id
int64 599M
3.26B
| number
int64 1
7.7k
| title
stringlengths 1
290
| body
stringlengths 0
228k
⌀ | state
stringclasses 2
values | html_url
stringlengths 46
51
| created_at
timestamp[s]date 2020-04-14 10:18:02
2025-07-23 08:04:53
| updated_at
timestamp[s]date 2020-04-27 16:04:17
2025-07-23 18:53:44
| closed_at
timestamp[s]date 2020-04-14 12:01:40
2025-07-23 16:44:42
⌀ | user
dict | labels
listlengths 0
4
| is_pull_request
bool 2
classes | comments
listlengths 0
0
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|
988,276,859
| 2,870
|
Fix three typos in two files for documentation
|
Changed "bacth_size" to "batch_size" (2x)
Changed "intsructions" to "instructions"
|
closed
|
https://github.com/huggingface/datasets/pull/2870
| 2021-09-04T11:49:43
| 2021-09-06T08:21:21
| 2021-09-06T08:19:35
|
{
"login": "leny-mi",
"id": 25124853,
"type": "User"
}
|
[] | true
|
[] |
987,676,420
| 2,869
|
TypeError: 'NoneType' object is not callable
|
## Describe the bug
TypeError: 'NoneType' object is not callable
## Steps to reproduce the bug
```python
from datasets import load_dataset, load_metric
dataset = datasets.load_dataset("glue", 'cola')
```
## Expected results
A clear and concise description of the expected results.
## Actual results
Specify the actual results or traceback.
## Environment info
<!-- You can run the command `datasets-cli env` and copy-and-paste its output below. -->
- `datasets` version: 1.11.0
- Platform:
- Python version: 3.7
- PyArrow version:
|
closed
|
https://github.com/huggingface/datasets/issues/2869
| 2021-09-03T11:27:39
| 2025-02-19T09:57:34
| 2021-09-08T09:24:55
|
{
"login": "Chenfei-Kang",
"id": 40911446,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
987,139,146
| 2,868
|
Add Common Objects in 3D (CO3D)
|
## Adding a Dataset
- **Name:** *Common Objects in 3D (CO3D)*
- **Description:** *See blog post [here](https://ai.facebook.com/blog/common-objects-in-3d-dataset-for-3d-reconstruction)*
- **Paper:** *[link to paper](https://arxiv.org/abs/2109.00512)*
- **Data:** *[link to data](https://ai.facebook.com/datasets/co3d-downloads/)*
- **Motivation:** *excerpt from above blog post:*
> As the first data set of its kind, CO3D will aptly enable reconstruction of real-life 3D objects. Indeed, CO3D already provides training data to enable our NeRFormer to tackle the new-view synthesis (NVS) task. Here, photorealistic NVS is a major step on the path to fully immersive AR/VR effects, where objects can be virtually transported across different environments, which will allow connecting users by sharing or recollecting their experiences.
>
> Besides practical applications in AR/VR, we hope that the data set will become a standard testbed for the recent proliferation of methods (including NeRFormer, Implicit Differentiable Renderer, NeRF, and others) that reconstruct 3D scenes by means of an implicit shape model.
>
Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md).
|
open
|
https://github.com/huggingface/datasets/issues/2868
| 2021-09-02T20:36:12
| 2024-01-17T12:03:59
| null |
{
"login": "nateraw",
"id": 32437151,
"type": "User"
}
|
[
{
"name": "dataset request",
"color": "e99695"
},
{
"name": "vision",
"color": "bfdadc"
}
] | false
|
[] |
986,971,224
| 2,867
|
Add CaSiNo dataset
|
Hi. I request you to add our dataset to the repository.
This data was recently published at NAACL 2021: https://aclanthology.org/2021.naacl-main.254.pdf
|
closed
|
https://github.com/huggingface/datasets/pull/2867
| 2021-09-02T17:06:23
| 2021-09-16T15:12:54
| 2021-09-16T09:23:44
|
{
"login": "kushalchawla",
"id": 8416863,
"type": "User"
}
|
[] | true
|
[] |
986,706,676
| 2,866
|
"counter" dataset raises an error in normal mode, but not in streaming mode
|
## Describe the bug
`counter` dataset raises an error on `load_dataset()`, but simply returns an empty iterator in streaming mode.
## Steps to reproduce the bug
```python
>>> import datasets as ds
>>> a = ds.load_dataset('counter', split="train", streaming=False)
Using custom data configuration default
Downloading and preparing dataset counter/default (download: 1.29 MiB, generated: 2.48 MiB, post-processed: Unknown size, total: 3.77 MiB) to /home/slesage/.cache/huggingface/datasets/counter/default/1.0.0/9f84962fa0f35bec5a34fe0bdff8681838d497008c457f7856c48654476ec0e9...
Traceback (most recent call last):
File "/home/slesage/hf/datasets-preview-backend/.venv/lib/python3.8/site-packages/datasets/builder.py", line 726, in _download_and_prepare
self._prepare_split(split_generator, **prepare_split_kwargs)
File "/home/slesage/hf/datasets-preview-backend/.venv/lib/python3.8/site-packages/datasets/builder.py", line 1124, in _prepare_split
for key, record in utils.tqdm(
File "/home/slesage/hf/datasets-preview-backend/.venv/lib/python3.8/site-packages/tqdm/std.py", line 1185, in __iter__
for obj in iterable:
File "/home/slesage/.cache/huggingface/modules/datasets_modules/datasets/counter/9f84962fa0f35bec5a34fe0bdff8681838d497008c457f7856c48654476ec0e9/counter.py", line 161, in _generate_examples
with derived_file.open(encoding="utf-8") as f:
File "/home/slesage/.pyenv/versions/3.8.11/lib/python3.8/pathlib.py", line 1222, in open
return io.open(self, mode, buffering, encoding, errors, newline,
File "/home/slesage/.pyenv/versions/3.8.11/lib/python3.8/pathlib.py", line 1078, in _opener
return self._accessor.open(self, flags, mode)
FileNotFoundError: [Errno 2] No such file or directory: '/home/slesage/.cache/huggingface/datasets/downloads/extracted/b57aa6db5601a738e57b95c1fd8cced54ff28fc540efcdaf0f6c4f1bb5dfe211/COUNTER/0032p.xml'
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/slesage/hf/datasets-preview-backend/.venv/lib/python3.8/site-packages/datasets/load.py", line 1112, in load_dataset
builder_instance.download_and_prepare(
File "/home/slesage/hf/datasets-preview-backend/.venv/lib/python3.8/site-packages/datasets/builder.py", line 636, in download_and_prepare
self._download_and_prepare(
File "/home/slesage/hf/datasets-preview-backend/.venv/lib/python3.8/site-packages/datasets/builder.py", line 728, in _download_and_prepare
raise OSError(
OSError: Cannot find data file.
Original error:
[Errno 2] No such file or directory: '/home/slesage/.cache/huggingface/datasets/downloads/extracted/b57aa6db5601a738e57b95c1fd8cced54ff28fc540efcdaf0f6c4f1bb5dfe211/COUNTER/0032p.xml'
```
```python
>>> import datasets as ds
>>> b = ds.load_dataset('counter', split="train", streaming=True)
Using custom data configuration default
>>> list(b)
[]
```
## Expected results
An exception should be raised in streaming mode
## Actual results
No exception is raised in streaming mode: there is no way to tell if something has broken or if the dataset is simply empty.
## Environment info
- `datasets` version: 1.11.1.dev0
- Platform: Linux-5.11.0-1016-aws-x86_64-with-glibc2.29
- Python version: 3.8.11
- PyArrow version: 4.0.1
|
closed
|
https://github.com/huggingface/datasets/issues/2866
| 2021-09-02T13:10:53
| 2021-10-14T09:24:09
| 2021-10-14T09:24:09
|
{
"login": "severo",
"id": 1676121,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
986,460,698
| 2,865
|
Add MultiEURLEX dataset
|
**Add new MultiEURLEX Dataset**
MultiEURLEX comprises 65k EU laws in 23 official EU languages (some low-ish resource). Each EU law has been annotated with EUROVOC concepts (labels) by the Publication Office of EU. As with the English EURLEX, the goal is to predict the relevant EUROVOC concepts (labels); this is multi-label classification task (given the text, predict multiple labels).
|
closed
|
https://github.com/huggingface/datasets/pull/2865
| 2021-09-02T09:42:24
| 2021-09-10T11:50:06
| 2021-09-10T11:50:06
|
{
"login": "iliaschalkidis",
"id": 1626984,
"type": "User"
}
|
[] | true
|
[] |
986,159,438
| 2,864
|
Fix data URL in ToTTo dataset
|
Data source host changed their data URL: google-research-datasets/ToTTo@cebeb43.
Fix #2860.
|
closed
|
https://github.com/huggingface/datasets/pull/2864
| 2021-09-02T05:25:08
| 2021-09-02T06:47:40
| 2021-09-02T06:47:40
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
986,156,755
| 2,863
|
Update dataset URL
| null |
closed
|
https://github.com/huggingface/datasets/pull/2863
| 2021-09-02T05:22:18
| 2021-09-02T08:10:50
| 2021-09-02T08:10:50
|
{
"login": "mrm8488",
"id": 3653789,
"type": "User"
}
|
[] | true
|
[] |
985,081,871
| 2,861
|
fix: 🐛 be more specific when catching exceptions
|
The same specific exception is catched in other parts of the same
function.
|
closed
|
https://github.com/huggingface/datasets/pull/2861
| 2021-09-01T12:18:12
| 2021-09-02T09:53:36
| 2021-09-02T09:52:03
|
{
"login": "severo",
"id": 1676121,
"type": "User"
}
|
[] | true
|
[] |
985,013,339
| 2,860
|
Cannot download TOTTO dataset
|
Error: Couldn't find file at https://storage.googleapis.com/totto/totto_data.zip
`datasets version: 1.11.0`
# How to reproduce:
```py
from datasets import load_dataset
dataset = load_dataset('totto')
```
|
closed
|
https://github.com/huggingface/datasets/issues/2860
| 2021-09-01T11:04:10
| 2021-09-02T06:47:40
| 2021-09-02T06:47:40
|
{
"login": "mrm8488",
"id": 3653789,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
984,324,500
| 2,859
|
Loading allenai/c4 in streaming mode does too many HEAD requests
|
This does 60,000+ HEAD requests to get all the ETags of all the data files:
```python
from datasets import load_dataset
load_dataset("allenai/c4", streaming=True)
```
It makes loading the dataset completely impractical.
