Datasets:
Refresh executable CB-Telemetry review artifact
Browse filesAdd released bottleneck feature tables, executable retrieval evaluation scripts, refreshed archive checksum, and the Zenodo draft review link. Remove stale wav viewer files.
- README.md +25 -20
- archive/cb_telemetry_v1_expert_aligned.tar.gz +2 -2
- archive/cb_telemetry_v1_expert_aligned.tar.gz.sha256 +1 -1
- audio_sample/yamanakako/20140402/cbt_yamanakako_old_20140402_20140402060501yamanakako_clip.m4a +0 -0
- audio_sample_wav/README.md +0 -3
- audio_sample_wav/chichibu/20120407/cbt_tetto_20120407_20120407053414tetto_clip.wav +0 -3
- audio_sample_wav/chichibu/20120407/cbt_tetto_20120407_20120407063415tetto_clip.wav +0 -3
- audio_sample_wav/chichibu/20120410/cbt_tetto_20120410_20120410053319tetto_clip.wav +0 -3
- audio_sample_wav/chichibu/20120410/cbt_tetto_20120410_20120410063320tetto_clip.wav +0 -3
- audio_sample_wav/chichibu/20120413/cbt_tetto_20120413_20120413050431tetto_clip.wav +0 -3
- audio_sample_wav/chichibu/20120413/cbt_tetto_20120413_20120413060432tetto_clip.wav +0 -3
- audio_sample_wav/chichibu/20120416/cbt_tetto_20120416_20120416070400tetto_clip.wav +0 -3
- audio_sample_wav/chichibu/20120425/cbt_tetto_20120425_20120425050327tetto_clip.wav +0 -3
- audio_sample_wav/furano/20200403/cbt_maeyama_20200403_20200403053159maeyama_clip.wav +0 -3
- audio_sample_wav/furano/20200411/cbt_maeyama_20200411_20200411053201maeyama_clip.wav +0 -3
- audio_sample_wav/furano/20200417/cbt_maeyama_20200417_20200417043209maeyama_clip.wav +0 -3
- audio_sample_wav/metadata.csv +0 -25
- audio_sample_wav/shiga/20200403/cbt_otanomo_20200403_20200403051249otanomo_clip.wav +0 -3
- audio_sample_wav/shiga/20200403/cbt_otanomo_20200403_20200403061257otanomo_clip.wav +0 -3
- audio_sample_wav/shiga/20200620/cbt_otanomo_20200620_20200620040430otanomo_clip.wav +0 -3
- audio_sample_wav/yamanakako/20140402/cbt_yamanakako_old_20140402_20140402060501yamanakako_clip.wav +0 -3
- audio_sample_wav/yamanakako/20150402/cbt_yamanakako_old_20150402_20150402055817yamanakako_clip.wav +0 -3
- audio_sample_wav/yamanakako/20160628/cbt_yamanakako_old_20160628_20160628054404yamanakako_clip.wav +0 -3
- audio_sample_wav/yatake/20120403/cbt_yatake_old_20120403_20120403061005yatake_clip.wav +0 -3
- audio_sample_wav/yatake/20120430/cbt_yatake_old_20120430_20120430045616yatake_clip.wav +0 -3
- audio_sample_wav/yatake/20140611/cbt_yatake_old_20140611_20140611040745yatake_clip.wav +0 -3
- audio_sample_wav/chichibu/20120404/cbt_tetto_20120404_20120404063333tetto_clip.wav → features/bottlenecks/default/opq_8bit_feature_table.csv.gz +2 -2
- audio_sample_wav/chichibu/20120401/cbt_tetto_20120401_20120401053323tetto_clip.wav → features/bottlenecks/default/pq_8bit_feature_table.csv.gz +2 -2
- audio_sample_wav/chichibu/20120401/cbt_tetto_20120401_20120401063324tetto_clip.wav → features/bottlenecks/default/standard_rvq_8bit_feature_table.csv.gz +2 -2
- audio_sample_wav/chichibu/20120404/cbt_tetto_20120404_20120404053332tetto_clip.wav → features/bottlenecks/strict_clean/opq_8bit_feature_table.csv.gz +2 -2
- features/bottlenecks/strict_clean/pq_8bit_feature_table.csv.gz +3 -0
- features/bottlenecks/strict_clean/standard_rvq_8bit_feature_table.csv.gz +3 -0
- hf_staging_validation.json +3 -3
- metadata.csv +24 -24
- requirements.txt +4 -0
- scripts/run_release_evaluation.py +366 -0
- scripts/run_retrieval_eval.py +629 -0
- scripts/validate_cb_telemetry.py +15 -6
README.md
CHANGED
|
@@ -21,11 +21,10 @@ size_categories:
|
|
| 21 |
- n<1K
|
| 22 |
configs:
|
| 23 |
- config_name: reviewer_audio
|
| 24 |
-
data_dir:
|
| 25 |
data_files:
|
| 26 |
- split: train
|
| 27 |
-
path: "**/*.
|
| 28 |
-
drop_labels: true
|
| 29 |
default: true
|
| 30 |
---
|
| 31 |
|
|
@@ -38,10 +37,11 @@ archive is preserved on Zenodo:
|
|
| 38 |
|
| 39 |
[https://doi.org/10.5281/zenodo.19951359](https://doi.org/10.5281/zenodo.19951359)
|
| 40 |
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
|
|
|
| 45 |
|
| 46 |
## What Is Hosted Here
|
| 47 |
|
|
@@ -51,6 +51,7 @@ This repository contains the lightweight ML-ready benchmark package:
|
|
| 51 |
- frozen v1 scored manifests with 805 expert-overlap recording rows;
|
| 52 |
- paired Default and Strict-Clean split metadata;
|
| 53 |
- frozen Default and Strict-Clean feature tables;
|
|
|
|
| 54 |
- refreshed representation, retrieval, metadata-control, and bottleneck
|
| 55 |
baseline outputs;
|
| 56 |
- validation and smoke-test scripts;
|
|
@@ -68,20 +69,18 @@ reconstructing the complete source-audio set from Cyberforest URLs.
|
|
| 68 |
## Reviewer Audio Viewer
|
| 69 |
|
| 70 |
The default Hugging Face viewer is configured for the reviewer audio sample in
|
| 71 |
-
`
|
| 72 |
-
same 24 M4A reviewer clips retained in `audio_sample/`. The viewer split can be
|
| 73 |
-
loaded locally with:
|
| 74 |
|
| 75 |
```python
|
| 76 |
from datasets import load_dataset
|
| 77 |
|
| 78 |
-
samples = load_dataset("audiofolder", data_dir="
|
| 79 |
print(samples)
|
| 80 |
```
|
| 81 |
|
| 82 |
The root-level `metadata.csv` points to the same 24 clips with paths relative to
|
| 83 |
-
the repository root. The `
|
| 84 |
-
|
| 85 |
|
| 86 |
## Benchmark Package Quickstart
|
| 87 |
|
|
@@ -92,6 +91,17 @@ python3 scripts/run_smoke_eval.py --root .
|
|
| 92 |
python3 scripts/validate_cb_telemetry.py --root . --write-report
|
| 93 |
```
|
| 94 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 95 |
To reconstruct the complete Default source-audio set:
|
| 96 |
|
| 97 |
```bash
|
|
@@ -120,7 +130,7 @@ The exact frozen v1 archive uploaded to Zenodo is also included under
|
|
| 120 |
`archive/`:
|
| 121 |
|
| 122 |
```text
|
| 123 |
-
|
| 124 |
```
|
| 125 |
|
| 126 |
The Zenodo DOI for the immutable v1 artifact is:
|
|
@@ -128,11 +138,7 @@ The Zenodo DOI for the immutable v1 artifact is:
|
|
| 128 |
```text
|
| 129 |
10.5281/zenodo.19951359
|
| 130 |
```
|
| 131 |
-
Reviewer Access:
|
| 132 |
-
Because this project is under double-blind peer review, the official Zenodo record remains an unpublished draft. The DOI (10.5281/zenodo.19951359) is reserved in the metadata but will return a 404 until de-anonymization.
|
| 133 |
|
| 134 |
-
To download the exact cb_telemetry_v1_expert_aligned.tar.gz archive, please use the anonymous preview link:
|
| 135 |
-
https://zenodo.org/records/19951359?preview=1&token=eyJhbGciOiJIUzUxMiJ9.eyJpZCI6IjgyMTMzMjc2LTQ1ZGYtNDM5MC1hZWQ0LTA0MDUwN2U5NTIzYyIsImRhdGEiOnt9LCJyYW5kb20iOiJjYjUxMzE1NDMwM2IxMDYzMjU3MTAzYzVhOTk0M2NjNCJ9.vFOFfG_HYwC1BDsesFQzwGOevxXbin5FaJ7y4nutnam61ch0HoQEdR4Jl8U4UnTDZA9ghw2ps2HOOfkdJx9z2g
|
| 136 |
## Provenance and Licensing
|
| 137 |
|
| 138 |
The combined CB-Telemetry package is distributed under CC BY-NC-SA 4.0.
|
|
@@ -172,8 +178,7 @@ breeding-season-centered protocol.
|
|
| 172 |
|
| 173 |
```text
|
| 174 |
annotations/ normalized J-STAGE expert listening records
|
| 175 |
-
audio_sample/ 24 short M4A clips plus metadata
|
| 176 |
-
audio_sample_wav/ 24 short WAV clips for the Hugging Face Dataset Viewer
|
| 177 |
baselines/ refreshed public baseline outputs
|
| 178 |
features/ frozen Default and Strict-Clean feature tables
|
| 179 |
manifests/ scored snapshot, split, audio, and sample manifests
|
|
|
|
| 21 |
- n<1K
|
| 22 |
configs:
|
| 23 |
- config_name: reviewer_audio
|
| 24 |
+
data_dir: audio_sample
|
| 25 |
data_files:
|
| 26 |
- split: train
|
| 27 |
+
path: "**/*.m4a"
|
|
|
|
| 28 |
default: true
|
| 29 |
---
|
| 30 |
|
|
|
|
| 37 |
|
| 38 |
[https://doi.org/10.5281/zenodo.19951359](https://doi.org/10.5281/zenodo.19951359)
|
| 39 |
|
| 40 |
+
**Note for reviewers:** The permanent DOI (`10.5281/zenodo.19951359`)
|
| 41 |
+
is currently reserved and will resolve upon camera-ready publication. During
|
| 42 |
+
the double-blind review phase, please access the frozen draft archive using
|
| 43 |
+
this anonymized secret link:
|
| 44 |
+
[https://zenodo.org/records/19951359?preview=1&token=eyJhbGciOiJIUzUxMiJ9.eyJpZCI6IjgyMTMzMjc2LTQ1ZGYtNDM5MC1hZWQ0LTA0MDUwN2U5NTIzYyIsImRhdGEiOnt9LCJyYW5kb20iOiJjYjUxMzE1NDMwM2IxMDYzMjU3MTAzYzVhOTk0M2NjNCJ9.vFOFfG_HYwC1BDsesFQzwGOevxXbin5FaJ7y4nutnam61ch0HoQEdR4Jl8U4UnTDZA9ghw2ps2HOOfkdJx9z2g](https://zenodo.org/records/19951359?preview=1&token=eyJhbGciOiJIUzUxMiJ9.eyJpZCI6IjgyMTMzMjc2LTQ1ZGYtNDM5MC1hZWQ0LTA0MDUwN2U5NTIzYyIsImRhdGEiOnt9LCJyYW5kb20iOiJjYjUxMzE1NDMwM2IxMDYzMjU3MTAzYzVhOTk0M2NjNCJ9.vFOFfG_HYwC1BDsesFQzwGOevxXbin5FaJ7y4nutnam61ch0HoQEdR4Jl8U4UnTDZA9ghw2ps2HOOfkdJx9z2g).
|
| 45 |
|
| 46 |
## What Is Hosted Here
|
| 47 |
|
|
|
|
| 51 |
- frozen v1 scored manifests with 805 expert-overlap recording rows;
|
| 52 |
- paired Default and Strict-Clean split metadata;
|
| 53 |
- frozen Default and Strict-Clean feature tables;
|
| 54 |
+
- released RVQ/PQ/OPQ bottleneck feature tables for metric recomputation;
|
| 55 |
- refreshed representation, retrieval, metadata-control, and bottleneck
|
| 56 |
baseline outputs;
|
| 57 |
- validation and smoke-test scripts;
|
|
|
|
| 69 |
## Reviewer Audio Viewer
|
| 70 |
|
| 71 |
The default Hugging Face viewer is configured for the reviewer audio sample in
|
| 72 |
+
`audio_sample/`. The same files can be loaded locally with:
|
|
|
|
|
|
|
| 73 |
|
| 74 |
```python
|
| 75 |
from datasets import load_dataset
|
| 76 |
|
| 77 |
+
samples = load_dataset("audiofolder", data_dir="audio_sample")
|
| 78 |
print(samples)
|
| 79 |
```
|
| 80 |
|
| 81 |
The root-level `metadata.csv` points to the same 24 clips with paths relative to
|
| 82 |
+
the repository root. The `audio_sample/metadata.csv` file points to the clips
|
| 83 |
+
relative to `audio_sample/`, following the AudioFolder convention.
