=None):
+        # decomp init
+        mean = torch.mean(x_enc, dim=1).unsqueeze(1).repeat(1, self.pred_len, 1)
+        zeros = torch.zeros([x_dec.shape[0], self.pred_len, x_dec.shape[2]], device=x_enc.device)
+        seasonal_init, trend_init = self.decomp(x_enc)
+        # decoder input
+        trend_init = torch.cat([trend_init[:, -self.label_len:, :], mean], dim=1)
+        seasonal_init = torch.cat([seasonal_init[:, -self.label_len:, :], zeros], dim=1)
+        # enc
+        enc_out = self.enc_embedding(x_enc, x_mark_enc)
+        enc_out, attns = self.encoder(enc_out, attn_mask=enc_self_mask)
+        # dec
+        dec_out = self.dec_embedding(seasonal_init, x_mark_dec)
+        seasonal_part, trend_part = self.decoder(dec_out, enc_out, x_mask=dec_self_mask, cross_mask=dec_enc_mask,
+                                                 trend=trend_init)
+        # final
+        dec_out = trend_part + seasonal_part
+
+        if self.output_attention:
+            return dec_out[:, -self.pred_len:, :], attns
+        else:
+            return dec_out[:, -self.pred_len:, :]  # [B, L, D]

time_series_forecasting/models/CycleNet.py

@@ -0,0 +1,67 @@
+import torch
+import torch.nn as nn
+
+class RecurrentCycle(torch.nn.Module):
+    # Thanks for the contribution of wayhoww.
+    # The new implementation uses index arithmetic with modulo to directly gather cyclic data in a single operation,
+    # while the original implementation manually rolls and repeats the data through looping.
+    # It achieves a significant speed improvement (2x ~ 3x acceleration).
+    # See https://github.com/ACAT-SCUT/CycleNet/pull/4 for more details.
+    def __init__(self, cycle_len, channel_size):
+        super(RecurrentCycle, self).__init__()
+        self.cycle_len = cycle_len
+        self.channel_size = channel_size
+        self.data = torch.nn.Parameter(torch.zeros(cycle_len, channel_size), requires_grad=True)
+
+    def forward(self, index, length):
+        gather_index = (index.view(-1, 1) + torch.arange(length, device=index.device).view(1, -1)) % self.cycle_len    
+        return self.data[gather_index]
+
+
+class Model(nn.Module):
+    def __init__(self, configs):
+        super(Model, self).__init__()
+
+        self.seq_len = configs.seq_len
+        self.pred_len = configs.pred_len
+        self.enc_in = configs.enc_in
+        self.cycle_len = configs.cycle
+        self.model_type = configs.model_type
+        self.d_model = configs.d_model
+        self.use_revin = configs.use_revin
+
+        self.cycleQueue = RecurrentCycle(cycle_len=self.cycle_len, channel_size=self.enc_in)
+
+        assert self.model_type in ['linear', 'mlp']
+        if self.model_type == 'linear':
+            self.model = nn.Linear(self.seq_len, self.pred_len)
+        elif self.model_type == 'mlp':
+            self.model = nn.Sequential(
+                nn.Linear(self.seq_len, self.d_model),
+                nn.ReLU(),
+                nn.Linear(self.d_model, self.pred_len)
+            )
+
+    def forward(self, x, cycle_index):
+        # x: (batch_size, seq_len, enc_in), cycle_index: (batch_size,)
+
+        # instance norm
+        if self.use_revin:
+            seq_mean = torch.mean(x, dim=1, keepdim=True)
+            seq_var = torch.var(x, dim=1, keepdim=True) + 1e-5
+            x = (x - seq_mean) / torch.sqrt(seq_var)
+
+        # remove the cycle of the input data
+        x = x - self.cycleQueue(cycle_index, self.seq_len)
+
+        # forecasting with channel independence (parameters-sharing)
+        y = self.model(x.permute(0, 2, 1)).permute(0, 2, 1)
+
+        # add back the cycle of the output data
+        y = y + self.cycleQueue((cycle_index + self.seq_len) % self.cycle_len, self.pred_len)
+
+        # instance denorm
+        if self.use_revin:
+            y = y * torch.sqrt(seq_var) + seq_mean
+
+        return y