The ETags are used to compute the config id (it must depend on the data files being used).
Instead of using the ETags, we could simply use the commit hash of the dataset repository on the hub, as well and the glob pattern used to resolve the files (here it's `*` by default, to load all the files of the repository)
|
closed
|
https://github.com/huggingface/datasets/issues/2859
| 2021-08-31T21:11:04
| 2021-10-12T07:35:52
| 2021-10-11T11:05:51
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
},
{
"name": "streaming",
"color": "fef2c0"
}
] | false
|
[] |
984,145,568
| 2,858
|
Fix s3fs version in CI
|
The latest s3fs version has new constrains on aiobotocore, and therefore on boto3 and botocore
This PR changes the constrains to avoid the new conflicts
In particular it pins the version of s3fs.
|
closed
|
https://github.com/huggingface/datasets/pull/2858
| 2021-08-31T18:05:43
| 2021-09-06T13:33:35
| 2021-08-31T21:29:51
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
984,093,938
| 2,857
|
Update: Openwebtext - update size
|
Update the size of the Openwebtext dataset
I also regenerated the dataset_infos.json but the data file checksum didn't change, and the number of examples either (8013769 examples)
Close #2839, close #726.
|
closed
|
https://github.com/huggingface/datasets/pull/2857
| 2021-08-31T17:11:03
| 2022-02-15T10:38:03
| 2021-09-07T09:44:32
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
983,876,734
| 2,856
|
fix: 🐛 remove URL's query string only if it's ?dl=1
|
A lot of URL use the query strings, for example
http://opus.nlpl.eu/download.php?f=Bianet/v1/moses/en-ku.txt.zip, we
must not remove it when trying to detect the protocol. We thus remove it
only in the case of the query string being ?dl=1 which occurs on dropbox
and dl.orangedox.com. Also: add unit tests.
See https://github.com/huggingface/datasets/pull/2843 for the original
discussion.
|
closed
|
https://github.com/huggingface/datasets/pull/2856
| 2021-08-31T13:40:07
| 2021-08-31T14:22:12
| 2021-08-31T14:22:12
|
{
"login": "severo",
"id": 1676121,
"type": "User"
}
|
[] | true
|
[] |
983,858,229
| 2,855
|
Fix windows CI CondaError
|
From this thread: https://github.com/conda/conda/issues/6057
We can fix the conda error
```
CondaError: Cannot link a source that does not exist.
C:\Users\...\Anaconda3\Scripts\conda.exe
```
by doing
```bash
conda update conda
```
before doing any install in the windows CI
|
closed
|
https://github.com/huggingface/datasets/pull/2855
| 2021-08-31T13:22:02
| 2021-08-31T13:35:34
| 2021-08-31T13:35:33
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
983,726,084
| 2,854
|
Fix caching when moving script
|
When caching the result of a `map` function, the hash that is computed depends on many properties of this function, such as all the python objects it uses, its code and also the location of this code.
Using the full path of the python script for the location of the code makes the hash change if a script like `run_mlm.py` is moved.
I changed this by simply using the base name of the script instead of the full path.
Note that this change also affects the hash of the code used from imported modules, but I think it's fine. Indeed it hashes the code of the imported modules anyway, so the location of the python files of the imported modules doesn't matter when computing the hash.
Close https://github.com/huggingface/datasets/issues/2825
|
closed
|
https://github.com/huggingface/datasets/pull/2854
| 2021-08-31T10:58:35
| 2021-08-31T13:13:36
| 2021-08-31T13:13:36
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
983,692,026
| 2,853
|
Add AMI dataset
|
This is an initial commit for AMI dataset
|
closed
|
https://github.com/huggingface/datasets/pull/2853
| 2021-08-31T10:19:01
| 2021-09-29T09:19:19
| 2021-09-29T09:19:19
|
{
"login": "cahya-wirawan",
"id": 7669893,
"type": "User"
}
|
[] | true
|
[] |
983,609,352
| 2,852
|
Fix: linnaeus - fix url
|
The url was causing a `ConnectionError` because of the "/" at the end
Close https://github.com/huggingface/datasets/issues/2821
|
closed
|
https://github.com/huggingface/datasets/pull/2852
| 2021-08-31T08:51:13
| 2021-08-31T13:12:10
| 2021-08-31T13:12:09
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
982,789,593
| 2,851
|
Update `column_names` showed as `:func:` in exploring.st
|
Hi,
One mention of `column_names` in exploring.st was showing it as `:func:` instead of `:attr:`.
|
closed
|
https://github.com/huggingface/datasets/pull/2851
| 2021-08-30T13:21:46
| 2021-09-01T08:42:11
| 2021-08-31T14:45:46
|
{
"login": "ClementRomac",
"id": 8899812,
"type": "User"
}
|
[] | true
|
[] |
982,654,644
| 2,850
|
Wound segmentation datasets
|
## Adding a Dataset
- **Name:** Wound segmentation datasets
- **Description:** annotated wound image dataset
- **Paper:** https://www.nature.com/articles/s41598-020-78799-w
- **Data:** https://github.com/uwm-bigdata/wound-segmentation
- **Motivation:** Interesting simple image dataset, useful for segmentation, with visibility due to http://www.miccai.org/special-interest-groups/challenges/ and https://fusc.grand-challenge.org/
Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md).
|
open
|
https://github.com/huggingface/datasets/issues/2850
| 2021-08-30T10:44:32
| 2021-12-08T12:02:00
| null |
{
"login": "osanseviero",
"id": 7246357,
"type": "User"
}
|
[
{
"name": "dataset request",
"color": "e99695"
},
{
"name": "vision",
"color": "bfdadc"
}
] | false
|
[] |
982,631,420
| 2,849
|
Add Open Catalyst Project Dataset
|
## Adding a Dataset
- **Name:** Open Catalyst 2020 (OC20) Dataset
- **Website:** https://opencatalystproject.org/
- **Data:** https://github.com/Open-Catalyst-Project/ocp/blob/master/DATASET.md
Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md).
|
open
|
https://github.com/huggingface/datasets/issues/2849
| 2021-08-30T10:14:39
| 2021-08-30T10:14:39
| null |
{
"login": "osanseviero",
"id": 7246357,
"type": "User"
}
|
[
{
"name": "dataset request",
"color": "e99695"
}
] | false
|
[] |
981,953,908
| 2,848
|
Update README.md
|
Changed 'Tain' to 'Train'.
|
closed
|
https://github.com/huggingface/datasets/pull/2848
| 2021-08-28T23:58:26
| 2021-09-07T09:40:32
| 2021-09-07T09:40:32
|
{
"login": "odellus",
"id": 4686956,
"type": "User"
}
|
[] | true
|
[] |
981,589,693
| 2,847
|
fix regex to accept negative timezone
|
fix #2846
|
closed
|
https://github.com/huggingface/datasets/pull/2847
| 2021-08-27T20:54:05
| 2021-09-13T20:39:50
| 2021-09-07T09:34:23
|
{
"login": "jadermcs",
"id": 7156771,
"type": "User"
}
|
[] | true
|
[] |
981,587,590
| 2,846
|
Negative timezone
|
## Describe the bug
The load_dataset method do not accept a parquet file with a negative timezone, as it has the following regex:
```
"^(s|ms|us|ns),\s*tz=([a-zA-Z0-9/_+:]*)$"
```
So a valid timestap ```timestamp[us, tz=-03:00]``` returns an error when loading parquet files.
## Steps to reproduce the bug
```python
# Where the timestamp column has a tz of -03:00
datasets = load_dataset('parquet', data_files={'train': train_files, 'validation': validation_files,
'test': test_files}, cache_dir="./cache_teste/")
```
## Expected results
The -03:00 is a valid tz so the regex should accept this without raising an error.
## Actual results
As this regex disaproves a valid tz it raises the following error:
```python
raise ValueError(
f"{datasets_dtype} is not a validly formatted string representation of a pyarrow timestamp."
f"Examples include timestamp[us] or timestamp[us, tz=America/New_York]"
f"See: https://arrow.apache.org/docs/python/generated/pyarrow.timestamp.html#pyarrow.timestamp"
)
```
## Environment info
<!-- You can run the command `datasets-cli env` and copy-and-paste its output below. -->
- `datasets` version: 1.11.0
- Platform: Ubuntu 20.04
- Python version: 3.8
- PyArrow version: 5.0.0
|
closed
|
https://github.com/huggingface/datasets/issues/2846
| 2021-08-27T20:50:33
| 2021-09-10T11:51:07
| 2021-09-10T11:51:07
|
{
"login": "jadermcs",
"id": 7156771,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
981,487,861
| 2,845
|
[feature request] adding easy to remember `datasets.cache_dataset()` + `datasets.is_dataset_cached()`
|
Often, there is a need to prepare a dataset but not use it immediately, e.g. think tests suite setup, so it'd be really useful to be able to do:
```
if not datasets.is_dataset_cached(ds): datasets.cache_dataset(ds)
```
This can already be done with:
```
builder = load_dataset_builder(ds)
if not os.path.idsir(builder.cache_dir):
builder.download_and_prepare()
```
but the current way is a way less intuitive and much harder to remember than the proposed API, IMHO.
One more way is to do:
```
_ = load_dataset(ds)
```
but it wastes resources loading the dataset when it's not needed.
this has been discussed at https://huggingface.slack.com/archives/C01229B19EX/p1630021912025800
Thank you!