|
| 84 |
|
| 85 |
## Benchmark Package Quickstart
|
| 86 |
|
|
|
|
| 91 |
python3 scripts/validate_cb_telemetry.py --root . --write-report
|
| 92 |
```
|
| 93 |
|
| 94 |
+
The companion code release can recompute the released retrieval and
|
| 95 |
+
shortcut-control tables from this repository with:
|
| 96 |
+
|
| 97 |
+
```bash
|
| 98 |
+
python3 -m pip install -r requirements.txt
|
| 99 |
+
python3 scripts/run_release_evaluation.py --root . --bootstrap-samples 2000
|
| 100 |
+
```
|
| 101 |
+
|
| 102 |
+
The recomputation uses the continuous feature tables and the released
|
| 103 |
+
source-frozen bottleneck feature tables under `features/bottlenecks/`.
|
| 104 |
+
|
| 105 |
To reconstruct the complete Default source-audio set:
|
| 106 |
|
| 107 |
```bash
|
|
|
|
| 130 |
`archive/`:
|
| 131 |
|
| 132 |
```text
|
| 133 |
+
701ad48d75ff34fabce82ee27271579f2ad83345a74a13468891beb0c6dc8f36 cb_telemetry_v1_expert_aligned.tar.gz
|
| 134 |
```
|
| 135 |
|
| 136 |
The Zenodo DOI for the immutable v1 artifact is:
|
|
|
|
| 138 |
```text
|
| 139 |
10.5281/zenodo.19951359
|
| 140 |
```
|
|
|
|
|
|
|
| 141 |
|
|
|
|
|
|
|
| 142 |
## Provenance and Licensing
|
| 143 |
|
| 144 |
The combined CB-Telemetry package is distributed under CC BY-NC-SA 4.0.
|
|
|
|
| 178 |
|
| 179 |
```text
|
| 180 |
annotations/ normalized J-STAGE expert listening records
|
| 181 |
+
audio_sample/ 24 short M4A clips plus AudioFolder metadata
|
|
|
|
| 182 |
baselines/ refreshed public baseline outputs
|
| 183 |
features/ frozen Default and Strict-Clean feature tables
|
| 184 |
manifests/ scored snapshot, split, audio, and sample manifests
|
archive/cb_telemetry_v1_expert_aligned.tar.gz
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:701ad48d75ff34fabce82ee27271579f2ad83345a74a13468891beb0c6dc8f36
|
| 3 |
+
size 48368883
|
archive/cb_telemetry_v1_expert_aligned.tar.gz.sha256
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 104
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4e7027fda93a0d3e5aabcc87de8aab7a7e9857aaa6cd980a2ebe1e9ef110f896
|
| 3 |
size 104
|
audio_sample/yamanakako/20140402/cbt_yamanakako_old_20140402_20140402060501yamanakako_clip.m4a
CHANGED
|
Binary files a/audio_sample/yamanakako/20140402/cbt_yamanakako_old_20140402_20140402060501yamanakako_clip.m4a and b/audio_sample/yamanakako/20140402/cbt_yamanakako_old_20140402_20140402060501yamanakako_clip.m4a differ
|
|
|
audio_sample_wav/README.md
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
# Reviewer Audio Sample WAV Split
|
| 2 |
-
|
| 3 |
-
This directory contains 16 kHz mono WAV transcodes of the 24 reviewer audio clips for the Hugging Face Dataset Viewer. The original M4A clips are retained in `audio_sample/`.
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/chichibu/20120407/cbt_tetto_20120407_20120407053414tetto_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:03acaad52ada9d573ad138d80eb31a840368a129c1e9c99db8577b8d91f47dd4
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/chichibu/20120407/cbt_tetto_20120407_20120407063415tetto_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:ee8cfe8cc6c00d2bc4c158c22ce0c926be68174501ca5a3a4a6bcaf23edab987
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/chichibu/20120410/cbt_tetto_20120410_20120410053319tetto_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:033e47eda01f404b5e1e95c9d98887a05c919c1ac0f9338ae2518a1745fd7b0c
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/chichibu/20120410/cbt_tetto_20120410_20120410063320tetto_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:ada0bc4fda0c5c6363f8028805e877654b505c629f4d3d987a23ce2ab77bfa15
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/chichibu/20120413/cbt_tetto_20120413_20120413050431tetto_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:9cf2f01df553cabf4d7b157f0707dd1aa49c23b55c59c73264945cc2c08580f8
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/chichibu/20120413/cbt_tetto_20120413_20120413060432tetto_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:c4aeefe5e0dfd3e67a973928763ad62dd9a4b16b46fdb9f789e5e40034dff8db
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/chichibu/20120416/cbt_tetto_20120416_20120416070400tetto_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:c96d7bfc27fca1f63345eea280d5371f36322027bf1dd17c676460f335128329
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/chichibu/20120425/cbt_tetto_20120425_20120425050327tetto_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:217413d30c3321414fea31f8f4c282d33ae162d42dba66d5df8fe257fbed5b8e
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/furano/20200403/cbt_maeyama_20200403_20200403053159maeyama_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:510427186d020db8cb7e13d2d5bb32f90067f0de3a8432546b445ff2ace48608
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/furano/20200411/cbt_maeyama_20200411_20200411053201maeyama_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:5e9ff140472ae190804b71e7c4cf58321a0b7f5aa957c7743af7a957ee87ed1b
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/furano/20200417/cbt_maeyama_20200417_20200417043209maeyama_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:4602f05bc6ebbc45eb458b7dad37429243353c2a740d8d63dca9ae04c18386fd
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/metadata.csv
DELETED
|
@@ -1,25 +0,0 @@
|
|
| 1 |
-
file_name,recording_id,site_en,date,year,behavior_types,jstage_event_count,jstage_species_count,clip_start_sec,clip_duration_sec,segment_start_sec,segment_end_sec,source_audio_license,source_annotation_doi
|
| 2 |
-
chichibu/20120425/cbt_tetto_20120425_20120425050327tetto_clip.wav,cbt_tetto_20120425_20120425050327tetto,Chichibu,20120425,2012,C|D|S,250,16,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 3 |
-
furano/20200403/cbt_maeyama_20200403_20200403053159maeyama_clip.wav,cbt_maeyama_20200403_20200403053159maeyama,Furano,20200403,2020,C|D|S,70,10,2370.388,8.0,2373.138,2375.638,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 4 |
-
shiga/20200403/cbt_otanomo_20200403_20200403061257otanomo_clip.wav,cbt_otanomo_20200403_20200403061257otanomo,Shiga,20200403,2020,C|D|S,83,10,775.312,8.0,779.026,779.598,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 5 |
-
yamanakako/20150402/cbt_yamanakako_old_20150402_20150402055817yamanakako_clip.wav,cbt_yamanakako_old_20150402_20150402055817yamanakako,Yamanakako,20150402,2015,C|D|S,208,17,2164.752,8.0,2167.502,2170.002,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 6 |
-
yatake/20120430/cbt_yatake_old_20120430_20120430045616yatake_clip.wav,cbt_yatake_old_20120430_20120430045616yatake,Yatake,20120430,2012,C|D|S,224,20,3100.528,8.0,3104.162,3104.894,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 7 |
-
chichibu/20120401/cbt_tetto_20120401_20120401053323tetto_clip.wav,cbt_tetto_20120401_20120401053323tetto,Chichibu,20120401,2012,C|S,69,7,361.476,8.0,364.226,366.726,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 8 |
-
furano/20200411/cbt_maeyama_20200411_20200411053201maeyama_clip.wav,cbt_maeyama_20200411_20200411053201maeyama,Furano,20200411,2020,C|S,91,10,2059.272,8.0,2062.022,2064.522,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 9 |
-
shiga/20200403/cbt_otanomo_20200403_20200403051249otanomo_clip.wav,cbt_otanomo_20200403_20200403051249otanomo,Shiga,20200403,2020,C,2,1,540.144,8.0,542.894,545.394,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 10 |
-
yamanakako/20140402/cbt_yamanakako_old_20140402_20140402060501yamanakako_clip.wav,cbt_yamanakako_old_20140402_20140402060501yamanakako,Yamanakako,20140402,2014,C|S,15,8,104.75,8.0,107.5,110.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 11 |
-
yatake/20120403/cbt_yatake_old_20120403_20120403061005yatake_clip.wav,cbt_yatake_old_20120403_20120403061005yatake,Yatake,20120403,2012,C|S,43,6,702.25,8.0,705.0,707.5,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 12 |
-
chichibu/20120416/cbt_tetto_20120416_20120416070400tetto_clip.wav,cbt_tetto_20120416_20120416070400tetto,Chichibu,20120416,2012,S,2,2,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 13 |
-
furano/20200417/cbt_maeyama_20200417_20200417043209maeyama_clip.wav,cbt_maeyama_20200417_20200417043209maeyama,Furano,20200417,2020,S,19,4,1669.752,8.0,1672.502,1675.002,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 14 |
-
shiga/20200620/cbt_otanomo_20200620_20200620040430otanomo_clip.wav,cbt_otanomo_20200620_20200620040430otanomo,Shiga,20200620,2020,S,4,2,2895.72,8.0,2899.634,2899.806,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 15 |
-
yamanakako/20160628/cbt_yamanakako_old_20160628_20160628054404yamanakako_clip.wav,cbt_yamanakako_old_20160628_20160628054404yamanakako,Yamanakako,20160628,2016,S,27,1,3119.75,8.0,3122.5,3125.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 16 |
-
yatake/20140611/cbt_yatake_old_20140611_20140611040745yatake_clip.wav,cbt_yatake_old_20140611_20140611040745yatake,Yatake,20140611,2014,S,24,7,162.25,8.0,165.0,167.5,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 17 |
-
chichibu/20120401/cbt_tetto_20120401_20120401063324tetto_clip.wav,cbt_tetto_20120401_20120401063324tetto,Chichibu,20120401,2012,C|S,176,12,1477.25,8.0,1480.0,1482.5,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 18 |
-
chichibu/20120404/cbt_tetto_20120404_20120404053332tetto_clip.wav,cbt_tetto_20120404_20120404053332tetto,Chichibu,20120404,2012,C|S,100,10,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 19 |
-
chichibu/20120404/cbt_tetto_20120404_20120404063333tetto_clip.wav,cbt_tetto_20120404_20120404063333tetto,Chichibu,20120404,2012,C|S,113,11,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 20 |
-
chichibu/20120407/cbt_tetto_20120407_20120407053414tetto_clip.wav,cbt_tetto_20120407_20120407053414tetto,Chichibu,20120407,2012,C|S,109,9,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 21 |
-
chichibu/20120407/cbt_tetto_20120407_20120407063415tetto_clip.wav,cbt_tetto_20120407_20120407063415tetto,Chichibu,20120407,2012,C|S,100,12,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 22 |
-
chichibu/20120410/cbt_tetto_20120410_20120410053319tetto_clip.wav,cbt_tetto_20120410_20120410053319tetto,Chichibu,20120410,2012,C|S,24,7,322.25,8.0,325.0,327.5,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 23 |
-
chichibu/20120410/cbt_tetto_20120410_20120410063320tetto_clip.wav,cbt_tetto_20120410_20120410063320tetto,Chichibu,20120410,2012,C|S,115,12,1294.75,8.0,1297.5,1300.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 24 |
-
chichibu/20120413/cbt_tetto_20120413_20120413050431tetto_clip.wav,cbt_tetto_20120413_20120413050431tetto,Chichibu,20120413,2012,C|S,103,11,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 25 |
-
chichibu/20120413/cbt_tetto_20120413_20120413060432tetto_clip.wav,cbt_tetto_20120413_20120413060432tetto,Chichibu,20120413,2012,C|S,236,16,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/shiga/20200403/cbt_otanomo_20200403_20200403051249otanomo_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:5b6e6f8b94590776c6be3d627e18aca6f7b75e637a48860ecb45f96fab76d693
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/shiga/20200403/cbt_otanomo_20200403_20200403061257otanomo_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:01afa600dd7172988d57ae8d5f63820d6058487d9c1d31f7ccac1e8563a3c5ad
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/shiga/20200620/cbt_otanomo_20200620_20200620040430otanomo_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:ba8c64f4128c8bb093986ddab150de120cc49662a3ba3234b738d1b718a32c2f
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/yamanakako/20140402/cbt_yamanakako_old_20140402_20140402060501yamanakako_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:f79c075a2ea3b55edc83c3781ac187dc21f903701c3ee3670a687796a71807e7