time_series_forecasting/models/DLinear.py

@@ -0,0 +1,87 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import numpy as np
+
+class moving_avg(nn.Module):
+    """
+    Moving average block to highlight the trend of time series
+    """
+    def __init__(self, kernel_size, stride):
+        super(moving_avg, self).__init__()
+        self.kernel_size = kernel_size
+        self.avg = nn.AvgPool1d(kernel_size=kernel_size, stride=stride, padding=0)
+
+    def forward(self, x):
+        # padding on the both ends of time series
+        front = x[:, 0:1, :].repeat(1, (self.kernel_size - 1) // 2, 1)
+        end = x[:, -1:, :].repeat(1, (self.kernel_size - 1) // 2, 1)
+        x = torch.cat([front, x, end], dim=1)
+        x = self.avg(x.permute(0, 2, 1))
+        x = x.permute(0, 2, 1)
+        return x
+
+
+class series_decomp(nn.Module):
+    """
+    Series decomposition block
+    """
+    def __init__(self, kernel_size):
+        super(series_decomp, self).__init__()
+        self.moving_avg = moving_avg(kernel_size, stride=1)
+
+    def forward(self, x):
+        moving_mean = self.moving_avg(x)
+        res = x - moving_mean
+        return res, moving_mean
+
+class Model(nn.Module):
+    """
+    Decomposition-Linear
+    """
+    def __init__(self, configs):
+        super(Model, self).__init__()
+        self.seq_len = configs.seq_len
+        self.pred_len = configs.pred_len
+
+        # Decompsition Kernel Size
+        kernel_size = 25
+        self.decompsition = series_decomp(kernel_size)
+        self.individual = configs.individual
+        self.channels = configs.enc_in
+
+        if self.individual:
+            self.Linear_Seasonal = nn.ModuleList()
+            self.Linear_Trend = nn.ModuleList()
+            
+            for i in range(self.channels):
+                self.Linear_Seasonal.append(nn.Linear(self.seq_len,self.pred_len))
+                self.Linear_Trend.append(nn.Linear(self.seq_len,self.pred_len))
+
+                # Use this two lines if you want to visualize the weights
+                # self.Linear_Seasonal[i].weight = nn.Parameter((1/self.seq_len)*torch.ones([self.pred_len,self.seq_len]))
+                # self.Linear_Trend[i].weight = nn.Parameter((1/self.seq_len)*torch.ones([self.pred_len,self.seq_len]))
+        else:
+            self.Linear_Seasonal = nn.Linear(self.seq_len,self.pred_len)
+            self.Linear_Trend = nn.Linear(self.seq_len,self.pred_len)
+            
+            # Use this two lines if you want to visualize the weights
+            # self.Linear_Seasonal.weight = nn.Parameter((1/self.seq_len)*torch.ones([self.pred_len,self.seq_len]))
+            # self.Linear_Trend.weight = nn.Parameter((1/self.seq_len)*torch.ones([self.pred_len,self.seq_len]))
+
+    def forward(self, x):
+        # x: [Batch, Input length, Channel]
+        seasonal_init, trend_init = self.decompsition(x)
+        seasonal_init, trend_init = seasonal_init.permute(0,2,1), trend_init.permute(0,2,1)
+        if self.individual:
+            seasonal_output = torch.zeros([seasonal_init.size(0),seasonal_init.size(1),self.pred_len],dtype=seasonal_init.dtype).to(seasonal_init.device)
+            trend_output = torch.zeros([trend_init.size(0),trend_init.size(1),self.pred_len],dtype=trend_init.dtype).to(trend_init.device)
+            for i in range(self.channels):
+                seasonal_output[:,i,:] = self.Linear_Seasonal[i](seasonal_init[:,i,:])
+                trend_output[:,i,:] = self.Linear_Trend[i](trend_init[:,i,:])
+        else:
+            seasonal_output = self.Linear_Seasonal(seasonal_init)
+            trend_output = self.Linear_Trend(trend_init)
+
+        x = seasonal_output + trend_output
+        return x.permute(0,2,1) # to [Batch, Output length, Channel]