@lhoestq
|
open
|
https://github.com/huggingface/datasets/issues/2845
| 2021-08-27T18:21:51
| 2021-08-27T18:24:05
| null |
{
"login": "stas00",
"id": 10676103,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
981,382,806
| 2,844
|
Fix: wikicorpus - fix keys
|
As mentioned in https://github.com/huggingface/datasets/issues/2552, there is a duplicate keys error in `wikicorpus`.
I fixed that by taking into account the file index in the keys
|
closed
|
https://github.com/huggingface/datasets/pull/2844
| 2021-08-27T15:56:06
| 2021-09-06T14:07:28
| 2021-09-06T14:07:27
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
981,317,775
| 2,843
|
Fix extraction protocol inference from urls with params
|
Previously it was unable to infer the compression protocol for files at URLs like
```
https://foo.bar/train.json.gz?dl=1
```
because of the query parameters.
I fixed that, this should allow 10+ datasets to work in streaming mode:
```
"discovery",
"emotion",
"grail_qa",
"guardian_authorship",
"pragmeval",
"simple_questions_v2",
"versae/adobo",
"w-nicole/childes_data",
"w-nicole/childes_data_no_tags_",
"w-nicole/childes_data_with_tags",
"w-nicole/childes_data_with_tags_"
```
cc @severo
|
closed
|
https://github.com/huggingface/datasets/pull/2843
| 2021-08-27T14:40:57
| 2021-08-30T17:11:49
| 2021-08-30T13:12:01
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
980,725,899
| 2,842
|
always requiring the username in the dataset name when there is one
|
Me and now another person have been bitten by the `datasets`'s non-strictness on requiring a dataset creator's username when it's due.
So both of us started with `stas/openwebtext-10k`, somewhere along the lines lost `stas/` and continued using `openwebtext-10k` and it all was good until we published the software and things broke, since there is no `openwebtext-10k`
So this feature request is asking to tighten the checking and not allow dataset loading if it was downloaded with the user prefix, but then attempted to be used w/o it.
The same in code:
```
# first run
python -c "from datasets import load_dataset; load_dataset('stas/openwebtext-10k')"
# now run immediately
python -c "from datasets import load_dataset; load_dataset('openwebtext-10k')"
# the second command should fail, but it doesn't fail now.
```
Please let me know if I explained myself clearly.
Thank you!
|
closed
|
https://github.com/huggingface/datasets/issues/2842
| 2021-08-26T23:31:53
| 2021-10-22T09:43:35
| 2021-10-22T09:43:35
|
{
"login": "stas00",
"id": 10676103,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
980,497,321
| 2,841
|
Adding GLUECoS Hinglish and Spanglish code-switching bemchmark
|
## Adding a Dataset
- **Name:** GLUECoS
- **Description:** a Microsoft Benchmark to evaluate code-switching for only two language pairs but a variety of tasks
- **Paper:** https://aclanthology.org/2020.acl-main.329/
- **Data:** https://github.com/microsoft/GLUECoS
- **Motivation:** We currently only have [one other](https://huggingface.co/datasets/lince) dataset for code-switching
Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md).
|
open
|
https://github.com/huggingface/datasets/issues/2841
| 2021-08-26T17:47:39
| 2021-10-20T18:41:20
| null |
{
"login": "yjernite",
"id": 10469459,
"type": "User"
}
|
[
{
"name": "dataset request",
"color": "e99695"
}
] | false
|
[] |
980,489,074
| 2,840
|
How can I compute BLEU-4 score use `load_metric` ?
|
I have found the sacrebleu metric. But, I do not know the difference between it and BLEU-4.
If I want to compute BLEU-4 score, what can i do?
|
closed
|
https://github.com/huggingface/datasets/issues/2840
| 2021-08-26T17:36:37
| 2021-08-27T08:13:24
| 2021-08-27T08:13:24
|
{
"login": "Doragd",
"id": 26213546,
"type": "User"
}
|
[] | false
|
[] |
980,271,715
| 2,839
|
OpenWebText: NonMatchingSplitsSizesError
|
## Describe the bug
When downloading `openwebtext`, I'm getting:
```
datasets.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='train', num_bytes=39769494896, num_examples=8013769, dataset_name='openwebtext'), 'recorded': SplitInfo(name='train', num_bytes=39611023912, num_examples=7982430, dataset_name='openwebtext')}]
```
I suspect that the file we download from has changed since the size doesn't look like to match with documentation
`Downloading: 0%| | 0.00/12.9G [00:00<?, ?B/s]` This suggest the total size is 12.9GB, whereas the one documented mentions `Size of downloaded dataset files: 12283.35 MB`.
## Steps to reproduce the bug
```python
from datasets import load_dataset
load_dataset("openwebtext", download_mode="force_redownload")
```
## Expected results
Loading is successful
## Actual results
Loading throws above error.
## Environment info
<!-- You can run the command `datasets-cli env` and copy-and-paste its output below. -->
- `datasets` version: 1.10.2
- Platform: linux (Redhat version 8.1)
- Python version: 3.8
- PyArrow version: 4.0.1
|
closed
|
https://github.com/huggingface/datasets/issues/2839
| 2021-08-26T13:50:26
| 2021-09-21T14:12:40
| 2021-09-21T14:09:43
|
{
"login": "thomasw21",
"id": 24695242,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
980,067,186
| 2,838
|
Add error_bad_chunk to the JSON loader
|
Add the `error_bad_chunk` parameter to the JSON loader.
Setting `error_bad_chunk=False` allows to skip an unparsable chunk of JSON data without raising an error.
Additional note:
In case of an unparsable JSON chunk, the JSON loader no longer tries to load the full JSON (which could take a lot of time in streaming mode) to get the JSON fields that the user may have forgotten to pass. Ex : for squad-like data, the user has to pass `field="data"` to tell the loader to get the list of examples from this field.
TODO: update docs
cc @lvwerra
|
open
|
https://github.com/huggingface/datasets/pull/2838
| 2021-08-26T10:07:32
| 2023-09-25T09:06:42
| null |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
979,298,297
| 2,837
|
prepare_module issue when loading from read-only fs
|
## Describe the bug
When we use prepare_module from a readonly file system, we create a FileLock using the `local_path`.
This path is not necessarily writable.
`lock_path = local_path + ".lock"`
## Steps to reproduce the bug
Run `load_dataset` on a readonly python loader file.
```python
ds = load_dataset(
python_loader, data_files={"train": train_path, "test": test_path}
)
```
where `python_loader` is a path to a file located in a readonly folder.
## Expected results
This should work I think?
## Actual results
```python
return load_dataset(
File "/usr/local/lib/python3.8/dist-packages/datasets/load.py", line 711, in load_dataset
module_path, hash, resolved_file_path = prepare_module(
File "/usr/local/lib/python3.8/dist-packages/datasets/load.py", line 465, in prepare_module
with FileLock(lock_path):
File "/usr/local/lib/python3.8/dist-packages/datasets/utils/filelock.py", line 314, in __enter__
self.acquire()
File "/usr/local/lib/python3.8/dist-packages/datasets/utils/filelock.py", line 263, in acquire
self._acquire()
File "/usr/local/lib/python3.8/dist-packages/datasets/utils/filelock.py", line 378, in _acquire
fd = os.open(self._lock_file, open_mode)
OSError: [Errno 30] Read-only file system: 'YOUR_FILE.py.lock'
```
## Environment info
<!-- You can run the command `datasets-cli env` and copy-and-paste its output below. -->
- `datasets` version: 1.7.0
- Platform: macOS-10.15.7-x86_64-i386-64bit
- Python version: 3.8.8
- PyArrow version: 3.0.0
|
closed
|
https://github.com/huggingface/datasets/issues/2837
| 2021-08-25T15:21:26
| 2021-10-05T17:58:22
| 2021-10-05T17:58:22
|
{
"login": "Dref360",
"id": 8976546,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
979,230,142
| 2,836
|
Optimize Dataset.filter to only compute the indices to keep
|
Optimize `Dataset.filter` to only compute the indices of the rows to keep, instead of creating a new Arrow table with the rows to keep. Creating a new table was an issue because it could take a lot of disk space.
This will be useful to process audio datasets for example cc @patrickvonplaten
|
closed
|
https://github.com/huggingface/datasets/pull/2836
| 2021-08-25T14:41:22
| 2021-09-14T14:51:53
| 2021-09-13T15:50:21
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
979,209,394
| 2,835
|
Update: timit_asr - make the dataset streamable
|
The TIMIT ASR dataset had two issues that was preventing it from being streamable:
1. it was missing a call to `open` before `pd.read_csv`
2. it was using `os.path.dirname` which is not supported for streaming
I made the dataset streamable by using `open` to load the CSV, and by adding the support for `os.path.dirname` in dataset scripts to stream data
You can now do
```python
from datasets import load_dataset
timit_asr = load_dataset("timit_asr", streaming=True)
print(next(iter(timit_asr["train"])))
```
prints:
```json
{"file": "zip://data/TRAIN/DR4/MMDM0/SI681.WAV::https://data.deepai.org/timit.zip",
"phonetic_detail": {"start": [0, 1960, 2466, 3480, 4000, 5960, 7480, 7880, 9400, 9960, 10680, 13480, 15680, 15880, 16920, 18297, 18882, 19480, 21723, 22516, 24040, 25190, 27080, 28160, 28560, 30120, 31832, 33240, 34640, 35968, 37720],
"utterance": ["h#", "w", "ix", "dcl", "s", "ah", "tcl", "ch", "ix", "n", "ae", "kcl", "t", "ix", "v", "r", "ix", "f", "y", "ux", "zh", "el", "bcl", "b", "iy", "y", "ux", "s", "f", "el", "h#"],
"stop": [1960, 2466, 3480, 4000, 5960, 7480, 7880, 9400, 9960, 10680, 13480, 15680, 15880, 16920, 18297, 18882, 19480, 21723, 22516, 24040, 25190, 27080, 28160, 28560, 30120, 31832, 33240, 34640, 35968, 37720, 39920]},
"sentence_type": "SI", "id": "SI681",
"speaker_id": "MMDM0",
"dialect_region": "DR4",
"text": "Would such an act of refusal be useful?",
"word_detail": {
"start": [1960, 4000, 9400, 10680, 15880, 18297, 27080, 30120],
"utterance": ["would", "such", "an", "act", "of", "refusal", "be", "useful"],
"stop": [4000, 9400, 10680, 15880, 18297, 27080, 30120, 37720]
}}
```
cc @patrickvonplaten @vrindaprabhu
|
closed
|
https://github.com/huggingface/datasets/pull/2835
| 2021-08-25T14:22:49
| 2021-09-07T13:15:47
| 2021-09-07T13:15:46
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
978,309,749
| 2,834
|
Fix IndexError by ignoring empty RecordBatch
|
We need to ignore the empty record batches for the interpolation search to work correctly when querying arrow tables
Close #2833
cc @SaulLu
|
closed
|
https://github.com/huggingface/datasets/pull/2834
| 2021-08-24T17:06:13
| 2021-08-24T17:21:18
| 2021-08-24T17:21:18
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
978,296,140
| 2,833
|
IndexError when accessing first element of a Dataset if first RecordBatch is empty
|
The computation of the offsets of the underlying Table of a Dataset has some issues if the first RecordBatch is empty.