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/yamanakako/20150402/cbt_yamanakako_old_20150402_20150402055817yamanakako_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:6a1efb4c8bf7168732577749ade21fd2da5bac258289884aacea65ec48b385b4
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/yamanakako/20160628/cbt_yamanakako_old_20160628_20160628054404yamanakako_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:4516bf37ab15384f602c460dd3792b48dd3f45f6d318979aeed99e2545bd11ab
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/yatake/20120403/cbt_yatake_old_20120403_20120403061005yatake_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:85ed5b7466698d5d2c5fc94df52a7f97afd9feca239384598b71429f0edb2c88
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/yatake/20120430/cbt_yatake_old_20120430_20120430045616yatake_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:606b3f6649c15cb580c5a077a30bfc5c64fb82d2e57294b3286c336b44af185a
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/yatake/20140611/cbt_yatake_old_20140611_20140611040745yatake_clip.wav
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:757dbb540ace49fc8c20587e12b8b8eb63f88d065e7e8ca8d910aa05b917f38b
|
| 3 |
-
size 260096
|
|
|
|
|
|
|
|
|
|
|
|
audio_sample_wav/chichibu/20120404/cbt_tetto_20120404_20120404063333tetto_clip.wav → features/bottlenecks/default/opq_8bit_feature_table.csv.gz
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c3359d5af297ab7cee9fab04aa0a9327e0f6370dcb89322d6ae04bbda6f1f265
|
| 3 |
+
size 4456952
|
audio_sample_wav/chichibu/20120401/cbt_tetto_20120401_20120401053323tetto_clip.wav → features/bottlenecks/default/pq_8bit_feature_table.csv.gz
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:57ed5dbe3466c44db6b9b64d25401666d80c97ad875cdc5cf510a4e9e2007e36
|
| 3 |
+
size 3474233
|
audio_sample_wav/chichibu/20120401/cbt_tetto_20120401_20120401063324tetto_clip.wav → features/bottlenecks/default/standard_rvq_8bit_feature_table.csv.gz
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:47cb104bcdfe52a3b3117777d7bf81519cf21945d0c6e8aaa4459f81a0ec4d5b
|
| 3 |
+
size 4684598
|
audio_sample_wav/chichibu/20120404/cbt_tetto_20120404_20120404053332tetto_clip.wav → features/bottlenecks/strict_clean/opq_8bit_feature_table.csv.gz
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c1dffda743db376217bd27cdd394a25bb04c5b2634ff96c1d779676afae5fb05
|
| 3 |
+
size 2484906
|
features/bottlenecks/strict_clean/pq_8bit_feature_table.csv.gz
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:22c3cbd4c1e19dd145aa89d17a32a955a259a257297403220c79065204a702eb
|
| 3 |
+
size 1908955
|
features/bottlenecks/strict_clean/standard_rvq_8bit_feature_table.csv.gz
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c41b83e8fabe004886e55e05451882514b79292112b1e026e9e3e35dd92fdb21
|
| 3 |
+
size 2760058
|
hf_staging_validation.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"status": "pass",
|
| 3 |
-
"root": "
|
| 4 |
"reviewer_audio_rows": 24,
|
| 5 |
"reviewer_audio_sites": [
|
| 6 |
"Chichibu",
|
|
@@ -15,9 +15,9 @@
|
|
| 15 |
"S"
|
| 16 |
],
|
| 17 |
"zenodo_doi": "10.5281/zenodo.19951359",
|
| 18 |
-
"archive_sha256": "
|
| 19 |
"archive_stats": {
|
| 20 |
-
"members":
|
| 21 |
"ds_store": 0,
|
| 22 |
"macosx": 0
|
| 23 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"status": "pass",
|
| 3 |
+
"root": "hf_cb_telemetry_v1",
|
| 4 |
"reviewer_audio_rows": 24,
|
| 5 |
"reviewer_audio_sites": [
|
| 6 |
"Chichibu",
|
|
|
|
| 15 |
"S"
|
| 16 |
],
|
| 17 |
"zenodo_doi": "10.5281/zenodo.19951359",
|
| 18 |
+
"archive_sha256": "701ad48d75ff34fabce82ee27271579f2ad83345a74a13468891beb0c6dc8f36",
|
| 19 |
"archive_stats": {
|
| 20 |
+
"members": 99,
|
| 21 |
"ds_store": 0,
|
| 22 |
"macosx": 0
|
| 23 |
}
|
metadata.csv
CHANGED
|
@@ -1,25 +1,25 @@
|
|
| 1 |
file_name,recording_id,site_en,date,year,behavior_types,jstage_event_count,jstage_species_count,clip_start_sec,clip_duration_sec,segment_start_sec,segment_end_sec,source_audio_license,source_annotation_doi
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
|
|
|
| 1 |
file_name,recording_id,site_en,date,year,behavior_types,jstage_event_count,jstage_species_count,clip_start_sec,clip_duration_sec,segment_start_sec,segment_end_sec,source_audio_license,source_annotation_doi
|
| 2 |
+
audio_sample/chichibu/20120425/cbt_tetto_20120425_20120425050327tetto_clip.m4a,cbt_tetto_20120425_20120425050327tetto,Chichibu,20120425,2012,C|D|S,250,16,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 3 |
+
audio_sample/furano/20200403/cbt_maeyama_20200403_20200403053159maeyama_clip.m4a,cbt_maeyama_20200403_20200403053159maeyama,Furano,20200403,2020,C|D|S,70,10,2370.388,8.0,2373.138,2375.638,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 4 |
+
audio_sample/shiga/20200403/cbt_otanomo_20200403_20200403061257otanomo_clip.m4a,cbt_otanomo_20200403_20200403061257otanomo,Shiga,20200403,2020,C|D|S,83,10,775.312,8.0,779.026,779.598,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 5 |
+
audio_sample/yamanakako/20150402/cbt_yamanakako_old_20150402_20150402055817yamanakako_clip.m4a,cbt_yamanakako_old_20150402_20150402055817yamanakako,Yamanakako,20150402,2015,C|D|S,208,17,2164.752,8.0,2167.502,2170.002,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 6 |
+
audio_sample/yatake/20120430/cbt_yatake_old_20120430_20120430045616yatake_clip.m4a,cbt_yatake_old_20120430_20120430045616yatake,Yatake,20120430,2012,C|D|S,224,20,3100.528,8.0,3104.162,3104.894,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 7 |
+
audio_sample/chichibu/20120401/cbt_tetto_20120401_20120401053323tetto_clip.m4a,cbt_tetto_20120401_20120401053323tetto,Chichibu,20120401,2012,C|S,69,7,361.476,8.0,364.226,366.726,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 8 |
+
audio_sample/furano/20200411/cbt_maeyama_20200411_20200411053201maeyama_clip.m4a,cbt_maeyama_20200411_20200411053201maeyama,Furano,20200411,2020,C|S,91,10,2059.272,8.0,2062.022,2064.522,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 9 |
+
audio_sample/shiga/20200403/cbt_otanomo_20200403_20200403051249otanomo_clip.m4a,cbt_otanomo_20200403_20200403051249otanomo,Shiga,20200403,2020,C,2,1,540.144,8.0,542.894,545.394,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 10 |
+
audio_sample/yamanakako/20140402/cbt_yamanakako_old_20140402_20140402060501yamanakako_clip.m4a,cbt_yamanakako_old_20140402_20140402060501yamanakako,Yamanakako,20140402,2014,C|S,15,8,104.75,8.0,107.5,110.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 11 |
+
audio_sample/yatake/20120403/cbt_yatake_old_20120403_20120403061005yatake_clip.m4a,cbt_yatake_old_20120403_20120403061005yatake,Yatake,20120403,2012,C|S,43,6,702.25,8.0,705.0,707.5,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 12 |
+
audio_sample/chichibu/20120416/cbt_tetto_20120416_20120416070400tetto_clip.m4a,cbt_tetto_20120416_20120416070400tetto,Chichibu,20120416,2012,S,2,2,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 13 |
+
audio_sample/furano/20200417/cbt_maeyama_20200417_20200417043209maeyama_clip.m4a,cbt_maeyama_20200417_20200417043209maeyama,Furano,20200417,2020,S,19,4,1669.752,8.0,1672.502,1675.002,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 14 |
+
audio_sample/shiga/20200620/cbt_otanomo_20200620_20200620040430otanomo_clip.m4a,cbt_otanomo_20200620_20200620040430otanomo,Shiga,20200620,2020,S,4,2,2895.72,8.0,2899.634,2899.806,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 15 |
+
audio_sample/yamanakako/20160628/cbt_yamanakako_old_20160628_20160628054404yamanakako_clip.m4a,cbt_yamanakako_old_20160628_20160628054404yamanakako,Yamanakako,20160628,2016,S,27,1,3119.75,8.0,3122.5,3125.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 16 |
+
audio_sample/yatake/20140611/cbt_yatake_old_20140611_20140611040745yatake_clip.m4a,cbt_yatake_old_20140611_20140611040745yatake,Yatake,20140611,2014,S,24,7,162.25,8.0,165.0,167.5,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 17 |
+
audio_sample/chichibu/20120401/cbt_tetto_20120401_20120401063324tetto_clip.m4a,cbt_tetto_20120401_20120401063324tetto,Chichibu,20120401,2012,C|S,176,12,1477.25,8.0,1480.0,1482.5,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 18 |
+
audio_sample/chichibu/20120404/cbt_tetto_20120404_20120404053332tetto_clip.m4a,cbt_tetto_20120404_20120404053332tetto,Chichibu,20120404,2012,C|S,100,10,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 19 |
+
audio_sample/chichibu/20120404/cbt_tetto_20120404_20120404063333tetto_clip.m4a,cbt_tetto_20120404_20120404063333tetto,Chichibu,20120404,2012,C|S,113,11,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 20 |
+
audio_sample/chichibu/20120407/cbt_tetto_20120407_20120407053414tetto_clip.m4a,cbt_tetto_20120407_20120407053414tetto,Chichibu,20120407,2012,C|S,109,9,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 21 |
+
audio_sample/chichibu/20120407/cbt_tetto_20120407_20120407063415tetto_clip.m4a,cbt_tetto_20120407_20120407063415tetto,Chichibu,20120407,2012,C|S,100,12,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 22 |
+
audio_sample/chichibu/20120410/cbt_tetto_20120410_20120410053319tetto_clip.m4a,cbt_tetto_20120410_20120410053319tetto,Chichibu,20120410,2012,C|S,24,7,322.25,8.0,325.0,327.5,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 23 |
+
audio_sample/chichibu/20120410/cbt_tetto_20120410_20120410063320tetto_clip.m4a,cbt_tetto_20120410_20120410063320tetto,Chichibu,20120410,2012,C|S,115,12,1294.75,8.0,1297.5,1300.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 24 |
+
audio_sample/chichibu/20120413/cbt_tetto_20120413_20120413050431tetto_clip.m4a,cbt_tetto_20120413_20120413050431tetto,Chichibu,20120413,2012,C|S,103,11,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
| 25 |
+
audio_sample/chichibu/20120413/cbt_tetto_20120413_20120413060432tetto_clip.m4a,cbt_tetto_20120413_20120413060432tetto,Chichibu,20120413,2012,C|S,236,16,0.0,8.0,0.0,8.0,CC BY-NC-SA 4.0,10.57368/data.birdresearch.26499895
|
requirements.txt
ADDED
|
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Python 3.9 or newer is recommended.
|
| 2 |
+
numpy>=1.23
|
| 3 |
+
pandas>=1.5
|
| 4 |
+
scikit-learn>=1.2
|
scripts/run_release_evaluation.py
ADDED
|
@@ -0,0 +1,366 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
from __future__ import annotations
|
| 3 |
+
|
| 4 |
+
import argparse
|
| 5 |
+
import csv
|
| 6 |
+
import json
|
| 7 |
+
import subprocess
|
| 8 |
+
import sys
|
| 9 |
+
from pathlib import Path
|
| 10 |
+
from typing import Any
|
| 11 |
+
|
| 12 |
+
import pandas as pd
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
DISPLAY_NAME = {
|
| 16 |
+
"continuous": "Continuous reference",
|
| 17 |
+
"standard_rvq_8bit": "standard RVQ 8-bit",
|
| 18 |
+
"pq_8bit": "PQ 8-bit",
|
| 19 |
+
"opq_8bit": "OPQ 8-bit",
|
| 20 |
+
"metadata_basic": "Metadata basic",
|
| 21 |
+
"metadata_calendar": "Metadata calendar",
|
| 22 |
+
"random_permuted_continuous": "Random-permuted continuous",
|
| 23 |
+
}
|
| 24 |
+
|
| 25 |
+
PUBLIC_METHODS = [
|
| 26 |
+
"continuous",
|
| 27 |
+
"standard_rvq_8bit",
|
| 28 |
+
"pq_8bit",
|
| 29 |
+
"opq_8bit",
|
| 30 |
+
"metadata_basic",
|
| 31 |
+
"metadata_calendar",
|
| 32 |
+
"random_permuted_continuous",
|
| 33 |
+
]
|
| 34 |
+
|
| 35 |
+
|
| 36 |
+
def parse_args() -> argparse.Namespace:
|
| 37 |
+
parser = argparse.ArgumentParser(
|
| 38 |
+
description="Recompute the released CB-Telemetry retrieval and shortcut-control tables."
|
| 39 |
+
)
|
| 40 |
+
parser.add_argument("--root", default=".", help="CB-Telemetry dataset root.")
|
| 41 |
+
parser.add_argument("--output-dir", default="evaluation_runs/release_retrieval", help="Output directory under --root.")
|
| 42 |
+
parser.add_argument("--bootstrap-samples", type=int, default=2000, help="Bootstrap resamples.")