time_series_forecasting/models/Informer.py

@@ -0,0 +1,101 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from utils.masking import TriangularCausalMask, ProbMask
+from layers.Transformer_EncDec import Decoder, DecoderLayer, Encoder, EncoderLayer, ConvLayer
+from layers.SelfAttention_Family import FullAttention, ProbAttention, AttentionLayer
+from layers.Embed import DataEmbedding,DataEmbedding_wo_pos,DataEmbedding_wo_temp,DataEmbedding_wo_pos_temp
+import numpy as np
+
+
+class Model(nn.Module):
+    """
+    Informer with Propspare attention in O(LlogL) complexity
+    """
+    def __init__(self, configs):
+        super(Model, self).__init__()
+        self.pred_len = configs.pred_len
+        self.output_attention = configs.output_attention
+
+        # Embedding
+        if configs.embed_type == 0:
+            self.enc_embedding = DataEmbedding(configs.enc_in, configs.d_model, configs.embed, configs.freq,
+                                            configs.dropout)
+            self.dec_embedding = DataEmbedding(configs.dec_in, configs.d_model, configs.embed, configs.freq,
+                                           configs.dropout)
+        elif configs.embed_type == 1:
+            self.enc_embedding = DataEmbedding(configs.enc_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+            self.dec_embedding = DataEmbedding(configs.dec_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+        elif configs.embed_type == 2:
+            self.enc_embedding = DataEmbedding_wo_pos(configs.enc_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+            self.dec_embedding = DataEmbedding_wo_pos(configs.dec_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+
+        elif configs.embed_type == 3:
+            self.enc_embedding = DataEmbedding_wo_temp(configs.enc_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+            self.dec_embedding = DataEmbedding_wo_temp(configs.dec_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+        elif configs.embed_type == 4:
+            self.enc_embedding = DataEmbedding_wo_pos_temp(configs.enc_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+            self.dec_embedding = DataEmbedding_wo_pos_temp(configs.dec_in, configs.d_model, configs.embed, configs.freq,
+                                                    configs.dropout)
+        # Encoder
+        self.encoder = Encoder(
+            [
+                EncoderLayer(
+                    AttentionLayer(
+                        ProbAttention(False, configs.factor, attention_dropout=configs.dropout,
+                                      output_attention=configs.output_attention),
+                        configs.d_model, configs.n_heads),
+                    configs.d_model,
+                    configs.d_ff,
+                    dropout=configs.dropout,
+                    activation=configs.activation
+                ) for l in range(configs.e_layers)
+            ],
+            [
+                ConvLayer(
+                    configs.d_model
+                ) for l in range(configs.e_layers - 1)
+            ] if configs.distil else None,
+            norm_layer=torch.nn.LayerNorm(configs.d_model)
+        )
+        # Decoder
+        self.decoder = Decoder(
+            [
+                DecoderLayer(
+                    AttentionLayer(
+                        ProbAttention(True, configs.factor, attention_dropout=configs.dropout, output_attention=False),
+                        configs.d_model, configs.n_heads),
+                    AttentionLayer(
+                        ProbAttention(False, configs.factor, attention_dropout=configs.dropout, output_attention=False),
+                        configs.d_model, configs.n_heads),
+                    configs.d_model,
+                    configs.d_ff,
+                    dropout=configs.dropout,
+                    activation=configs.activation,
+                )
+                for l in range(configs.d_layers)
+            ],
+            norm_layer=torch.nn.LayerNorm(configs.d_model),
+            projection=nn.Linear(configs.d_model, configs.c_out, bias=True)
+        )
+
+    def forward(self, x_enc, x_mark_enc, x_dec, x_mark_dec,
+                enc_self_mask=None, dec_self_mask=None, dec_enc_mask=None):
+
+        enc_out = self.enc_embedding(x_enc, x_mark_enc)
+        enc_out, attns = self.encoder(enc_out, attn_mask=enc_self_mask)
+
+        dec_out = self.dec_embedding(x_dec, x_mark_dec)
+        dec_out = self.decoder(dec_out, enc_out, x_mask=dec_self_mask, cross_mask=dec_enc_mask)
+
+        if self.output_attention:
+            return dec_out[:, -self.pred_len:, :], attns
+        else:
+            return dec_out[:, -self.pred_len:, :]  # [B, L, D]