```python
from datasets import Dataset
import pyarrow as pa
pa_table = pa.Table.from_pydict({"a": [1]})
pa_table2 = pa.Table.from_pydict({"a": []}, schema=pa_table.schema)
ds_table = pa.concat_tables([pa_table2, pa_table])
dataset = Dataset(ds_table)
print([len(b) for b in dataset.data._batches])
# [0, 1]
print(dataset.data._offsets)
# [0 0 1] (should be [0, 1])
dataset[0]
```
raises
```python
---------------------------------------------------------------------------
IndexError Traceback (most recent call last)
/usr/local/lib/python3.7/dist-packages/datasets/table.py in _interpolation_search(arr, x)
90 else:
91 i, j = i, k
---> 92 raise IndexError(f"Invalid query '{x}' for size {arr[-1] if len(arr) else 'none'}.")
93
94
IndexError: Invalid query '0' for size 1.
```
This can be fixed by ignoring empty batches when computing `table._batches` and `table._offsets`
cc @SaulLu
|
closed
|
https://github.com/huggingface/datasets/issues/2833
| 2021-08-24T16:49:20
| 2021-08-24T17:21:17
| 2021-08-24T17:21:17
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | false
|
[] |
978,012,800
| 2,832
|
Logging levels not taken into account
|
## Describe the bug
The `logging` module isn't working as intended relative to the levels to set.
## Steps to reproduce the bug
```python
from datasets import logging
logging.set_verbosity_debug()
logger = logging.get_logger()
logger.error("ERROR")
logger.warning("WARNING")
logger.info("INFO")
logger.debug("DEBUG"
```
## Expected results
I expect all logs to be output since I'm putting a `debug` level.
## Actual results
Only the two first logs are output.
## Environment info
- `datasets` version: 1.11.0
- Platform: Linux-5.13.9-arch1-1-x86_64-with-glibc2.33
- Python version: 3.9.6
- PyArrow version: 5.0.0
## To go further
This logging issue appears in `datasets` but not in `transformers`. It happens because there is no handler defined for the logger. When no handler is defined, the `logging` library will output a one-off error to stderr, using a `StderrHandler` with level `WARNING`.
`transformers` sets a default `StreamHandler` [here](https://github.com/huggingface/transformers/blob/5c6eca71a983bae2589eed01e5c04fcf88ba5690/src/transformers/utils/logging.py#L86)
|
closed
|
https://github.com/huggingface/datasets/issues/2832
| 2021-08-24T11:50:41
| 2023-07-12T17:19:30
| 2023-07-12T17:19:29
|
{
"login": "LysandreJik",
"id": 30755778,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
977,864,600
| 2,831
|
ArrowInvalid when mapping dataset with missing values
|
## Describe the bug
I encountered an `ArrowInvalid` when mapping dataset with missing values.
Here are the files for a minimal example. The exception is only thrown when the first line in the csv has a missing value (if you move the last line to the top it isn't thrown).
[data_small.csv](https://github.com/huggingface/datasets/files/7037838/data_small.csv)
[data.csv](https://github.com/huggingface/datasets/files/7037842/data.csv)
## Steps to reproduce the bug
```python
from datasets import load_dataset
datasets = load_dataset("csv", data_files=['data_small.csv'])
datasets = datasets.map(lambda e: {'labels': e['match']},
remove_columns=['id'])
```
## Expected results
No error
## Actual results
```
File "pyarrow/error.pxi", line 84, in pyarrow.lib.check_status
pyarrow.lib.ArrowInvalid: Invalid null value
```
## Environment info
- `datasets` version: 1.5.0
- Platform: Linux-5.11.0-25-generic-x86_64-with-glibc2.29
- Python version: 3.8.10
- PyTorch version (GPU?): 1.7.1+cpu (False)
- Tensorflow version (GPU?): 2.4.1 (False)
- Using GPU in script?: no
- Using distributed or parallel set-up in script?: no
|
open
|
https://github.com/huggingface/datasets/issues/2831
| 2021-08-24T08:50:42
| 2021-08-31T14:15:34
| null |
{
"login": "uniquefine",
"id": 12694730,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
977,563,947
| 2,830
|
Add imagefolder dataset
|
A generic imagefolder dataset inspired by `torchvision.datasets.ImageFolder`.
Resolves #2508
---
Example Usage:
[](https://colab.research.google.com/gist/nateraw/954fa8cba4ff806f6147a782fa9efd1a/imagefolder-official-example.ipynb)
|
closed
|
https://github.com/huggingface/datasets/pull/2830
| 2021-08-23T23:34:06
| 2022-03-01T16:29:44
| 2022-03-01T16:29:44
|
{
"login": "nateraw",
"id": 32437151,
"type": "User"
}
|
[] | true
|
[] |
977,233,360
| 2,829
|
Optimize streaming from TAR archives
|
Hi ! As you know TAR has some constraints for data streaming. While it is optimized for buffering, the files in the TAR archive **need to be streamed in order**. It means that we can't choose which file to stream from, and this notation is to be avoided for TAR archives:
```
tar://books_large_p1.txt::https://storage.googleapis.com/huggingface-nlp/datasets/bookcorpus/bookcorpus.tar.bz2
```
Instead, I suggest we implement `iter_archive` for the `StreamingDownloadManager`.
The regular `DownloadManager` already has it.
Then we will have to update the json/txt/csv/etc. loaders to make them use `iter_archive` on TAR archives.
That's also what Tensorflow Datasets is doing in this case.
See this [dataset](https://github.com/tensorflow/datasets/blob/93895059c80a9e05805e8f32a2e310f66a23fc98/tensorflow_datasets/image_classification/flowers.py) for example.
Therefore instead of doing
```python
uncompressed = dl_manager.extract(tar_archive)
filename = "books_large_p1.txt"
with open(os.path.join(uncompressed, filename)) as f:
for line in f:
...
```
we'll do
```python
for filename, f in dl_manager.iter_archive(tar_archive):
for line in f:
...
```
|
closed
|
https://github.com/huggingface/datasets/issues/2829
| 2021-08-23T16:56:40
| 2022-09-21T14:29:46
| 2022-09-21T14:08:39
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
},
{
"name": "streaming",
"color": "fef2c0"
}
] | false
|
[] |
977,181,517
| 2,828
|
Add code-mixed Kannada Hope speech dataset
|
## Adding a Dataset
- **Name:** *KanHope*
- **Description:** *A code-mixed English-Kannada dataset for Hope speech detection*
- **Paper:** *https://arxiv.org/abs/2108.04616*
- **Data:** *https://github.com/adeepH/KanHope/tree/main/dataset*
- **Motivation:** *The dataset is amongst the very few resources available for code-mixed low-resourced Dravidian languages of India*
|
closed
|
https://github.com/huggingface/datasets/pull/2828
| 2021-08-23T15:55:09
| 2021-10-01T17:21:03
| 2021-10-01T17:21:03
|
{
"login": "adeepH",
"id": 46108405,
"type": "User"
}
|
[] | true
|
[] |
976,976,552
| 2,827
|
add a text classification dataset
| null |
closed
|
https://github.com/huggingface/datasets/pull/2827
| 2021-08-23T12:24:41
| 2021-08-23T15:51:18
| 2021-08-23T15:51:18
|
{
"login": "adeepH",
"id": 46108405,
"type": "User"
}
|
[] | true
|
[] |
976,974,254
| 2,826
|
Add a Text Classification dataset: KanHope
|
## Adding a Dataset
- **Name:** *KanHope*
- **Description:** *A code-mixed English-Kannada dataset for Hope speech detection*
- **Paper:** *https://arxiv.org/abs/2108.04616* (I am the author of the paper}
- **Author:** *[AdeepH](https://github.com/adeepH)*
- **Data:** *https://github.com/adeepH/KanHope/tree/main/dataset*
- **Motivation:** *The dataset is amongst the very few resources available for code-mixed Dravidian languages*
- I tried following the steps as per the instructions. However, could not resolve an error. Any help would be appreciated.
- The dataset card and the scripts for the dataset *https://github.com/adeepH/datasets/tree/multilingual-hope-speech/datasets/mhs_eval*
```
Using custom data configuration default
Downloading and preparing dataset bn_hate_speech/default (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /root/.cache/huggingface/datasets/bn_hate_speech/default/0.0.0/5f417ddc89777278abd29988f909f39495f0ec802090f7d8fa63b5bffb121762...