|
| 43 |
+
parser.add_argument("--bootstrap-seed", type=int, default=42, help="Bootstrap random seed.")
|
| 44 |
+
parser.add_argument("--random-seed", type=int, default=42, help="Random-permuted control seed.")
|
| 45 |
+
parser.add_argument("--include-gap2", action="store_true", help="Also run the supplemental gap=2 retrieval controls.")
|
| 46 |
+
parser.add_argument("--tolerance", type=float, default=5e-4, help="Rounded-table comparison tolerance.")
|
| 47 |
+
return parser.parse_args()
|
| 48 |
+
|
| 49 |
+
|
| 50 |
+
def resolve(root: Path, raw_path: str) -> Path:
|
| 51 |
+
path = Path(raw_path).expanduser()
|
| 52 |
+
if not path.is_absolute():
|
| 53 |
+
path = (root / path).resolve()
|
| 54 |
+
return path
|
| 55 |
+
|
| 56 |
+
|
| 57 |
+
def release_path(root: Path, path: Path) -> str:
|
| 58 |
+
try:
|
| 59 |
+
return path.resolve().relative_to(root).as_posix()
|
| 60 |
+
except ValueError:
|
| 61 |
+
return path.as_posix()
|
| 62 |
+
|
| 63 |
+
|
| 64 |
+
def track_feature_table(track: str) -> str:
|
| 65 |
+
return "features/feature_table_default.csv.gz" if track == "default" else "features/feature_table_strict_clean.csv.gz"
|
| 66 |
+
|
| 67 |
+
|
| 68 |
+
def bottleneck_representations(root: Path, track: str) -> list[str]:
|
| 69 |
+
result = []
|
| 70 |
+
for method in ["standard_rvq_8bit", "pq_8bit", "opq_8bit"]:
|
| 71 |
+
rel = f"features/bottlenecks/{track}/{method}_feature_table.csv.gz"
|
| 72 |
+
if (root / rel).exists():
|
| 73 |
+
result.append(f"{method}={rel}")
|
| 74 |
+
return result
|
| 75 |
+
|
| 76 |
+
|
| 77 |
+
def run_command(cmd: list[str]) -> None:
|
| 78 |
+
subprocess.run(cmd, check=True)
|
| 79 |
+
|
| 80 |
+
|
| 81 |
+
def run_retrieval_suite(
|
| 82 |
+
root: Path,
|
| 83 |
+
output_dir: Path,
|
| 84 |
+
bootstrap_samples: int,
|
| 85 |
+
bootstrap_seed: int,
|
| 86 |
+
random_seed: int,
|
| 87 |
+
include_gap2: bool,
|
| 88 |
+
) -> list[dict[str, Any]]:
|
| 89 |
+
script = Path(__file__).resolve().parent / "run_retrieval_eval.py"
|
| 90 |
+
run_rows: list[dict[str, Any]] = []
|
| 91 |
+
gaps = [1, 2] if include_gap2 else [1]
|
| 92 |
+
for track in ["default", "strict_clean"]:
|
| 93 |
+
feature_table = track_feature_table(track)
|
| 94 |
+
representations = [f"continuous={feature_table}", *bottleneck_representations(root, track)]
|
| 95 |
+
for scope_label, archive_scope in [("global", "global"), ("same-archive", "same_archive_only")]:
|
| 96 |
+
for gap in gaps:
|
| 97 |
+
run_dir = output_dir / track / f"{archive_scope}_gap{gap}"
|
| 98 |
+
cmd = [
|
| 99 |
+
sys.executable,
|
| 100 |
+
script.as_posix(),
|
| 101 |
+
"--root",
|
| 102 |
+
root.as_posix(),
|
| 103 |
+
"--feature-table",
|
| 104 |
+
feature_table,
|
| 105 |
+
"--output-dir",
|
| 106 |
+
run_dir.as_posix(),
|
| 107 |
+
"--archive-scope",
|
| 108 |
+
archive_scope,
|
| 109 |
+
"--max-slot-gap",
|
| 110 |
+
str(gap),
|
| 111 |
+
"--bootstrap-samples",
|
| 112 |
+
str(bootstrap_samples),
|
| 113 |
+
"--bootstrap-seed",
|
| 114 |
+
str(bootstrap_seed),
|
| 115 |
+
"--random-seed",
|
| 116 |
+
str(random_seed),
|
| 117 |
+
"--include-metadata-controls",
|
| 118 |
+
]
|
| 119 |
+
for item in representations:
|
| 120 |
+
cmd.extend(["--representation", item])
|
| 121 |
+
run_command(cmd)
|
| 122 |
+
run_rows.append(
|
| 123 |
+
{
|
| 124 |
+
"track": track,
|
| 125 |
+
"scope": scope_label,
|
| 126 |
+
"gap": gap,
|
| 127 |
+
"archive_scope": archive_scope,
|
| 128 |
+
"summary_json_path": (run_dir / "summary.json").as_posix(),
|
| 129 |
+
"summary_json": release_path(root, run_dir / "summary.json"),
|
| 130 |
+
}
|
| 131 |
+
)
|
| 132 |
+
return run_rows
|
| 133 |
+
|
| 134 |
+
|
| 135 |
+
def load_json(path: Path) -> dict[str, Any]:
|
| 136 |
+
return json.loads(path.read_text(encoding="utf-8"))
|
| 137 |
+
|
| 138 |
+
|
| 139 |
+
def build_overview(run_rows: list[dict[str, Any]]) -> pd.DataFrame:
|
| 140 |
+
rows: list[dict[str, Any]] = []
|
| 141 |
+
for run in run_rows:
|
| 142 |
+
summary = load_json(Path(run["summary_json_path"]))
|
| 143 |
+
aggregate_df = pd.DataFrame(summary["aggregate_rows"]).set_index("method")
|
| 144 |
+
bootstrap_df = pd.DataFrame(summary["bootstrap_rows"])
|
| 145 |
+
for _, boot in bootstrap_df.iterrows():
|
| 146 |
+
method = str(boot["method"])
|
| 147 |
+
aggregate = aggregate_df.loc[method]
|
| 148 |
+
rows.append(
|
| 149 |
+
{
|
| 150 |
+
"track": run["track"],
|
| 151 |
+
"scope": run["scope"],
|
| 152 |
+
"gap": int(run["gap"]),
|
| 153 |
+
"archive_scope": run["archive_scope"],
|
| 154 |
+
"method": method,
|
| 155 |
+
"display_name": DISPLAY_NAME.get(method, method),
|
| 156 |
+
"bootstrap_unit": str(boot["bootstrap_unit"]),
|
| 157 |
+
"top1": float(aggregate["mean_top1_hit_rate"]),
|
| 158 |
+
"top1_ci_low": float(boot["top1_hit_rate_ci_low"]),
|
| 159 |
+
"top1_ci_high": float(boot["top1_hit_rate_ci_high"]),
|
| 160 |
+
"mrr": float(aggregate["mean_mrr"]),
|
| 161 |
+
"mrr_ci_low": float(boot["mrr_ci_low"]),
|
| 162 |
+
"mrr_ci_high": float(boot["mrr_ci_high"]),
|
| 163 |
+
"top5": float(aggregate["mean_top5_hit_rate"]),
|
| 164 |
+
"top5_ci_low": float(boot["top5_hit_rate_ci_low"]),
|
| 165 |
+
"top5_ci_high": float(boot["top5_hit_rate_ci_high"]),
|
| 166 |
+
"mean_candidate_size": float(aggregate["mean_candidate_size"]),
|
| 167 |
+
"mean_chance_top1": float(aggregate["mean_chance_top1"]),
|
| 168 |
+
"summary_json": str(run["summary_json"]),
|
| 169 |
+
}
|
| 170 |
+
)
|
| 171 |
+
return pd.DataFrame(rows).sort_values(["track", "scope", "gap", "bootstrap_unit", "method"]).reset_index(drop=True)
|
| 172 |
+
|
| 173 |
+
|
| 174 |
+
def date_row(df: pd.DataFrame, track: str, method: str, scope: str, gap: int) -> pd.Series:
|
| 175 |
+
rows = df[
|
| 176 |
+
(df["track"] == track)
|
| 177 |
+
& (df["method"] == method)
|
| 178 |
+
& (df["scope"] == scope)
|
| 179 |
+
& (df["gap"] == gap)
|
| 180 |
+
& (df["bootstrap_unit"] == "date")
|
| 181 |
+
]
|
| 182 |
+
if rows.empty:
|
| 183 |
+
raise KeyError(f"Missing row: track={track}, method={method}, scope={scope}, gap={gap}")
|
| 184 |
+
return rows.iloc[0]
|
| 185 |
+
|
| 186 |
+
|
| 187 |
+
def ci_text(row: pd.Series, metric: str) -> str:
|
| 188 |
+
return f"{float(row[metric]):.4f} [{float(row[f'{metric}_ci_low']):.4f}, {float(row[f'{metric}_ci_high']):.4f}]"
|
| 189 |
+
|
| 190 |
+
|
| 191 |
+
def build_table3_recomputed(overview: pd.DataFrame) -> pd.DataFrame:
|
| 192 |
+
rows = []
|
| 193 |
+
for method in PUBLIC_METHODS:
|
| 194 |
+
global_row = date_row(overview, "default", method, "global", 1)
|
| 195 |
+
same_row = date_row(overview, "default", method, "same-archive", 1)
|
| 196 |
+
rows.append(
|
| 197 |
+
{
|
| 198 |
+
"method": method,
|
| 199 |
+
"display_name": DISPLAY_NAME.get(method, method),
|
| 200 |
+
"global_top1": round(float(global_row["top1"]), 4),
|
| 201 |
+
"global_mrr": round(float(global_row["mrr"]), 4),
|
| 202 |
+
"global_top5": round(float(global_row["top5"]), 4),
|
| 203 |
+
"same_archive_top1": round(float(same_row["top1"]), 4),
|
| 204 |
+
"same_archive_mrr": round(float(same_row["mrr"]), 4),
|
| 205 |
+
"same_archive_top5": round(float(same_row["top5"]), 4),
|
| 206 |
+
"global_mrr_ci": ci_text(global_row, "mrr"),
|
| 207 |
+
"same_archive_mrr_ci": ci_text(same_row, "mrr"),
|
| 208 |
+
}
|
| 209 |
+
)
|
| 210 |
+
return pd.DataFrame(rows)
|
| 211 |
+
|
| 212 |
+
|
| 213 |
+
def build_dual_track_recomputed(root: Path, overview: pd.DataFrame) -> pd.DataFrame:
|
| 214 |
+
rows = []
|
| 215 |
+
for track in ["default", "strict_clean"]:
|
| 216 |
+
feature_df = pd.read_csv(root / track_feature_table(track), usecols=["date"])
|
| 217 |
+
for scope_label, archive_scope in [("global", "global"), ("same-archive", "same_archive_only")]:
|
| 218 |
+
row = date_row(overview, track, "continuous", scope_label, 1)
|
| 219 |
+
rows.append(
|
| 220 |
+
{
|
| 221 |
+
"track": f"{track}__{archive_scope}",
|
| 222 |
+
"subset": track,
|
| 223 |
+
"archive_scope": archive_scope,
|
| 224 |
+
"rows": int(feature_df.shape[0]),
|
| 225 |
+
"date_count": int(feature_df["date"].astype(str).nunique()),
|
| 226 |
+
"top1": round(float(row["top1"]), 4),
|
| 227 |
+
"mrr": round(float(row["mrr"]), 4),
|
| 228 |
+
"top5": round(float(row["top5"]), 4),
|
| 229 |
+
"chance_top1": round(float(row["mean_chance_top1"]), 4),
|
| 230 |
+
}
|
| 231 |
+
)
|
| 232 |
+
return pd.DataFrame(rows)
|
| 233 |
+
|
| 234 |
+
|
| 235 |
+
def compare_table(
|
| 236 |
+
expected_path: Path,
|
| 237 |
+
actual_df: pd.DataFrame,
|
| 238 |
+
key_columns: list[str],
|
| 239 |
+
metric_columns: list[str],
|
| 240 |
+
tolerance: float,
|
| 241 |
+
) -> list[dict[str, Any]]:
|
| 242 |
+
mismatches: list[dict[str, Any]] = []
|
| 243 |
+
if not expected_path.exists():
|
| 244 |
+
return [{"table": expected_path.name, "issue": "missing_expected_table"}]
|
| 245 |
+
expected_df = pd.read_csv(expected_path)
|
| 246 |
+
merged = expected_df.merge(actual_df, on=key_columns, how="outer", suffixes=("_expected", "_actual"), indicator=True)
|
| 247 |
+
for _, row in merged.iterrows():
|
| 248 |
+
key = {column: row[column] for column in key_columns}
|
| 249 |
+
if row["_merge"] != "both":
|
| 250 |
+
mismatches.append({"table": expected_path.name, "key": key, "issue": str(row["_merge"])})
|
| 251 |
+
continue
|
| 252 |
+
for column in metric_columns:
|
| 253 |
+
expected = float(row[f"{column}_expected"])
|
| 254 |
+
actual = float(row[f"{column}_actual"])
|
| 255 |
+
if abs(expected - actual) > tolerance:
|
| 256 |
+
mismatches.append(
|
| 257 |
+
{
|
| 258 |
+
"table": expected_path.name,
|
| 259 |
+
"key": key,
|
| 260 |
+
"metric": column,
|
| 261 |
+
"expected": expected,
|
| 262 |
+
"actual": actual,
|