time_series_forecasting/models/LightTS.py

@@ -0,0 +1,114 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+class IEBlock(nn.Module):
+    def __init__(self, input_dim, hid_dim, output_dim, num_node):
+        super(IEBlock, self).__init__()
+
+        self.input_dim = input_dim
+        self.hid_dim = hid_dim
+        self.output_dim = output_dim
+        self.num_node = num_node
+
+        self._build()
+
+    def _build(self):
+        self.spatial_proj = nn.Sequential(
+            nn.Linear(self.input_dim, self.hid_dim),
+            nn.LeakyReLU(),
+            nn.Linear(self.hid_dim, self.hid_dim // 4)
+        )
+
+        self.channel_proj = nn.Linear(self.num_node, self.num_node)
+        torch.nn.init.eye_(self.channel_proj.weight)
+
+        self.output_proj = nn.Linear(self.hid_dim // 4, self.output_dim)
+
+
+    def forward(self, x):
+        x = self.spatial_proj(x.permute(0, 2, 1))
+        x = x.permute(0, 2, 1) + self.channel_proj(x.permute(0, 2, 1))
+        x = self.output_proj(x.permute(0, 2, 1))
+
+        x = x.permute(0, 2, 1)
+
+        return x
+
+
+class Model(nn.Module):
+    def __init__(self, config):
+        super(Model, self).__init__()
+        self.lookback = config.seq_len
+        self.lookahead = config.pred_len
+        self.chunk_size = config.chunk_size
+        assert(self.lookback % self.chunk_size == 0)
+        self.num_chunks = self.lookback // self.chunk_size
+        self.hid_dim = config.d_model
+        self.num_node = config.enc_in
+        self.dropout = config.dropout
+        self._build()
+
+    def _build(self):
+        self.layer_1 = IEBlock(
+            input_dim=self.chunk_size,
+            hid_dim=self.hid_dim // 4,
+            output_dim=self.hid_dim // 4,
+            num_node=self.num_chunks
+        )
+
+        self.chunk_proj_1 = nn.Linear(self.num_chunks, 1)
+
+        self.layer_2 = IEBlock(
+            input_dim=self.chunk_size,
+            hid_dim=self.hid_dim // 4,
+            output_dim=self.hid_dim // 4,
+            num_node=self.num_chunks
+        )
+
+        self.chunk_proj_2 = nn.Linear(self.num_chunks, 1)
+
+        self.layer_3 = IEBlock(
+            input_dim=self.hid_dim // 2,
+            hid_dim=self.hid_dim // 2,
+            output_dim=self.lookahead,
+            num_node=self.num_node
+        )
+
+        # self.ar = nn.Sequential(
+        #     nn.Linear(self.lookback, self.hid_dim //4),
+        #     nn.LeakyReLU(),
+        #     nn.Linear(self.hid_dim // 4, self.lookahead)
+        # )
+        self.ar = nn.Linear(self.lookback, self.lookahead)
+
+    def forward(self, x):
+        B, T, N = x.size()
+
+        highway = self.ar(x.permute(0, 2, 1))
+        highway = highway.permute(0, 2, 1)
+
+        # continuous sampling
+        x1 = x.reshape(B, self.num_chunks, self.chunk_size, N)
+        x1 = x1.permute(0, 3, 2, 1)
+        x1 = x1.reshape(-1, self.chunk_size, self.num_chunks)
+        x1 = self.layer_1(x1)
+        x1 = self.chunk_proj_1(x1).squeeze(dim=-1)
+        
+        # interval sampling
+        x2 = x.reshape(B, self.chunk_size, self.num_chunks, N)
+        x2 = x2.permute(0, 3, 1, 2)
+        x2 = x2.reshape(-1, self.chunk_size, self.num_chunks)
+        x2 = self.layer_2(x2)
+        x2 = self.chunk_proj_2(x2).squeeze(dim=-1)
+
+        x3 = torch.cat([x1, x2], dim=-1)
+
+        x3 = x3.reshape(B, N, -1)
+        x3 = x3.permute(0, 2, 1)
+
+        out = self.layer_3(x3)
+
+        out = out + highway
+        return out

time_series_forecasting/models/Linear.py

@@ -0,0 +1,21 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import numpy as np
+
+class Model(nn.Module):
+    """
+    Just one Linear layer
+    """
+    def __init__(self, configs):
+        super(Model, self).__init__()
+        self.seq_len = configs.seq_len
+        self.pred_len = configs.pred_len
+        self.Linear = nn.Linear(self.seq_len, self.pred_len)
+        # Use this line if you want to visualize the weights
+        # self.Linear.weight = nn.Parameter((1/self.seq_len)*torch.ones([self.pred_len,self.seq_len]))
+
+    def forward(self, x):
+        # x: [Batch, Input length, Channel]
+        x = self.Linear(x.permute(0,2,1)).permute(0,2,1)
+        return x # [Batch, Output length, Channel]