---------------------------------------------------------------------------
KeyError Traceback (most recent call last)
<ipython-input-114-4a9cdb519e4c> in <module>()
1 from datasets import load_dataset
2
----> 3 data = load_dataset('/content/bn')
9 frames
/usr/local/lib/python3.7/dist-packages/datasets/load.py in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, ignore_verifications, keep_in_memory, save_infos, script_version, use_auth_token, task, streaming, **config_kwargs)
850 ignore_verifications=ignore_verifications,
851 try_from_hf_gcs=try_from_hf_gcs,
--> 852 use_auth_token=use_auth_token,
853 )
854
/usr/local/lib/python3.7/dist-packages/datasets/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, try_from_hf_gcs, dl_manager, base_path, use_auth_token, **download_and_prepare_kwargs)
614 if not downloaded_from_gcs:
615 self._download_and_prepare(
--> 616 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
617 )
618 # Sync info
/usr/local/lib/python3.7/dist-packages/datasets/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
691 try:
692 # Prepare split will record examples associated to the split
--> 693 self._prepare_split(split_generator, **prepare_split_kwargs)
694 except OSError as e:
695 raise OSError(
/usr/local/lib/python3.7/dist-packages/datasets/builder.py in _prepare_split(self, split_generator)
1107 disable=bool(logging.get_verbosity() == logging.NOTSET),
1108 ):
-> 1109 example = self.info.features.encode_example(record)
1110 writer.write(example, key)
1111 finally:
/usr/local/lib/python3.7/dist-packages/datasets/features.py in encode_example(self, example)
1015 """
1016 example = cast_to_python_objects(example)
-> 1017 return encode_nested_example(self, example)
1018
1019 def encode_batch(self, batch):
/usr/local/lib/python3.7/dist-packages/datasets/features.py in encode_nested_example(schema, obj)
863 if isinstance(schema, dict):
864 return {
--> 865 k: encode_nested_example(sub_schema, sub_obj) for k, (sub_schema, sub_obj) in utils.zip_dict(schema, obj)
866 }
867 elif isinstance(schema, (list, tuple)):
/usr/local/lib/python3.7/dist-packages/datasets/features.py in <dictcomp>(.0)
863 if isinstance(schema, dict):
864 return {
--> 865 k: encode_nested_example(sub_schema, sub_obj) for k, (sub_schema, sub_obj) in utils.zip_dict(schema, obj)
866 }
867 elif isinstance(schema, (list, tuple)):
/usr/local/lib/python3.7/dist-packages/datasets/features.py in encode_nested_example(schema, obj)
890 # ClassLabel will convert from string to int, TranslationVariableLanguages does some checks
891 elif isinstance(schema, (ClassLabel, TranslationVariableLanguages, Value, _ArrayXD)):
--> 892 return schema.encode_example(obj)
893 # Other object should be directly convertible to a native Arrow type (like Translation and Translation)
894 return obj
/usr/local/lib/python3.7/dist-packages/datasets/features.py in encode_example(self, example_data)
665 # If a string is given, convert to associated integer
666 if isinstance(example_data, str):
--> 667 example_data = self.str2int(example_data)
668
669 # Allowing -1 to mean no label.
/usr/local/lib/python3.7/dist-packages/datasets/features.py in str2int(self, values)
623 if value not in self._str2int:
624 value = str(value).strip()
--> 625 output.append(self._str2int[str(value)])
626 else:
627 # No names provided, try to integerize
KeyError: ' '
```
|
closed
|
https://github.com/huggingface/datasets/issues/2826
| 2021-08-23T12:21:58
| 2021-10-01T18:06:59
| 2021-10-01T18:06:59
|
{
"login": "adeepH",
"id": 46108405,
"type": "User"
}
|
[
{
"name": "dataset request",
"color": "e99695"
}
] | false
|
[] |
976,584,926
| 2,825
|
The datasets.map function does not load cached dataset after moving python script
|
## Describe the bug
The datasets.map function caches the processed data to a certain directory. When the map function is called another time with totally the same parameters, the cached data are supposed to be reloaded instead of re-processing. However, it doesn't reuse cached data sometimes. I use the common data processing in different tasks, the datasets are processed again, the only difference is that I run them in different files.
## Steps to reproduce the bug
Just run the following codes in different .py files.
```python
if __name__ == '__main__':
from datasets import load_dataset
from transformers import AutoTokenizer
raw_datasets = load_dataset("wikitext", "wikitext-2-raw-v1")
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
def tokenize_function(examples):
return tokenizer(examples["text"], padding="max_length", truncation=True)
tokenized_datasets = raw_datasets.map(tokenize_function, batched=True)
```
## Expected results
The map function should reload data in the second or any later runs.
## Actual results
The processing happens in each run.
## Environment info
<!-- You can run the command `datasets-cli env` and copy-and-paste its output below. -->
- `datasets` version: 1.8.0
- Platform: linux
- Python version: 3.7.6
- PyArrow version: 3.0.0
This is the first time I report a bug. If there is any problem or confusing description, please let me know 😄.
|
closed
|
https://github.com/huggingface/datasets/issues/2825
| 2021-08-23T03:23:37
| 2024-07-29T11:25:50
| 2021-08-31T13:13:36
|
{
"login": "hobbitlzy",
"id": 35392624,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
976,394,721
| 2,824
|
Fix defaults in cache_dir docstring in load.py
|
Fix defaults in the `cache_dir` docstring.
|
closed
|
https://github.com/huggingface/datasets/pull/2824
| 2021-08-22T14:48:37
| 2021-08-26T13:23:32
| 2021-08-26T11:55:16
|
{
"login": "mariosasko",
"id": 47462742,
"type": "User"
}
|
[] | true
|
[] |
976,135,355
| 2,823
|
HF_DATASETS_CACHE variable in Windows
|
I can't seem to use a custom Cache directory in Windows. I have tried:
set HF_DATASETS_CACHE = "C:\Datasets"
set HF_DATASETS_CACHE = "C:/Datasets"
set HF_DATASETS_CACHE = "C:\\Datasets"
set HF_DATASETS_CACHE = "r'C:\Datasets'"
set HF_DATASETS_CACHE = "\Datasets"
set HF_DATASETS_CACHE = "/Datasets"
In each instance I get the "[WinError 123] The filename, directory name, or volume label syntax is incorrect" error when attempting to load a dataset
|
closed
|
https://github.com/huggingface/datasets/issues/2823
| 2021-08-21T13:17:44
| 2021-08-21T13:20:11
| 2021-08-21T13:20:11
|
{
"login": "rp2839",
"id": 8453798,
"type": "User"
}
|
[] | false
|
[] |
975,744,463
| 2,822
|
Add url prefix convention for many compression formats
|
## Intro
When doing dataset streaming, the uncompression of compressed files is done on the fly using `fsspec`.
In particular, the download manager method `download_and_extract` doesn't return a path to the local download and extracted file, but instead a chained URL so that the uncompression can be done when the file is opened. A few examples of chained URLS:
- `gz://file.txt::https://foo.bar/file.txt.gz`
- `bz2://file.txt::https://foo.bar/file.txt.bz2`
- `zip://::https://foo.bar/archive.zip`
- `tar://::https://foo.bar/archive.tar.gz` (the TAR uncompression includes gz, bz2 etc. uncompression in `fsspec`)
This syntax is highly inspired by the `fsspec` URL chaining syntax from https://filesystem-spec.readthedocs.io/en/latest/features.html#url-chaining
This url prefixing allows `open` to know what kind of uncompression to do in a dataset script when doing
```python
def _generate_examples(self, urlpath):
with open(urlpath) as f:
....
```
## What it changes
This changes the previous behavior from https://github.com/huggingface/datasets/pull/2786 , in which `open` was trying to infer the compression automatically. Infering the compression made it impossible to know whether the user wanted `open` to return compressed data (as the default behavior of the buitin open), or the uncompressed data. By adding uncompression prefixes to the URL, `open` know directly if it has to uncompress or not, and also which protocol to use.
## Additional notes
This PR should close https://github.com/huggingface/datasets/issues/2813
It should also close this PR https://github.com/huggingface/datasets/pull/2811 since the oscar dataset script won't try to uncompress twice anymore
Note that I had to temporarily remove the support for passing tar and zip files to `data_files` for streaming to make it work, since it makes it ambiguous whether a zip file passed as `data_files` should be uncompressed or not. IMO we can make it work again by changing the syntax to make the glob explicit:
```python
load_dataset("json", data_files="zip://*.jsonl::https://foo.bar/archive.zip")
```
This is the exact same convention as fsspec and it removes all ambiguities
cc @albertvillanova @lewtun
|
closed
|
https://github.com/huggingface/datasets/pull/2822
| 2021-08-20T16:11:23
| 2021-08-23T15:59:16
| 2021-08-23T15:59:14
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
975,556,032
| 2,821
|
Cannot load linnaeus dataset
|
## Describe the bug
The [linnaeus](https://huggingface.co/datasets/linnaeus) dataset cannot be loaded. To reproduce:
```
from datasets import load_dataset
datasets = load_dataset("linnaeus")
```
This results in:
```
Downloading and preparing dataset linnaeus/linnaeus (download: 17.36 MiB, generated: 8.74 MiB, post-processed: Unknown size, total: 26.10 MiB) to /root/.cache/huggingface/datasets/linnaeus/linnaeus/1.0.0/2ff05dbc256108233262f596e09e322dbc3db067202de14286913607cd9cb704...
---------------------------------------------------------------------------
ConnectionError Traceback (most recent call last)
<ipython-input-4-7ef3a88f6276> in <module>()
1 from datasets import load_dataset
2
----> 3 datasets = load_dataset("linnaeus")
11 frames
/usr/local/lib/python3.7/dist-packages/datasets/utils/file_utils.py in get_from_cache(url, cache_dir, force_download, proxies, etag_timeout, resume_download, user_agent, local_files_only, use_etag, max_retries, use_auth_token)
603 raise FileNotFoundError("Couldn't find file at {}".format(url))
604 _raise_if_offline_mode_is_enabled(f"Tried to reach {url}")
--> 605 raise ConnectionError("Couldn't reach {}".format(url))
606
607 # Try a second time
ConnectionError: Couldn't reach https://drive.google.com/u/0/uc?id=1OletxmPYNkz2ltOr9pyT0b0iBtUWxslh&export=download/
```
|
closed
|
https://github.com/huggingface/datasets/issues/2821
| 2021-08-20T12:15:15
| 2021-08-31T13:13:02
| 2021-08-31T13:12:09
|
{
"login": "NielsRogge",
"id": 48327001,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
975,210,712
| 2,820
|
Downloading “reddit” dataset keeps timing out.
|
## Describe the bug
A clear and concise description of what the bug is.