| 263 |
+
"abs_delta": abs(expected - actual),
|
| 264 |
+
}
|
| 265 |
+
)
|
| 266 |
+
return mismatches
|
| 267 |
+
|
| 268 |
+
|
| 269 |
+
def write_markdown(path: Path, check: dict[str, Any]) -> None:
|
| 270 |
+
lines = [
|
| 271 |
+
"# CB-Telemetry Release Evaluation Check",
|
| 272 |
+
"",
|
| 273 |
+
f"- status: `{check['status']}`",
|
| 274 |
+
f"- bootstrap_samples: `{check['bootstrap_samples']}`",
|
| 275 |
+
f"- overview_rows: `{check['overview_rows']}`",
|
| 276 |
+
f"- mismatches: `{len(check['mismatches'])}`",
|
| 277 |
+
"",
|
| 278 |
+
"## Mismatches",
|
| 279 |
+
"",
|
| 280 |
+
]
|
| 281 |
+
if check["mismatches"]:
|
| 282 |
+
for item in check["mismatches"]:
|
| 283 |
+
lines.append(f"- `{item}`")
|
| 284 |
+
else:
|
| 285 |
+
lines.append("- none")
|
| 286 |
+
path.write_text("\n".join(lines) + "\n", encoding="utf-8")
|
| 287 |
+
|
| 288 |
+
|
| 289 |
+
def main() -> None:
|
| 290 |
+
args = parse_args()
|
| 291 |
+
root = Path(args.root).expanduser().resolve()
|
| 292 |
+
output_dir = resolve(root, args.output_dir)
|
| 293 |
+
output_dir.mkdir(parents=True, exist_ok=True)
|
| 294 |
+
|
| 295 |
+
run_rows = run_retrieval_suite(
|
| 296 |
+
root=root,
|
| 297 |
+
output_dir=output_dir,
|
| 298 |
+
bootstrap_samples=int(args.bootstrap_samples),
|
| 299 |
+
bootstrap_seed=int(args.bootstrap_seed),
|
| 300 |
+
random_seed=int(args.random_seed),
|
| 301 |
+
include_gap2=bool(args.include_gap2),
|
| 302 |
+
)
|
| 303 |
+
overview = build_overview(run_rows)
|
| 304 |
+
overview_path = output_dir / "release_retrieval_overview.csv"
|
| 305 |
+
overview.to_csv(overview_path, index=False)
|
| 306 |
+
|
| 307 |
+
table3 = build_table3_recomputed(overview)
|
| 308 |
+
table3_path = output_dir / "table_3_retrieval_baselines_and_controls_recomputed.csv"
|
| 309 |
+
table3.to_csv(table3_path, index=False)
|
| 310 |
+
|
| 311 |
+
dual = build_dual_track_recomputed(root, overview)
|
| 312 |
+
dual_path = output_dir / "table_3_dual_track_retrieval_recomputed.csv"
|
| 313 |
+
dual.to_csv(dual_path, index=False)
|
| 314 |
+
|
| 315 |
+
mismatches: list[dict[str, Any]] = []
|
| 316 |
+
mismatches.extend(
|
| 317 |
+
compare_table(
|
| 318 |
+
expected_path=root / "baselines" / "table_3_retrieval_baselines_and_controls.csv",
|
| 319 |
+
actual_df=table3,
|
| 320 |
+
key_columns=["method"],
|
| 321 |
+
metric_columns=[
|
| 322 |
+
"global_top1",
|
| 323 |
+
"global_mrr",
|
| 324 |
+
"global_top5",
|
| 325 |
+
"same_archive_top1",
|
| 326 |
+
"same_archive_mrr",
|
| 327 |
+
"same_archive_top5",
|
| 328 |
+
],
|
| 329 |
+
tolerance=float(args.tolerance),
|
| 330 |
+
)
|
| 331 |
+
)
|
| 332 |
+
mismatches.extend(
|
| 333 |
+
compare_table(
|
| 334 |
+
expected_path=root / "baselines" / "table_3_dual_track_retrieval.csv",
|
| 335 |
+
actual_df=dual,
|
| 336 |
+
key_columns=["track"],
|
| 337 |
+
metric_columns=["top1", "mrr", "top5", "chance_top1"],
|
| 338 |
+
tolerance=float(args.tolerance),
|
| 339 |
+
)
|
| 340 |
+
)
|
| 341 |
+
|
| 342 |
+
check = {
|
| 343 |
+
"status": "pass" if not mismatches else "fail",
|
| 344 |
+
"root": root.name,
|
| 345 |
+
"output_dir": release_path(root, output_dir),
|
| 346 |
+
"bootstrap_samples": int(args.bootstrap_samples),
|
| 347 |
+
"include_gap2": bool(args.include_gap2),
|
| 348 |
+
"overview_rows": int(overview.shape[0]),
|
| 349 |
+
"files": {
|
| 350 |
+
"overview": release_path(root, overview_path),
|
| 351 |
+
"table3_recomputed": release_path(root, table3_path),
|
| 352 |
+
"dual_track_recomputed": release_path(root, dual_path),
|
| 353 |
+
},
|
| 354 |
+
"mismatches": mismatches,
|
| 355 |
+
}
|
| 356 |
+
(output_dir / "release_evaluation_check.json").write_text(
|
| 357 |
+
json.dumps(check, ensure_ascii=False, indent=2) + "\n",
|
| 358 |
+
encoding="utf-8",
|
| 359 |
+
)
|
| 360 |
+
write_markdown(output_dir / "release_evaluation_check.md", check)
|
| 361 |
+
print(json.dumps(check, ensure_ascii=False, indent=2))
|
| 362 |
+
raise SystemExit(0 if check["status"] == "pass" else 1)
|
| 363 |
+
|
| 364 |
+
|
| 365 |
+
if __name__ == "__main__":
|
| 366 |
+
main()
|
scripts/run_retrieval_eval.py
ADDED
|
@@ -0,0 +1,629 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
from __future__ import annotations
|
| 3 |
+
|
| 4 |
+
import argparse
|
| 5 |
+
import json
|
| 6 |
+
from pathlib import Path
|
| 7 |
+
from typing import Any
|
| 8 |
+
|
| 9 |
+
import numpy as np
|
| 10 |
+
import pandas as pd
|
| 11 |
+
from sklearn.metrics import pairwise_distances
|
| 12 |
+
from sklearn.preprocessing import StandardScaler
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
def parse_args() -> argparse.Namespace:
|
| 16 |
+
parser = argparse.ArgumentParser(
|
| 17 |
+
description="Run the CB-Telemetry time-matched retrieval protocol on released feature tables."
|
| 18 |
+
)
|
| 19 |
+
parser.add_argument("--root", default=".", help="CB-Telemetry dataset root.")
|
| 20 |
+
parser.add_argument("--feature-table", required=True, help="Anchor feature table used to define rows and slots.")
|
| 21 |
+
parser.add_argument("--output-dir", required=True, help="Directory for retrieval outputs.")
|
| 22 |
+
parser.add_argument("--max-slot-gap", type=int, default=1, help="Maximum neighboring-slot gap.")
|
| 23 |
+
parser.add_argument("--min-dates-per-pair", type=int, default=12, help="Minimum dates required per slot pair.")
|
| 24 |
+
parser.add_argument("--bootstrap-samples", type=int, default=2000, help="Bootstrap resamples for intervals.")
|
| 25 |
+
parser.add_argument("--bootstrap-seed", type=int, default=42, help="Bootstrap random seed.")
|
| 26 |
+
parser.add_argument("--random-seed", type=int, default=42, help="Random seed for the permuted-control baseline.")
|
| 27 |
+
parser.add_argument(
|
| 28 |
+
"--archive-scope",
|
| 29 |
+
choices=["global", "same_archive_only"],
|
| 30 |
+
default="global",
|
| 31 |
+
help="Candidate pool: all target rows, or rows from the same archive only.",
|
| 32 |
+
)
|
| 33 |
+
parser.add_argument(
|
| 34 |
+
"--representation",
|
| 35 |
+
action="append",
|
| 36 |
+
default=[],
|
| 37 |
+
help="Repeat as name=feature_table_path. Paths are resolved relative to --root unless absolute.",
|
| 38 |
+
)
|
| 39 |
+
parser.add_argument(
|
| 40 |
+
"--include-metadata-controls",
|
| 41 |
+
action="store_true",
|
| 42 |
+
help="Add metadata_basic, metadata_calendar, and random_permuted_continuous controls.",
|
| 43 |
+
)
|
| 44 |
+
return parser.parse_args()
|
| 45 |
+
|
| 46 |
+
|
| 47 |
+
def resolve_path(root: Path, raw_path: str) -> Path:
|
| 48 |
+
path = Path(raw_path).expanduser()
|
| 49 |
+
if not path.is_absolute():
|
| 50 |
+
path = (root / path).resolve()
|
| 51 |
+
return path
|
| 52 |
+
|
| 53 |
+
|
| 54 |
+
def release_path(root: Path, path: Path) -> str:
|
| 55 |
+
try:
|
| 56 |
+
return path.resolve().relative_to(root).as_posix()
|
| 57 |
+
except ValueError:
|
| 58 |
+
return path.as_posix()
|
| 59 |
+
|
| 60 |
+
|
| 61 |
+
def build_key(df: pd.DataFrame) -> pd.Series:
|
| 62 |
+
return (
|
| 63 |
+
df["archive"].astype(str).str.strip()
|
| 64 |
+
+ "||"
|
| 65 |
+
+ df["date"].astype(str).str.strip()
|
| 66 |
+
+ "||"
|
| 67 |
+
+ df["filename"].astype(str).str.strip()
|
| 68 |
+
)
|
| 69 |
+
|
| 70 |
+
|
| 71 |
+
def add_slot_metadata(df: pd.DataFrame) -> pd.DataFrame:
|
| 72 |
+
result = df.copy()
|
| 73 |
+
result["timestamp_dt"] = pd.to_datetime(result["timestamp"], errors="raise")
|
| 74 |
+
result["date"] = result["date"].astype(str)
|
| 75 |
+
result["archive"] = result["archive"].astype(str)
|
| 76 |
+
result["filename"] = result["filename"].astype(str)
|
| 77 |
+
result["key"] = build_key(result)
|
| 78 |
+
result["dawn_slot_id"] = 0
|
| 79 |
+
result["slot_count_on_date"] = 0
|
| 80 |
+
result["slot_hour_signature"] = ""
|
| 81 |
+
for date_str, group in result.groupby("date", sort=True):
|
| 82 |
+
ordered = group.sort_values(["timestamp_dt", "filename"]).copy()
|
| 83 |
+
slot_ids = np.arange(1, len(ordered) + 1, dtype=np.int32)
|
| 84 |
+
result.loc[ordered.index, "dawn_slot_id"] = slot_ids
|
| 85 |
+
result.loc[ordered.index, "slot_count_on_date"] = int(len(ordered))
|
| 86 |
+
signature = "-".join(str(int(item)) for item in ordered["hour"].tolist())
|
| 87 |
+
result.loc[ordered.index, "slot_hour_signature"] = signature
|
| 88 |
+
result["dawn_slot_id"] = result["dawn_slot_id"].astype(int)
|
| 89 |
+
result["slot_count_on_date"] = result["slot_count_on_date"].astype(int)
|
| 90 |
+
return result
|
| 91 |
+
|
| 92 |
+
|
| 93 |
+
def feature_columns(df: pd.DataFrame) -> list[str]:
|
| 94 |
+
columns = [col for col in df.columns if col.startswith("f_")]
|
| 95 |
+
if not columns:
|
| 96 |
+
raise RuntimeError("No f_* feature columns were found.")
|
| 97 |
+
return columns
|
| 98 |
+
|
| 99 |
+
|
| 100 |
+
def standardize(matrix: np.ndarray) -> np.ndarray:
|
| 101 |
+
return StandardScaler().fit_transform(matrix.astype(np.float32, copy=False)).astype(np.float32)
|
| 102 |
+
|
| 103 |
+
|
| 104 |
+
def load_representation_matrix(root: Path, path_text: str, anchor_df: pd.DataFrame) -> np.ndarray:
|
| 105 |
+
path = resolve_path(root, path_text)
|
| 106 |
+
df = pd.read_csv(path, low_memory=False).copy()
|
| 107 |
+
df["date"] = df["date"].astype(str)
|
| 108 |
+
df["archive"] = df["archive"].astype(str)
|
| 109 |
+
df["filename"] = df["filename"].astype(str)
|
| 110 |
+
df["key"] = build_key(df)
|
| 111 |
+
columns = feature_columns(df)
|
| 112 |
+
indexed = df.set_index("key")
|
| 113 |
+
keys = anchor_df["key"].tolist()
|
| 114 |
+
missing = [key for key in keys if key not in indexed.index]
|
| 115 |
+
if missing:
|
| 116 |
+
raise RuntimeError(f"{path} is missing {len(missing)} anchor rows.")
|
| 117 |
+
matrix = indexed.loc[keys, columns].to_numpy(dtype=np.float32)
|
| 118 |
+
return standardize(matrix)
|
| 119 |
+
|
| 120 |
+
|
| 121 |
+
def parse_representations(items: list[str]) -> list[tuple[str, str]]:
|
| 122 |
+
if not items:
|
| 123 |
+
raise ValueError("At least one --representation name=path is required.")
|
| 124 |
+
result: list[tuple[str, str]] = []
|
| 125 |
+
for item in items:
|
| 126 |
+
if "=" not in item:
|
| 127 |
+
raise ValueError(f"Invalid representation argument: {item}")
|
| 128 |
+
name, raw_path = item.split("=", 1)
|
| 129 |
+
result.append((name.strip(), raw_path.strip()))
|
| 130 |
+
return result
|
| 131 |
+
|
| 132 |
+
|
| 133 |
+
def build_metadata_matrix(anchor_df: pd.DataFrame, include_calendar: bool) -> np.ndarray:
|
| 134 |
+
archive_dummies = pd.get_dummies(anchor_df["archive"].astype(str), prefix="archive", dtype=np.float32)
|
| 135 |
+
slot_dummies = pd.get_dummies(anchor_df["dawn_slot_id"].astype(int), prefix="slot", dtype=np.float32)
|
| 136 |
+
blocks = [
|
| 137 |
+
anchor_df["hour"].astype(np.float32).to_numpy()[:, None],
|
| 138 |
+
archive_dummies.to_numpy(dtype=np.float32),
|
| 139 |
+
slot_dummies.to_numpy(dtype=np.float32),
|
| 140 |
+
]
|
| 141 |
+
if include_calendar:
|
| 142 |
+
date_dt = pd.to_datetime(anchor_df["date"].astype(str), format="%Y%m%d", errors="raise")
|
| 143 |
+
month_angle = 2.0 * np.pi * (date_dt.dt.month.to_numpy(dtype=np.float32) - 1.0) / 12.0
|
| 144 |
+
year_offset = date_dt.dt.year.to_numpy(dtype=np.float32) - float(date_dt.dt.year.min())
|
| 145 |
+
blocks.extend(
|
| 146 |
+
[
|
| 147 |
+
np.sin(month_angle).astype(np.float32)[:, None],
|
| 148 |
+
np.cos(month_angle).astype(np.float32)[:, None],
|
| 149 |
+
year_offset.astype(np.float32)[:, None],
|
| 150 |
+
]
|
| 151 |
+
)
|
| 152 |
+
return standardize(np.column_stack(blocks).astype(np.float32))
|
| 153 |
+
|
| 154 |
+
|
| 155 |
+
def cosine_rank_indices(query_matrix: np.ndarray, target_matrix: np.ndarray) -> np.ndarray:
|
| 156 |
+
distances = pairwise_distances(query_matrix, target_matrix, metric="cosine")
|
| 157 |
+
return np.argsort(distances, axis=1)
|
| 158 |
+
|
| 159 |
+
|
| 160 |
+
def reciprocal_rank(rank_index_zero_based: int) -> float:
|
| 161 |
+
return 1.0 / float(rank_index_zero_based + 1)
|
| 162 |
+
|
| 163 |
+
|
| 164 |
+
def summarize_query_rows(query_df: pd.DataFrame) -> dict[str, float]:
|
| 165 |
+
rank_positions = query_df["rank_position"].to_numpy(dtype=np.float32)
|
| 166 |
+
return {
|
| 167 |
+
"top1_hit_rate": float(query_df["top1_hit"].mean()),
|
| 168 |
+
"top5_hit_rate": float(query_df["top5_hit"].mean()),
|
| 169 |
+
"mrr": float(query_df["reciprocal_rank"].mean()),
|
| 170 |
+
"median_rank": float(np.median(rank_positions)),
|
| 171 |
+
"mean_rank": float(np.mean(rank_positions)),
|
| 172 |
+
}
|
| 173 |
+
|
| 174 |
+
|
| 175 |
+
def evaluate_pair_subset(
|
| 176 |
+
matrix: np.ndarray,
|
| 177 |
+
pair_df: pd.DataFrame,
|
| 178 |
+
query_slot: int,
|
| 179 |
+
target_slot: int,
|
| 180 |
+
archive_label: str,
|
| 181 |
+
) -> dict[str, Any]:
|
| 182 |
+
eligible_dates = pair_df.groupby("date")["dawn_slot_id"].nunique().reset_index(name="slot_count")
|
| 183 |
+
eligible_dates = eligible_dates[eligible_dates["slot_count"] == 2]["date"].astype(str).tolist()
|
| 184 |
+
pair_df = pair_df[pair_df["date"].isin(eligible_dates)].copy()
|
| 185 |
+
query_df = pair_df[pair_df["dawn_slot_id"] == query_slot].copy().sort_values("date").reset_index(drop=True)
|
| 186 |
+
target_df = pair_df[pair_df["dawn_slot_id"] == target_slot].copy().sort_values("date").reset_index(drop=True)
|
| 187 |
+
|
| 188 |
+
if query_df.empty or target_df.empty:
|
| 189 |
+
raise RuntimeError(f"slot_pair {query_slot}->{target_slot} has no valid rows.")
|
| 190 |
+
if query_df["date"].tolist() != target_df["date"].tolist():
|
| 191 |
+
raise RuntimeError(f"slot_pair {query_slot}->{target_slot} query and target dates are not aligned.")
|
| 192 |
+
|
| 193 |
+
query_matrix = matrix[query_df["anchor_row_index"].to_numpy()]
|
| 194 |
+
target_matrix = matrix[target_df["anchor_row_index"].to_numpy()]
|
| 195 |
+
ranks = cosine_rank_indices(query_matrix, target_matrix)
|
| 196 |
+
|
| 197 |
+
positive_dates = query_df["date"].tolist()
|
| 198 |
+
target_dates = target_df["date"].tolist()
|
| 199 |
+
query_rows: list[dict[str, Any]] = []
|
| 200 |
+
for row_idx, expected_date in enumerate(positive_dates):
|
| 201 |
+
ranked_dates = [target_dates[item] for item in ranks[row_idx].tolist()]
|
| 202 |
+
positive_rank = ranked_dates.index(expected_date)
|
| 203 |
+
query_rows.append(
|
| 204 |
+
{
|
| 205 |
+
"date": str(expected_date),
|
| 206 |
+
"query_slot": int(query_slot),
|
| 207 |
+
"target_slot": int(target_slot),
|
| 208 |
+
"archive_scope_label": archive_label,
|
| 209 |
+
"candidate_size_per_query": int(len(target_df)),
|
| 210 |
+
"chance_top1": 1.0 / float(len(target_df)),
|
| 211 |
+
"rank_position": int(positive_rank + 1),
|
| 212 |
+
"top1_hit": 1.0 if positive_rank == 0 else 0.0,
|
| 213 |
+
"top5_hit": 1.0 if positive_rank < min(5, len(ranked_dates)) else 0.0,
|
| 214 |
+
"reciprocal_rank": reciprocal_rank(positive_rank),
|
| 215 |
+
}
|
| 216 |
+
)
|
| 217 |
+
|
| 218 |
+
query_result_df = pd.DataFrame(query_rows)
|
| 219 |
+
metrics = summarize_query_rows(query_result_df)
|
| 220 |
+
return {
|
| 221 |
+
"query_slot": int(query_slot),
|
| 222 |
+
"target_slot": int(target_slot),
|
| 223 |
+
"archive_scope_label": archive_label,
|
| 224 |
+
"query_rows": int(len(query_df)),
|
| 225 |
+
"candidate_rows": int(len(target_df)),
|
| 226 |
+
"candidate_size_per_query": int(len(target_df)),
|
| 227 |
+
"chance_top1": 1.0 / float(len(target_df)),
|
| 228 |
+
"top1_hit_rate": metrics["top1_hit_rate"],
|
| 229 |
+
"top5_hit_rate": metrics["top5_hit_rate"],
|
| 230 |
+
"mrr": metrics["mrr"],
|
| 231 |
+
"median_rank": metrics["median_rank"],
|
| 232 |
+
"mean_rank": metrics["mean_rank"],
|
| 233 |
+
"date_examples": positive_dates[:5],
|
| 234 |
+
"query_details": query_rows,
|
| 235 |
+
}
|
| 236 |
+
|
| 237 |
+
|
| 238 |
+
def evaluate_slot_pair(
|
| 239 |
+
matrix: np.ndarray,
|
| 240 |
+
anchor_df: pd.DataFrame,
|
| 241 |
+
query_slot: int,
|
| 242 |
+
target_slot: int,
|
| 243 |
+
archive_scope: str,
|
| 244 |
+
min_dates_per_pair: int,
|
| 245 |
+
) -> list[dict[str, Any]]:
|
| 246 |
+
pair_df = anchor_df[anchor_df["dawn_slot_id"].isin([query_slot, target_slot])].copy()
|
| 247 |
+
if archive_scope == "global":
|
| 248 |
+
return [
|
| 249 |
+
evaluate_pair_subset(
|
| 250 |
+
matrix,
|
| 251 |
+
pair_df=pair_df,
|
| 252 |
+
query_slot=query_slot,
|
| 253 |
+
target_slot=target_slot,
|
| 254 |
+
archive_label="all",
|
| 255 |
+
)
|
| 256 |
+
]
|
| 257 |
+
|
| 258 |
+
rows: list[dict[str, Any]] = []
|
| 259 |
+
for archive_name, archive_df in pair_df.groupby("archive", sort=True):
|
| 260 |
+
if archive_df["dawn_slot_id"].nunique() < 2:
|
| 261 |
+
continue
|
| 262 |
+
eligible_dates = archive_df.groupby("date")["dawn_slot_id"].nunique().reset_index(name="slot_count")
|
| 263 |
+
eligible_dates = eligible_dates[eligible_dates["slot_count"] == 2]
|
| 264 |
+
if int(len(eligible_dates)) < min_dates_per_pair:
|
| 265 |
+
continue
|
| 266 |
+
rows.append(
|
| 267 |
+
evaluate_pair_subset(
|
| 268 |
+
matrix,
|
| 269 |
+
pair_df=archive_df.copy(),
|
| 270 |
+
query_slot=query_slot,
|
| 271 |
+
target_slot=target_slot,
|
| 272 |
+
archive_label=str(archive_name),
|
| 273 |
+
)
|
| 274 |
+
)
|
| 275 |
+
return rows
|
| 276 |
+
|
| 277 |
+
|
| 278 |
+
def build_slot_pairs(
|
| 279 |
+
anchor_df: pd.DataFrame,
|
| 280 |
+
max_slot_gap: int,
|
| 281 |
+
min_dates_per_pair: int,
|
| 282 |
+
archive_scope: str,
|
| 283 |
+
) -> list[tuple[int, int]]:
|
| 284 |
+
max_slot_id = int(anchor_df["dawn_slot_id"].max())
|
| 285 |
+
pairs: list[tuple[int, int]] = []
|
| 286 |
+
for query_slot in range(1, max_slot_id + 1):
|
| 287 |
+
for target_slot in range(1, max_slot_id + 1):
|
| 288 |
+
if query_slot == target_slot:
|
| 289 |
+
continue
|
| 290 |
+
if abs(query_slot - target_slot) > max_slot_gap:
|
| 291 |
+
continue
|
| 292 |
+
pair_df = anchor_df[anchor_df["dawn_slot_id"].isin([query_slot, target_slot])].copy()
|
| 293 |
+
if archive_scope == "global":
|
| 294 |
+
eligible_dates = pair_df.groupby("date")["dawn_slot_id"].nunique().reset_index(name="slot_count")
|
| 295 |
+
eligible_dates = eligible_dates[eligible_dates["slot_count"] == 2]
|
| 296 |
+
if int(len(eligible_dates)) >= min_dates_per_pair:
|
| 297 |
+
pairs.append((query_slot, target_slot))
|
| 298 |
+
continue
|
| 299 |
+
|
| 300 |
+
archive_has_valid_pair = False
|
| 301 |
+
for _, archive_df in pair_df.groupby("archive", sort=True):
|
| 302 |
+
if archive_df["dawn_slot_id"].nunique() < 2:
|
| 303 |
+
continue
|
| 304 |
+
eligible_dates = archive_df.groupby("date")["dawn_slot_id"].nunique().reset_index(name="slot_count")
|
| 305 |
+
eligible_dates = eligible_dates[eligible_dates["slot_count"] == 2]
|
| 306 |
+
if int(len(eligible_dates)) >= min_dates_per_pair:
|
| 307 |
+
archive_has_valid_pair = True
|
| 308 |
+
break
|
| 309 |
+
if archive_has_valid_pair:
|
| 310 |
+
pairs.append((query_slot, target_slot))
|
| 311 |
+
if not pairs:
|
| 312 |
+
raise RuntimeError("No valid slot pairs satisfy min_dates_per_pair.")