Everytime I try and download the reddit dataset it times out before finishing and I have to try again.
There is some timeout error that I will post once it happens again.
## Steps to reproduce the bug
```python
from datasets import load_dataset
dataset = load_dataset("reddit", ignore_verifications=True, cache_dir="/Volumes/My Passport for Mac/og-chat-data")
```
## Expected results
A clear and concise description of the expected results.
I would expect the download to finish, or at least provide a parameter to extend the read timeout window.
## Actual results
Specify the actual results or traceback.
Shown below in error message.
## Environment info
<!-- You can run the command `datasets-cli env` and copy-and-paste its output below. -->
- `datasets` version: 1.11.0
- Platform: macOS
- Python version: 3.9.6 (conda env)
- PyArrow version: N/A
|
closed
|
https://github.com/huggingface/datasets/issues/2820
| 2021-08-20T02:52:36
| 2021-09-08T14:52:02
| 2021-09-08T14:52:02
|
{
"login": "smeyerhot",
"id": 43877130,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
974,683,155
| 2,819
|
Added XL-Sum dataset
|
Added XL-Sum dataset published in ACL-IJCNLP 2021. (https://aclanthology.org/2021.findings-acl.413/). The default timeout values in `src/datasets/utils/file_utls.py` were increased to enable downloading from the original google drive links.
|
closed
|
https://github.com/huggingface/datasets/pull/2819
| 2021-08-19T13:47:45
| 2021-09-29T08:13:44
| 2021-09-23T17:49:05
|
{
"login": "abhik1505040",
"id": 49608995,
"type": "User"
}
|
[] | true
|
[] |
974,552,009
| 2,818
|
cannot load data from my loacal path
|
## Describe the bug
I just want to directly load data from my local path,but find a bug.And I compare it with pandas to provide my local path is real.
here is my code
```python3
# print my local path
print(config.train_path)
# read data and print data length
tarin=pd.read_csv(config.train_path)
print(len(tarin))
# loading data by load_dataset
data = load_dataset('csv',data_files=config.train_path)
print(len(data))
```
## Steps to reproduce the bug
```python
C:\Users\wie\Documents\项目\文本分类\data\train.csv
7613
Traceback (most recent call last):
File "c:/Users/wie/Documents/项目/文本分类/lib/DataPrecess.py", line 17, in <module>
data = load_dataset('csv',data_files=config.train_path)
File "C:\Users\wie\Miniconda3\lib\site-packages\datasets\load.py", line 830, in load_dataset
**config_kwargs,
File "C:\Users\wie\Miniconda3\lib\site-packages\datasets\load.py", line 710, in load_dataset_builder
**config_kwargs,
File "C:\Users\wie\Miniconda3\lib\site-packages\datasets\builder.py", line 271, in __init__
**config_kwargs,
File "C:\Users\wie\Miniconda3\lib\site-packages\datasets\builder.py", line 386, in _create_builder_config
config_kwargs, custom_features=custom_features, use_auth_token=self.use_auth_token
File "C:\Users\wie\Miniconda3\lib\site-packages\datasets\builder.py", line 156, in create_config_id
raise ValueError("Please provide a valid `data_files` in `DatasetBuilder`")
ValueError: Please provide a valid `data_files` in `DatasetBuilder`
```
## Expected results
A clear and concise description of the expected results.
## Actual results
Specify the actual results or traceback.
## Environment info
<!-- You can run the command `datasets-cli env` and copy-and-paste its output below. -->
- `datasets` version: 1.11.0
- Platform: win10
- Python version: 3.7.9
- PyArrow version: 5.0.0
|
closed
|
https://github.com/huggingface/datasets/issues/2818
| 2021-08-19T11:13:30
| 2023-07-25T17:42:15
| 2023-07-25T17:42:15
|
{
"login": "yang-collect",
"id": 46920280,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
974,486,051
| 2,817
|
Rename The Pile subsets
|
After discussing with @yjernite we think it's better to have the subsets of The Pile explicitly have "the_pile" in their names.
I'm doing the changes for the subsets that @richarddwang added:
- [x] books3 -> the_pile_books3 https://github.com/huggingface/datasets/pull/2801
- [x] stack_exchange -> the_pile_stack_exchange https://github.com/huggingface/datasets/pull/2803
- [x] openwebtext2 -> the_pile_openwebtext2 https://github.com/huggingface/datasets/pull/2802
For consistency we should also rename `bookcorpusopen` to `the_pile_bookcorpus` IMO, but let me know what you think.
(we can just add a deprecation message to `bookcorpusopen` for now and add `the_pile_bookcorpus`)
|
closed
|
https://github.com/huggingface/datasets/pull/2817
| 2021-08-19T09:56:22
| 2021-08-23T16:24:10
| 2021-08-23T16:24:09
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
974,031,404
| 2,816
|
Add Mostly Basic Python Problems Dataset
|
## Adding a Dataset
- **Name:** Mostly Basic Python Problems Dataset
- **Description:** The benchmark consists of around 1,000 crowd-sourced Python programming problems, designed to be solvable by entry level programmers, covering programming fundamentals, standard library functionality, and so on. Each problem consists of a task description, code solution and 3 automated test cases.
- **Paper:** *link to the dataset paper if available*
- **Data:** https://github.com/google-research/google-research/tree/master/mbpp
- **Motivation:** Simple, small dataset related to coding problems.
Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md).
|
open
|
https://github.com/huggingface/datasets/issues/2816
| 2021-08-18T20:28:39
| 2021-09-10T08:04:20
| null |
{
"login": "osanseviero",
"id": 7246357,
"type": "User"
}
|
[
{
"name": "dataset request",
"color": "e99695"
}
] | false
|
[] |
973,862,024
| 2,815
|
Tiny typo fixes of "fo" -> "of"
|
Noticed a few of these when reading docs- feel free to ignore the PR and just fix on some main contributor branch if more helpful. Thanks for the great library! :)
|
closed
|
https://github.com/huggingface/datasets/pull/2815
| 2021-08-18T16:36:11
| 2021-08-19T08:03:02
| 2021-08-19T08:03:02
|
{
"login": "aronszanto",
"id": 9934829,
"type": "User"
}
|
[] | true
|
[] |
973,632,645
| 2,814
|
Bump tqdm version
|
The recently released tqdm 4.62.1 includes a fix for PermissionError on Windows (submitted by me in https://github.com/tqdm/tqdm/pull/1207), which means we can remove expensive `gc.collect` calls by bumping tqdm to that version. This PR does exactly that and, additionally, fixes a `disable_tqdm` definition that would previously, if used, raise a PermissionError on Windows.
|
closed
|
https://github.com/huggingface/datasets/pull/2814
| 2021-08-18T12:51:29
| 2021-08-18T13:44:11
| 2021-08-18T13:39:50
|
{
"login": "mariosasko",
"id": 47462742,
"type": "User"
}
|
[] | true
|
[] |
973,470,580
| 2,813
|
Remove compression from xopen
|
We implemented support for streaming with 2 requirements:
- transparent use for the end user: just needs to pass the parameter `streaming=True`
- no additional work for the contributors: previous loading scripts should also work in streaming mode with no (or minor) changes; and new loading scripts should not involve additional code to support streaming
In order to fulfill these requirements, streaming implementation patched some Python functions:
- the `open(urlpath)` function was patched with `fsspec.open(urlpath)`
- the `os.path.join(urlpath, *others)` function was patched in order to add to `urlpath` hops (`::`) and extractor protocols (`zip://`), which are required by `fsspec.open`
Recently, we implemented support for streaming all archive+compression formats: zip, tar, gz, bz2, lz4, xz, zst; tar.gz, tar.bz2,...
Under the hood, the implementation:
- passes an additional parameter `compression` to `fsspec.open`, so that it performs the decompression on the fly: `fsspec.open(urlpath, compression=...)`
Some concerns have been raised about passing the parameter `compression` to `fsspec.open`:
- https://github.com/huggingface/datasets/pull/2786#discussion_r689550254
- #2811
The main argument is that if `open` decompresses the file and afterwards we call `gzip.open` on it, that will raise an error in `oscar` dataset:
```python
gzip.open(open(urlpath
```
While this is true:
- it is not natural/usual to call `open` inside `gzip.open` (never seen this before)
- indeed, this was recently (2 months ago) coded that way in `datasets` in order to allow streaming support (with previous implementation of streaming)
In this particular case, there is a natural fix solution: #2811:
- Revert the `open` inside the `gzip.open` (change done 2 months ago): `gzip.open(open(urlpath` => `gzip.open(urlpath`
- Patch `gzip.open(urlpath` with `fsspec.open(urlpath, compression="gzip"`
Are there other issues apart from this?
Note that there is an issue just because the open inside of the gzip.open. There is no issue in the other cases where datasets loading scripts use just
- `gzip.open`
- `open` (after having called dl_manager.download_and_extract)
TODO:
- [ ] Is this really an issue? Please enumerate the `datasets` loading scripts where this is problematic.
- For the moment, there are only 3 datasets where we have an `open` inside a `gzip.open`:
- oscar (since 23 June), mc4 (since 2 July) and c4 (since 2 July)
- In the 3 datasets, the only reason to put an open inside a gzip.open was indeed to force supporting streaming
- [ ] If this is indeed an issue, which are the possible alternatives? Pros/cons?
|
closed
|
https://github.com/huggingface/datasets/issues/2813
| 2021-08-18T09:35:59
| 2021-08-23T15:59:14
| 2021-08-23T15:59:14
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[
{
"name": "generic discussion",
"color": "c5def5"
}
] | false
|
[] |
972,936,889
| 2,812
|
arXiv Dataset verification problem
|
## Describe the bug
`dataset_infos.json` for `arxiv_dataset` contains a fixed number of training examples, however the data (downloaded from an external source) is updated every week with additional examples.