|
| 313 |
+
return sorted(pairs)
|
| 314 |
+
|
| 315 |
+
|
| 316 |
+
def aggregate_pair_rows(method_name: str, pair_df: pd.DataFrame) -> dict[str, float | str]:
|
| 317 |
+
return {
|
| 318 |
+
"method": method_name,
|
| 319 |
+
"mean_top1_hit_rate": float(pair_df["top1_hit_rate"].mean()),
|
| 320 |
+
"mean_top5_hit_rate": float(pair_df["top5_hit_rate"].mean()),
|
| 321 |
+
"mean_mrr": float(pair_df["mrr"].mean()),
|
| 322 |
+
"mean_median_rank": float(pair_df["median_rank"].mean()),
|
| 323 |
+
"mean_candidate_size": float(pair_df["candidate_size_per_query"].mean()),
|
| 324 |
+
"mean_chance_top1": float(pair_df["chance_top1"].mean()),
|
| 325 |
+
}
|
| 326 |
+
|
| 327 |
+
|
| 328 |
+
def evaluate_representation(
|
| 329 |
+
method_name: str,
|
| 330 |
+
matrix: np.ndarray,
|
| 331 |
+
anchor_df: pd.DataFrame,
|
| 332 |
+
slot_pairs: list[tuple[int, int]],
|
| 333 |
+
archive_scope: str,
|
| 334 |
+
min_dates_per_pair: int,
|
| 335 |
+
) -> dict[str, Any]:
|
| 336 |
+
pair_rows: list[dict[str, Any]] = []
|
| 337 |
+
query_rows: list[dict[str, Any]] = []
|
| 338 |
+
for query_slot, target_slot in slot_pairs:
|
| 339 |
+
rows = evaluate_slot_pair(
|
| 340 |
+
matrix,
|
| 341 |
+
anchor_df,
|
| 342 |
+
query_slot=query_slot,
|
| 343 |
+
target_slot=target_slot,
|
| 344 |
+
archive_scope=archive_scope,
|
| 345 |
+
min_dates_per_pair=min_dates_per_pair,
|
| 346 |
+
)
|
| 347 |
+
for row in rows:
|
| 348 |
+
raw_query_details = row.pop("query_details")
|
| 349 |
+
row["method"] = method_name
|
| 350 |
+
pair_rows.append(row)
|
| 351 |
+
for query_detail in raw_query_details:
|
| 352 |
+
query_detail["method"] = method_name
|
| 353 |
+
query_rows.append(query_detail)
|
| 354 |
+
if not pair_rows:
|
| 355 |
+
raise RuntimeError(f"method={method_name} has no valid pair rows.")
|
| 356 |
+
aggregate = aggregate_pair_rows(method_name, pd.DataFrame(pair_rows))
|
| 357 |
+
return {"aggregate": aggregate, "pairs": pair_rows, "queries": query_rows}
|
| 358 |
+
|
| 359 |
+
|
| 360 |
+
def percentile_interval(samples: list[float]) -> tuple[float, float]:
|
| 361 |
+
values = np.asarray(samples, dtype=np.float64)
|
| 362 |
+
return float(np.percentile(values, 2.5)), float(np.percentile(values, 97.5))
|
| 363 |
+
|
| 364 |
+
|
| 365 |
+
def bootstrap_pair_level(
|
| 366 |
+
method_name: str,
|
| 367 |
+
pair_df: pd.DataFrame,
|
| 368 |
+
bootstrap_samples: int,
|
| 369 |
+
rng: np.random.Generator,
|
| 370 |
+
) -> dict[str, Any]:
|
| 371 |
+
metric_names = ["top1_hit_rate", "mrr", "top5_hit_rate"]
|
| 372 |
+
point_estimates = {metric: float(pair_df[metric].mean()) for metric in metric_names}
|
| 373 |
+
boot_values = {metric: [] for metric in metric_names}
|
| 374 |
+
pair_count = len(pair_df)
|
| 375 |
+
for _ in range(bootstrap_samples):
|
| 376 |
+
sampled_indices = rng.integers(0, pair_count, size=pair_count)
|
| 377 |
+
sampled = pair_df.iloc[sampled_indices]
|
| 378 |
+
for metric in metric_names:
|
| 379 |
+
boot_values[metric].append(float(sampled[metric].mean()))
|
| 380 |
+
|
| 381 |
+
result: dict[str, Any] = {
|
| 382 |
+
"method": method_name,
|
| 383 |
+
"bootstrap_unit": "pair",
|
| 384 |
+
"bootstrap_samples": int(bootstrap_samples),
|
| 385 |
+
"group_count": int(pair_count),
|
| 386 |
+
}
|
| 387 |
+
for metric in metric_names:
|
| 388 |
+
ci_low, ci_high = percentile_interval(boot_values[metric])
|
| 389 |
+
result[f"{metric}_point_estimate"] = point_estimates[metric]
|
| 390 |
+
result[f"{metric}_ci_low"] = ci_low
|
| 391 |
+
result[f"{metric}_ci_high"] = ci_high
|
| 392 |
+
return result
|
| 393 |
+
|
| 394 |
+
|
| 395 |
+
def bootstrap_date_level(
|
| 396 |
+
method_name: str,
|
| 397 |
+
query_df: pd.DataFrame,
|
| 398 |
+
bootstrap_samples: int,
|
| 399 |
+
rng: np.random.Generator,
|
| 400 |
+
) -> dict[str, Any]:
|
| 401 |
+
grouped_queries = {
|
| 402 |
+
group_key: group.copy().reset_index(drop=True)
|
| 403 |
+
for group_key, group in query_df.groupby(["query_slot", "target_slot", "archive_scope_label"], sort=True)
|
| 404 |
+
}
|
| 405 |
+
point_pair_rows: list[dict[str, float | str]] = []
|
| 406 |
+
for group_key, group in grouped_queries.items():
|
| 407 |
+
metrics = summarize_query_rows(group)
|
| 408 |
+
point_pair_rows.append(
|
| 409 |
+
{
|
| 410 |
+
"group_key": str(group_key),
|
| 411 |
+
"top1_hit_rate": metrics["top1_hit_rate"],
|
| 412 |
+
"mrr": metrics["mrr"],
|
| 413 |
+
"top5_hit_rate": metrics["top5_hit_rate"],
|
| 414 |
+
}
|
| 415 |
+
)
|
| 416 |
+
point_pair_df = pd.DataFrame(point_pair_rows)
|
| 417 |
+
point_estimates = {
|
| 418 |
+
"top1_hit_rate": float(point_pair_df["top1_hit_rate"].mean()),
|
| 419 |
+
"mrr": float(point_pair_df["mrr"].mean()),
|
| 420 |
+
"top5_hit_rate": float(point_pair_df["top5_hit_rate"].mean()),
|
| 421 |
+
}
|
| 422 |
+
boot_values = {"top1_hit_rate": [], "mrr": [], "top5_hit_rate": []}
|
| 423 |
+
for _ in range(bootstrap_samples):
|
| 424 |
+
sampled_pair_rows: list[dict[str, float | str]] = []
|
| 425 |
+
for group_key, group in grouped_queries.items():
|
| 426 |
+
sampled_indices = rng.integers(0, len(group), size=len(group))
|
| 427 |
+
sampled_group = group.iloc[sampled_indices].reset_index(drop=True)
|
| 428 |
+
metrics = summarize_query_rows(sampled_group)
|
| 429 |
+
sampled_pair_rows.append(
|
| 430 |
+
{
|
| 431 |
+
"group_key": str(group_key),
|
| 432 |
+
"top1_hit_rate": metrics["top1_hit_rate"],
|
| 433 |
+
"mrr": metrics["mrr"],
|
| 434 |
+
"top5_hit_rate": metrics["top5_hit_rate"],
|
| 435 |
+
}
|
| 436 |
+
)
|
| 437 |
+
sampled_pair_df = pd.DataFrame(sampled_pair_rows)
|
| 438 |
+
boot_values["top1_hit_rate"].append(float(sampled_pair_df["top1_hit_rate"].mean()))
|
| 439 |
+
boot_values["mrr"].append(float(sampled_pair_df["mrr"].mean()))
|
| 440 |
+
boot_values["top5_hit_rate"].append(float(sampled_pair_df["top5_hit_rate"].mean()))
|
| 441 |
+
|
| 442 |
+
result: dict[str, Any] = {
|
| 443 |
+
"method": method_name,
|
| 444 |
+
"bootstrap_unit": "date",
|
| 445 |
+
"bootstrap_samples": int(bootstrap_samples),
|
| 446 |
+
"group_count": int(len(grouped_queries)),
|
| 447 |
+
"query_row_count": int(len(query_df)),
|
| 448 |
+
}
|
| 449 |
+
for metric in ["top1_hit_rate", "mrr", "top5_hit_rate"]:
|
| 450 |
+
ci_low, ci_high = percentile_interval(boot_values[metric])
|
| 451 |
+
result[f"{metric}_point_estimate"] = point_estimates[metric]
|
| 452 |
+
result[f"{metric}_ci_low"] = ci_low
|
| 453 |
+
result[f"{metric}_ci_high"] = ci_high
|
| 454 |
+
return result
|
| 455 |
+
|
| 456 |
+
|
| 457 |
+
def bootstrap_rows(
|
| 458 |
+
method_name: str,
|
| 459 |
+
pair_df: pd.DataFrame,
|
| 460 |
+
query_df: pd.DataFrame,
|
| 461 |
+
bootstrap_samples: int,
|
| 462 |
+
bootstrap_seed: int,
|
| 463 |
+
) -> list[dict[str, Any]]:
|
| 464 |
+
pair_rng = np.random.default_rng(bootstrap_seed)
|
| 465 |
+
date_rng = np.random.default_rng(bootstrap_seed + 1000)
|
| 466 |
+
return [
|
| 467 |
+
bootstrap_date_level(method_name, query_df, bootstrap_samples=bootstrap_samples, rng=date_rng),
|
| 468 |
+
bootstrap_pair_level(method_name, pair_df, bootstrap_samples=bootstrap_samples, rng=pair_rng),
|
| 469 |
+
]
|
| 470 |
+
|
| 471 |
+
|
| 472 |
+
def write_markdown(output_path: Path, summary: dict[str, Any]) -> None:
|
| 473 |
+
lines = [
|
| 474 |
+
"# CB-Telemetry Time-Matched Retrieval",
|
| 475 |
+
"",
|
| 476 |
+
f"- feature_table: `{summary['feature_table']}`",
|
| 477 |
+
f"- max_slot_gap: `{summary['max_slot_gap']}`",
|
| 478 |
+
f"- min_dates_per_pair: `{summary['min_dates_per_pair']}`",
|
| 479 |
+
f"- archive_scope: `{summary['archive_scope']}`",
|
| 480 |
+
f"- bootstrap_samples: `{summary['bootstrap_samples']}`",
|
| 481 |
+
f"- slot_pairs: `{summary['slot_pairs']}`",
|
| 482 |
+
"",
|
| 483 |
+
"## Aggregate",
|
| 484 |
+
"",
|
| 485 |
+
"| method | top1 | chance_top1 | mrr | top5 | mean_candidate_size |",
|
| 486 |
+
"| --- | --- | --- | --- | --- | --- |",
|
| 487 |
+
]
|
| 488 |
+
for row in summary["aggregate_rows"]:
|
| 489 |
+
lines.append(
|
| 490 |
+
f"| {row['method']} | {row['mean_top1_hit_rate']:.4f} | {row['mean_chance_top1']:.4f} | "
|
| 491 |
+
f"{row['mean_mrr']:.4f} | {row['mean_top5_hit_rate']:.4f} | {row['mean_candidate_size']:.1f} |"
|
| 492 |
+
)
|
| 493 |
+
lines.extend(["", "## Bootstrap 95% CI", "", "| method | unit | top1 | mrr | top5 |", "| --- | --- | --- | --- | --- |"])
|
| 494 |
+
for row in summary["bootstrap_rows"]:
|
| 495 |
+
lines.append(
|
| 496 |
+
f"| {row['method']} | {row['bootstrap_unit']} | "
|
| 497 |
+
f"{row['top1_hit_rate_point_estimate']:.4f} [{row['top1_hit_rate_ci_low']:.4f}, {row['top1_hit_rate_ci_high']:.4f}] | "
|
| 498 |
+
f"{row['mrr_point_estimate']:.4f} [{row['mrr_ci_low']:.4f}, {row['mrr_ci_high']:.4f}] | "
|
| 499 |
+
f"{row['top5_hit_rate_point_estimate']:.4f} [{row['top5_hit_rate_ci_low']:.4f}, {row['top5_hit_rate_ci_high']:.4f}] |"
|
| 500 |
+
)
|
| 501 |
+
output_path.write_text("\n".join(lines) + "\n", encoding="utf-8")
|
| 502 |
+
|
| 503 |
+
|
| 504 |
+
def main() -> None:
|
| 505 |
+
args = parse_args()
|
| 506 |
+
root = Path(args.root).expanduser().resolve()
|
| 507 |
+
output_dir = resolve_path(root, args.output_dir)
|
| 508 |
+
output_dir.mkdir(parents=True, exist_ok=True)
|
| 509 |
+
|
| 510 |
+
feature_table_path = resolve_path(root, args.feature_table)
|
| 511 |
+
anchor_df = add_slot_metadata(pd.read_csv(feature_table_path, low_memory=False))
|
| 512 |
+