Therefore, loading the dataset without `ignore_verifications=True` results in a verification error.
|
open
|
https://github.com/huggingface/datasets/issues/2812
| 2021-08-17T18:01:48
| 2022-01-19T14:15:35
| null |
{
"login": "eladsegal",
"id": 13485709,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
},
{
"name": "dataset bug",
"color": "2edb81"
}
] | false
|
[] |
972,522,480
| 2,811
|
Fix stream oscar
|
Previously, an additional `open` was added to oscar to make it stream-compatible: 587bbb94e891b22863b312b99696e32708c379f4.
This was argued that might be problematic: https://github.com/huggingface/datasets/pull/2786#discussion_r690045921
This PR:
- removes that additional `open`
- patches `gzip.open` with `xopen` + `compression="gzip"`
|
closed
|
https://github.com/huggingface/datasets/pull/2811
| 2021-08-17T10:10:59
| 2021-08-26T10:26:15
| 2021-08-26T10:26:14
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
972,040,022
| 2,810
|
Add WIT Dataset
|
Adds Google's [WIT](https://github.com/google-research-datasets/wit) dataset.
|
closed
|
https://github.com/huggingface/datasets/pull/2810
| 2021-08-16T19:34:09
| 2022-05-06T12:27:29
| 2022-05-06T12:26:16
|
{
"login": "hassiahk",
"id": 13920778,
"type": "User"
}
|
[] | true
|
[] |
971,902,613
| 2,809
|
Add Beans Dataset
|
Adds the [beans](https://github.com/AI-Lab-Makerere/ibean/) image classification dataset.
|
closed
|
https://github.com/huggingface/datasets/pull/2809
| 2021-08-16T16:22:33
| 2021-08-26T11:42:27
| 2021-08-26T11:42:27
|
{
"login": "nateraw",
"id": 32437151,
"type": "User"
}
|
[] | true
|
[] |
971,882,320
| 2,808
|
Enable streaming for Wikipedia corpora
|
**Is your feature request related to a problem? Please describe.**
Several of the [Wikipedia corpora](https://huggingface.co/datasets?search=wiki) on the Hub involve quite large files that would be a good candidate for streaming. Currently it is not possible to stream these corpora:
```python
from datasets import load_dataset
# Throws ValueError: Builder wikipedia is not streamable.
wiki_dataset_streamed = load_dataset("wikipedia", "20200501.en", split="train", streaming=True)
```
Given that these corpora are derived from Wikipedia dumps in XML format which are then processed with Apache Beam, I am not sure whether streaming is possible in principle. The goal of this issue is to discuss whether this feature even makes sense :)
**Describe the solution you'd like**
It would be nice to be able to stream Wikipedia corpora from the Hub with something like
```python
from datasets import load_dataset
wiki_dataset_streamed = load_dataset("wikipedia", "20200501.en", split="train", streaming=True)
```
|
closed
|
https://github.com/huggingface/datasets/issues/2808
| 2021-08-16T15:59:12
| 2023-07-20T13:45:30
| 2023-07-20T13:45:30
|
{
"login": "lewtun",
"id": 26859204,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
971,849,863
| 2,807
|
Add cats_vs_dogs dataset
|
Adds Microsoft's [Cats vs. Dogs](https://www.microsoft.com/en-us/download/details.aspx?id=54765) dataset.
|
closed
|
https://github.com/huggingface/datasets/pull/2807
| 2021-08-16T15:21:11
| 2021-08-30T16:35:25
| 2021-08-30T16:35:24
|
{
"login": "nateraw",
"id": 32437151,
"type": "User"
}
|
[] | true
|
[] |
971,625,449
| 2,806
|
Fix streaming tar files from canonical datasets
|
Previous PR #2800 implemented support to stream remote tar files when passing the parameter `data_files`: they required a glob string `"*"`.
However, this glob string creates an error when streaming canonical datasets (with a `join` after the `open`).
This PR fixes this issue and allows streaming tar files both from:
- canonical datasets scripts and
- data files.
This PR also adds support for compressed tar files: `.tar.gz`, `.tar.bz2`,...
|
closed
|
https://github.com/huggingface/datasets/pull/2806
| 2021-08-16T11:10:28
| 2021-10-13T09:04:03
| 2021-10-13T09:04:02
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
971,436,456
| 2,805
|
Fix streaming zip files from canonical datasets
|
Previous PR #2798 fixed streaming remote zip files when passing the parameter `data_files`.
However, that broke streaming zip files used in canonical `datasets` scripts, which normally have a subsequent `join()` (patched with `xjoin()`) after the `StreamingDownloadManager.download_and_extract()` is called.
This PR fixes this issue and allows streaming zip files both from:
- canonical datasets scripts and
- data files.
|
closed
|
https://github.com/huggingface/datasets/pull/2805
| 2021-08-16T07:11:40
| 2021-08-16T10:34:00
| 2021-08-16T10:34:00
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
971,353,437
| 2,804
|
Add Food-101
|
Adds image classification dataset [Food-101](https://data.vision.ee.ethz.ch/cvl/datasets_extra/food-101/).
|
closed
|
https://github.com/huggingface/datasets/pull/2804
| 2021-08-16T04:26:15
| 2021-08-20T14:31:33
| 2021-08-19T12:48:06
|
{
"login": "nateraw",
"id": 32437151,
"type": "User"
}
|
[] | true
|
[] |
970,858,928
| 2,803
|
add stack exchange
|
stack exchange is part of EleutherAI/The Pile, but AFAIK, The Pile dataset blend all sub datasets together thus we are not able to use just one of its sub dataset from The Pile data. So I create an independent dataset using The Pile preliminary components.
I also change default `timeout` to 100 seconds instead of 10 seconds, otherwise I keep getting read time out when downloading source data of stack exchange and cc100 dataset.
When I was creating dataset card. I found there is room for creating / editing dataset card. I've made it an issue. #2797
Also I am wondering whether the import of The Pile dataset is actively undertaken (because I may need it recently)? #1675
|
closed
|
https://github.com/huggingface/datasets/pull/2803
| 2021-08-14T08:11:02
| 2021-08-19T10:07:33
| 2021-08-19T08:07:38
|
{
"login": "richarddwang",
"id": 17963619,
"type": "User"
}
|
[] | true
|
[] |
970,848,302
| 2,802
|
add openwebtext2
|
openwebtext2 is part of EleutherAI/The Pile, but AFAIK, The Pile dataset blend all sub datasets together thus we are not able to use just one of its sub dataset from The Pile data. So I create an independent dataset using The Pile preliminary components.
When I was creating dataset card. I found there is room for creating / editing dataset card. I've made it an issue. #2797
Also I am wondering whether the import of The Pile dataset is actively undertaken (because I may need it recently)? #1675
|
closed
|
https://github.com/huggingface/datasets/pull/2802
| 2021-08-14T07:09:03
| 2021-08-23T14:06:14
| 2021-08-23T14:06:14
|
{
"login": "richarddwang",
"id": 17963619,
"type": "User"
}
|
[] | true
|
[] |
970,844,617
| 2,801
|
add books3
|
books3 is part of EleutherAI/The Pile, but AFAIK, The Pile dataset blend all sub datasets together thus we are not able to use just one of its sub dataset from The Pile data. So I create an independent dataset using The Pile preliminary components.
When I was creating dataset card. I found there is room for creating / editing dataset card. I've made it an issue. #2797
Also I am wondering whether the import of The Pile dataset is actively undertaken (because I may need it recently)? #1675
|
closed
|
https://github.com/huggingface/datasets/pull/2801
| 2021-08-14T07:04:25
| 2021-08-19T16:43:09
| 2021-08-18T15:36:59
|
{
"login": "richarddwang",
"id": 17963619,
"type": "User"
}
|
[] | true
|
[] |
970,819,988
| 2,800
|
Support streaming tar files
|
This PR adds support to stream tar files by using the `fsspec` tar protocol.
It also uses the custom `readline` implemented in PR #2786.
The corresponding test is implemented in PR #2786.
|
closed
|
https://github.com/huggingface/datasets/pull/2800
| 2021-08-14T04:40:17
| 2021-08-26T10:02:30
| 2021-08-14T04:55:57
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
970,507,351
| 2,799
|
Loading JSON throws ArrowNotImplementedError
|
## Describe the bug
I have created a [dataset](https://huggingface.co/datasets/lewtun/github-issues-test) of GitHub issues in line-separated JSON format and am finding that I cannot load it with the `json` loading script (see stack trace below).
Curiously, there is no problem loading the dataset with `pandas` which suggests some incorrect type inference is being made on the `datasets` side. For example, the stack trace indicates that some URL fields are being parsed as timestamps.
You can find a Colab notebook which reproduces the error [here](https://colab.research.google.com/drive/1YUCM0j1vx5ZrouQbYSzal6RwB4-Aoh4o?usp=sharing).
**Edit:** If one repeatedly tries to load the dataset, it _eventually_ works but I think it would still be good to understand why it fails in the first place :)
## Steps to reproduce the bug
```python
from datasets import load_dataset
from huggingface_hub import hf_hub_url
import pandas as pd
# returns https://huggingface.co/datasets/lewtun/github-issues-test/resolve/main/issues-datasets.jsonl
data_files = hf_hub_url(repo_id="lewtun/github-issues-test", filename="issues-datasets.jsonl", repo_type="dataset")
# throws ArrowNotImplementedError
dset = load_dataset("json", data_files=data_files, split="test")
# no problem with pandas ...
df = pd.read_json(data_files, orient="records", lines=True)
df.head()
```
## Expected results
I can load any line-separated JSON file, similar to `pandas`.