anchor_df = anchor_df.sort_values(["date", "timestamp_dt", "filename"]).reset_index(drop=True)
|
| 513 |
+
anchor_df["anchor_row_index"] = np.arange(len(anchor_df), dtype=np.int32)
|
| 514 |
+
slot_pairs = build_slot_pairs(
|
| 515 |
+
anchor_df,
|
| 516 |
+
max_slot_gap=args.max_slot_gap,
|
| 517 |
+
min_dates_per_pair=args.min_dates_per_pair,
|
| 518 |
+
archive_scope=args.archive_scope,
|
| 519 |
+
)
|
| 520 |
+
|
| 521 |
+
representation_matrices: list[tuple[str, np.ndarray, str]] = []
|
| 522 |
+
for method_name, raw_path in parse_representations(args.representation):
|
| 523 |
+
representation_matrices.append((method_name, load_representation_matrix(root, raw_path, anchor_df), raw_path))
|
| 524 |
+
|
| 525 |
+
if args.include_metadata_controls:
|
| 526 |
+
representation_matrices.append(("metadata_basic", build_metadata_matrix(anchor_df, include_calendar=False), "generated"))
|
| 527 |
+
representation_matrices.append(("metadata_calendar", build_metadata_matrix(anchor_df, include_calendar=True), "generated"))
|
| 528 |
+
continuous_columns = feature_columns(anchor_df)
|
| 529 |
+
continuous_matrix = anchor_df[continuous_columns].to_numpy(dtype=np.float32)
|
| 530 |
+
rng = np.random.default_rng(args.random_seed)
|
| 531 |
+
permuted = continuous_matrix[rng.permutation(len(anchor_df))]
|
| 532 |
+
representation_matrices.append(("random_permuted_continuous", standardize(permuted), "generated"))
|
| 533 |
+
|
| 534 |
+
aggregate_rows: list[dict[str, Any]] = []
|
| 535 |
+
pair_rows: list[dict[str, Any]] = []
|
| 536 |
+
query_rows: list[dict[str, Any]] = []
|
| 537 |
+
all_bootstrap_rows: list[dict[str, Any]] = []
|
| 538 |
+
detailed_results: dict[str, Any] = {}
|
| 539 |
+
|
| 540 |
+
for method_name, matrix, source_path in representation_matrices:
|
| 541 |
+
result = evaluate_representation(
|
| 542 |
+
method_name=method_name,
|
| 543 |
+
matrix=matrix,
|
| 544 |
+
anchor_df=anchor_df,
|
| 545 |
+
slot_pairs=slot_pairs,
|
| 546 |
+
archive_scope=args.archive_scope,
|
| 547 |
+
min_dates_per_pair=args.min_dates_per_pair,
|
| 548 |
+
)
|
| 549 |
+
aggregate_rows.append(result["aggregate"])
|
| 550 |
+
pair_rows.extend(result["pairs"])
|
| 551 |
+
query_rows.extend(result["queries"])
|
| 552 |
+
method_pair_df = pd.DataFrame(result["pairs"])
|
| 553 |
+
method_query_df = pd.DataFrame(result["queries"])
|
| 554 |
+
method_bootstrap_rows = bootstrap_rows(
|
| 555 |
+
method_name=method_name,
|
| 556 |
+
pair_df=method_pair_df,
|
| 557 |
+
query_df=method_query_df,
|
| 558 |
+
bootstrap_samples=args.bootstrap_samples,
|
| 559 |
+
bootstrap_seed=args.bootstrap_seed,
|
| 560 |
+
)
|
| 561 |
+
all_bootstrap_rows.extend(method_bootstrap_rows)
|
| 562 |
+
detailed_results[method_name] = {
|
| 563 |
+
"feature_table": source_path,
|
| 564 |
+
"aggregate": result["aggregate"],
|
| 565 |
+
"bootstrap": method_bootstrap_rows,
|
| 566 |
+
}
|
| 567 |
+
|
| 568 |
+
aggregate_df = pd.DataFrame(aggregate_rows).sort_values(
|
| 569 |
+
["mean_top1_hit_rate", "mean_mrr", "mean_top5_hit_rate"],
|
| 570 |
+
ascending=False,
|
| 571 |
+
).reset_index(drop=True)
|
| 572 |
+
pair_df = pd.DataFrame(pair_rows).sort_values(["method", "query_slot", "target_slot"]).reset_index(drop=True)
|
| 573 |
+
query_df = pd.DataFrame(query_rows).sort_values(
|
| 574 |
+
["method", "query_slot", "target_slot", "archive_scope_label", "date"]
|
| 575 |
+
).reset_index(drop=True)
|
| 576 |
+
bootstrap_df = pd.DataFrame(all_bootstrap_rows).sort_values(
|
| 577 |
+
["bootstrap_unit", "top1_hit_rate_point_estimate", "mrr_point_estimate", "top5_hit_rate_point_estimate"],
|
| 578 |
+
ascending=[True, False, False, False],
|
| 579 |
+
).reset_index(drop=True)
|
| 580 |
+
|
| 581 |
+
aggregate_df.to_csv(output_dir / "aggregate_summary.csv", index=False)
|
| 582 |
+
pair_df.to_csv(output_dir / "pair_summary.csv", index=False)
|
| 583 |
+
query_df.to_csv(output_dir / "query_summary.csv", index=False)
|
| 584 |
+
bootstrap_df.to_csv(output_dir / "bootstrap_summary.csv", index=False)
|
| 585 |
+
|
| 586 |
+
summary: dict[str, Any] = {
|
| 587 |
+
"feature_table": release_path(root, feature_table_path),
|
| 588 |
+
"output_dir": release_path(root, output_dir),
|
| 589 |
+
"files": {
|
| 590 |
+
"aggregate_summary_csv": release_path(root, output_dir / "aggregate_summary.csv"),
|
| 591 |
+
"pair_summary_csv": release_path(root, output_dir / "pair_summary.csv"),
|
| 592 |
+
"query_summary_csv": release_path(root, output_dir / "query_summary.csv"),
|
| 593 |
+
"bootstrap_summary_csv": release_path(root, output_dir / "bootstrap_summary.csv"),
|
| 594 |
+
},
|
| 595 |
+
"max_slot_gap": int(args.max_slot_gap),
|
| 596 |
+
"min_dates_per_pair": int(args.min_dates_per_pair),
|
| 597 |
+
"archive_scope": args.archive_scope,
|
| 598 |
+
"bootstrap_samples": int(args.bootstrap_samples),
|
| 599 |
+
"bootstrap_seed": int(args.bootstrap_seed),
|
| 600 |
+
"slot_pairs": [f"{src}->{dst}" for src, dst in slot_pairs],
|
| 601 |
+
"rows": int(len(anchor_df)),
|
| 602 |
+
"date_count": int(anchor_df["date"].nunique()),
|
| 603 |
+
"slot_count_distribution": {
|
| 604 |
+
str(key): int(value)
|
| 605 |
+
for key, value in anchor_df.groupby("date")["dawn_slot_id"].max().value_counts().sort_index().to_dict().items()
|
| 606 |
+
},
|
| 607 |
+
"aggregate_rows": aggregate_df.to_dict(orient="records"),
|
| 608 |
+
"bootstrap_rows": bootstrap_df.to_dict(orient="records"),
|
| 609 |
+
"pair_rows": pair_df.to_dict(orient="records"),
|
| 610 |
+
"detailed_results": detailed_results,
|
| 611 |
+
}
|
| 612 |
+
(output_dir / "summary.json").write_text(json.dumps(summary, ensure_ascii=False, indent=2), encoding="utf-8")
|
| 613 |
+
write_markdown(output_dir / "summary.md", summary)
|
| 614 |
+
print(
|
| 615 |
+
json.dumps(
|
| 616 |
+
{
|
| 617 |
+
"output_dir": summary["output_dir"],
|
| 618 |
+
"archive_scope": summary["archive_scope"],
|
| 619 |
+
"max_slot_gap": summary["max_slot_gap"],
|
| 620 |
+
"aggregate_rows": summary["aggregate_rows"],
|
| 621 |
+
},
|
| 622 |
+
ensure_ascii=False,
|
| 623 |
+
indent=2,
|
| 624 |
+
)
|
| 625 |
+
)
|
| 626 |
+
|
| 627 |
+
|
| 628 |
+
if __name__ == "__main__":
|
| 629 |
+
main()
|
scripts/validate_cb_telemetry.py
CHANGED
|
@@ -29,10 +29,10 @@ LOCAL_PATH_PATTERNS = [
|
|
| 29 |
re.compile(r"\\Users\\"),
|
| 30 |
]
|
| 31 |
|
| 32 |
-
|
| 33 |
-
re.compile(r"\
|
| 34 |
-
re.compile(r"\
|
| 35 |
-
re.compile(
|
| 36 |
]
|
| 37 |
|
| 38 |
JUNK_FILE_NAMES = {".DS_Store", "Thumbs.db"}
|
|
@@ -111,9 +111,9 @@ def scan_text(root: Path, errors: list[dict[str, str]], warnings: list[dict[str,
|
|
| 111 |
for pattern in LOCAL_PATH_PATTERNS:
|
| 112 |
if pattern.search(text):
|
| 113 |
add_issue(errors, "local_path_leak", f"Local path pattern found in {rel}: {pattern.pattern}")
|
| 114 |
-
for pattern in
|
| 115 |
if pattern.search(text):
|
| 116 |
-
add_issue(warnings, "
|
| 117 |
|
| 118 |
|
| 119 |
def validate_ids(root: Path, errors: list[dict[str, str]], warnings: list[dict[str, str]]) -> dict[str, int]:
|
|
@@ -341,10 +341,19 @@ def main() -> None:
|
|
| 341 |
"manifests/splits.csv",
|
| 342 |
"features/feature_table_default.csv.gz",
|
| 343 |
"features/feature_table_strict_clean.csv.gz",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 344 |
"baselines/table_2_representation_structure.csv",
|
| 345 |
"baselines/table_3_retrieval_baselines_and_controls.csv",
|
|
|
|
| 346 |
"scripts/download_audio_recordings.py",
|
| 347 |
"scripts/run_smoke_eval.py",
|
|
|
|
|
|
|
| 348 |
],
|
| 349 |
errors,
|
| 350 |
)
|
|
|
|
| 29 |
re.compile(r"\\Users\\"),
|
| 30 |
]
|
| 31 |
|
| 32 |
+
DRAFT_TEXT_PATTERNS = [
|
| 33 |
+
re.compile(r"\b" + "T" + "BD" + r"\b", re.IGNORECASE),
|
| 34 |
+
re.compile(r"\b" + "TO" + "DO" + r"\b", re.IGNORECASE),
|
| 35 |
+
re.compile("place" + "holder", re.IGNORECASE),
|
| 36 |
]
|
| 37 |
|
| 38 |
JUNK_FILE_NAMES = {".DS_Store", "Thumbs.db"}
|
|
|
|
| 111 |
for pattern in LOCAL_PATH_PATTERNS:
|
| 112 |
if pattern.search(text):
|
| 113 |
add_issue(errors, "local_path_leak", f"Local path pattern found in {rel}: {pattern.pattern}")
|
| 114 |
+
for pattern in DRAFT_TEXT_PATTERNS:
|
| 115 |
if pattern.search(text):
|
| 116 |
+
add_issue(warnings, "draft_text", f"Draft-like marker found in {rel}: {pattern.pattern}")
|
| 117 |
|
| 118 |
|
| 119 |
def validate_ids(root: Path, errors: list[dict[str, str]], warnings: list[dict[str, str]]) -> dict[str, int]:
|
|
|
|
| 341 |
"manifests/splits.csv",
|
| 342 |
"features/feature_table_default.csv.gz",
|
| 343 |
"features/feature_table_strict_clean.csv.gz",
|
| 344 |
+
"features/bottlenecks/default/standard_rvq_8bit_feature_table.csv.gz",
|
| 345 |
+
"features/bottlenecks/default/pq_8bit_feature_table.csv.gz",
|
| 346 |
+
"features/bottlenecks/default/opq_8bit_feature_table.csv.gz",
|
| 347 |
+
"features/bottlenecks/strict_clean/standard_rvq_8bit_feature_table.csv.gz",
|
| 348 |
+
"features/bottlenecks/strict_clean/pq_8bit_feature_table.csv.gz",
|
| 349 |
+
"features/bottlenecks/strict_clean/opq_8bit_feature_table.csv.gz",
|
| 350 |
"baselines/table_2_representation_structure.csv",
|
| 351 |
"baselines/table_3_retrieval_baselines_and_controls.csv",
|
| 352 |
+
"requirements.txt",
|
| 353 |
"scripts/download_audio_recordings.py",
|
| 354 |
"scripts/run_smoke_eval.py",
|
| 355 |
+
"scripts/run_retrieval_eval.py",
|
| 356 |
+
"scripts/run_release_evaluation.py",
|
| 357 |
],
|
| 358 |
errors,
|
| 359 |
)
|