## Actual results
```
---------------------------------------------------------------------------
ArrowNotImplementedError Traceback (most recent call last)
<ipython-input-7-5b8e82b6c3a2> in <module>()
----> 1 dset = load_dataset("json", data_files=data_files, split="test")
9 frames
/usr/local/lib/python3.7/dist-packages/pyarrow/error.pxi in pyarrow.lib.check_status()
ArrowNotImplementedError: JSON conversion to struct<url: timestamp[s], html_url: timestamp[s], labels_url: timestamp[s], id: int64, node_id: timestamp[s], number: int64, title: timestamp[s], description: timestamp[s], creator: struct<login: timestamp[s], id: int64, node_id: timestamp[s], avatar_url: timestamp[s], gravatar_id: timestamp[s], url: timestamp[s], html_url: timestamp[s], followers_url: timestamp[s], following_url: timestamp[s], gists_url: timestamp[s], starred_url: timestamp[s], subscriptions_url: timestamp[s], organizations_url: timestamp[s], repos_url: timestamp[s], events_url: timestamp[s], received_events_url: timestamp[s], type: timestamp[s], site_admin: bool>, open_issues: int64, closed_issues: int64, state: timestamp[s], created_at: timestamp[s], updated_at: timestamp[s], due_on: timestamp[s], closed_at: timestamp[s]> is not supported
```
## Environment info
<!-- You can run the command `datasets-cli env` and copy-and-paste its output below. -->
- `datasets` version: 1.11.0
- Platform: Linux-5.4.104+-x86_64-with-Ubuntu-18.04-bionic
- Python version: 3.7.11
- PyArrow version: 3.0.0
|
closed
|
https://github.com/huggingface/datasets/issues/2799
| 2021-08-13T15:31:48
| 2022-01-10T18:59:32
| 2022-01-10T18:59:32
|
{
"login": "lewtun",
"id": 26859204,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
970,493,126
| 2,798
|
Fix streaming zip files
|
Currently, streaming remote zip data files gives `FileNotFoundError` message:
```python
data_files = f"https://huggingface.co/datasets/albertvillanova/datasets-tests-compression/resolve/main/sample.zip"
ds = load_dataset("json", split="train", data_files=data_files, streaming=True)
next(iter(ds))
```
This PR fixes it by adding a glob string.
The corresponding test is implemented in PR #2786.
|
closed
|
https://github.com/huggingface/datasets/pull/2798
| 2021-08-13T15:17:01
| 2021-08-16T14:16:50
| 2021-08-13T15:38:28
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
970,331,634
| 2,797
|
Make creating/editing dataset cards easier, by editing on site and dumping info from test command.
|
**Is your feature request related to a problem? Please describe.**
Creating and editing dataset cards should be but not that easy
- If other else know Some information I don't know (bias of dataset, dataset curation, supported dataset, ...), he/she should know the description on hf.co comes from README.md under github huggingface/datasets/datasets/the dataset, and willing to make a pr to add or fix information.
- Many information is also saved in `dataset_info.json` (citaion, description), but still need to write it down to README.md again.
- Contributor need to pip install and start a local server just for tagging the dataset's size. And contributor may be creating the dataset on lab's server, which can't open a browser.
- if any one proposes a new tag, it doesn't show in the list that another creator see. (a stackoverflow way may be ideal)
- dataset card generator web app doesn't generate the necessary subsecion `Contributions` for us.
**Describe the solution you'd like**
- Everyone (or at least the author/contributor) can edit the description, information, tags of the dataset, on hf.co website. Just like wikipedia+stackoverflow
- We can infer the actual data size, citation, data instance, ... from `dataset_info.json` and `dataset.arrow` via `dataset-cli test`
|
open
|
https://github.com/huggingface/datasets/issues/2797
| 2021-08-13T11:54:49
| 2021-08-14T08:42:09
| null |
{
"login": "richarddwang",
"id": 17963619,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
970,235,846
| 2,796
|
add cedr dataset
| null |
closed
|
https://github.com/huggingface/datasets/pull/2796
| 2021-08-13T09:37:35
| 2021-08-27T16:01:36
| 2021-08-27T16:01:36
|
{
"login": "naumov-al",
"id": 22640075,
"type": "User"
}
|
[] | true
|
[] |
969,728,545
| 2,794
|
Warnings and documentation about pickling incorrect
|
## Describe the bug
I have a docs bug and a closely related docs enhancement suggestion!
### Bug
The warning and documentation say "either `dill` or `pickle`" for fingerprinting. But it seems that `dill`, which is installed by `datasets` by default, _must_ work, or else the fingerprinting fails.
Warning:
https://github.com/huggingface/datasets/blob/450b9174765374111e5c6daab0ed294bc3d9b639/src/datasets/fingerprint.py#L262
Docs:
> For a transform to be hashable, it needs to be pickleable using dill or pickle.
> – [docs](https://huggingface.co/docs/datasets/processing.html#fingerprinting)
For my code, `pickle` works, but `dill` fails. The `dill` failure has already been reported in https://github.com/huggingface/datasets/issues/2643. However, the `dill` failure causes a hashing failure in the datasets library, without any backing off to `pickle`. This implies that it's not the case that either `dill` **or** `pickle` can work, but that `dill` must work if it is installed. I think this is more accurate wording, since it is installed and used by default:
https://github.com/huggingface/datasets/blob/c93525dc291346e54212567fa72d7d607befe937/setup.py#L83
... and the hashing will fail if it fails.
### Enhancement
I think it'd be very helpful to add to the documentation how to debug hashing failures. It took me a while to figure out how to diagnose this. There is a very nice two-liner by @lhoestq in https://github.com/huggingface/datasets/issues/2516#issuecomment-865173139:
```python
from datasets.fingerprint import Hasher
Hasher.hash(my_object)
```
I think add this to the docs will help future users quickly debug any hashing troubles of their own :-)
## Steps to reproduce the bug
`dill` but not `pickle` hashing failure in https://github.com/huggingface/datasets/issues/2643
## Expected results
If either `dill` or `pickle` can successfully hash, the hashing will succeed.
## Actual results
If `dill` or `pickle` cannot hash, the hashing fails.
## Environment info
- `datasets` version: 1.9.0
- Platform: Linux-5.8.0-1038-gcp-x86_64-with-glibc2.31
- Python version: 3.9.6
- PyArrow version: 4.0.1
|
open
|
https://github.com/huggingface/datasets/issues/2794
| 2021-08-12T23:09:13
| 2021-08-12T23:09:31
| null |
{
"login": "mbforbes",
"id": 1170062,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
968,967,773
| 2,793
|
Fix type hint for data_files
|
Fix type hint for `data_files` in signatures and docstrings.
|
closed
|
https://github.com/huggingface/datasets/pull/2793
| 2021-08-12T14:42:37
| 2021-08-12T15:35:29
| 2021-08-12T15:35:29
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
968,650,274
| 2,792
|
Update: GooAQ - add train/val/test splits
|
[GooAQ](https://github.com/allenai/gooaq) dataset was recently updated after splits were added for the same. This PR contains new updated GooAQ with train/val/test splits and updated README as well.
|
closed
|
https://github.com/huggingface/datasets/pull/2792
| 2021-08-12T11:40:18
| 2021-08-27T15:58:45
| 2021-08-27T15:58:14
|
{
"login": "bhavitvyamalik",
"id": 19718818,
"type": "User"
}
|
[] | true
|
[] |
968,360,314
| 2,791
|
Fix typo in cnn_dailymail
| null |
closed
|
https://github.com/huggingface/datasets/pull/2791
| 2021-08-12T08:38:42
| 2021-08-12T11:17:59
| 2021-08-12T11:17:59
|
{
"login": "omaralsayed",
"id": 42531544,
"type": "User"
}
|
[] | true
|
[] |
967,772,181
| 2,790
|
Fix typo in test_dataset_common
| null |
closed
|
https://github.com/huggingface/datasets/pull/2790
| 2021-08-12T01:10:29
| 2021-08-12T11:31:29
| 2021-08-12T11:31:29
|
{
"login": "nateraw",
"id": 32437151,
"type": "User"
}
|
[] | true
|
[] |
967,361,934
| 2,789
|
Updated dataset description of DaNE
| null |
closed
|
https://github.com/huggingface/datasets/pull/2789
| 2021-08-11T19:58:48
| 2021-08-12T16:10:59
| 2021-08-12T16:06:01
|
{
"login": "KennethEnevoldsen",
"id": 23721977,
"type": "User"
}
|
[] | true
|
[] |
967,149,389
| 2,788
|
How to sample every file in a list of files making up a split in a dataset when loading?
|
I am loading a dataset with multiple train, test, and validation files like this:
```
data_files_dict = {
"train": [train_file1, train_file2],
"test": [test_file1, test_file2],
"val": [val_file1, val_file2]
}
dataset = datasets.load_dataset(
"csv",
data_files=data_files_dict,
split=['train[:8]', 'test[:8]', 'val[:8]']
)
```
However, this only selects the first 8 rows from train_file1, test_file1, val_file1, since they are the first files in the lists.
I'm trying to formulate a split argument that can sample from each file specified in my list of files that make up each split.
Is this type of splitting supported? If so, how can I do it?
|
closed
|
https://github.com/huggingface/datasets/issues/2788
| 2021-08-11T17:43:21
| 2023-07-25T17:40:50
| 2023-07-25T17:40:50
|
{
"login": "brijow",
"id": 11220949,
"type": "User"
}
|
[] | false
|
[] |
967,018,406
| 2,787
|
ConnectionError: Couldn't reach https://raw.githubusercontent.com
|
Hello,
I am trying to run run_glue.py and it gives me this error -
Traceback (most recent call last):
File "E:/BERT/pytorch_hugging/transformers/examples/pytorch/text-classification/run_glue.py", line 546, in <module>
main()
File "E:/BERT/pytorch_hugging/transformers/examples/pytorch/text-classification/run_glue.py", line 250, in main
datasets = load_dataset("glue", data_args.task_name, cache_dir=model_args.cache_dir)
File "C:\install\Anaconda3\envs\huggingface\lib\site-packages\datasets\load.py", line 718, in load_dataset
use_auth_token=use_auth_token,
File "C:\install\Anaconda3\envs\huggingface\lib\site-packages\datasets\load.py", line 320, in prepare_module
local_path = cached_path(file_path, download_config=download_config)
File "C:\install\Anaconda3\envs\huggingface\lib\site-packages\datasets\utils\file_utils.py", line 291, in cached_path
use_auth_token=download_config.use_auth_token,
File "C:\install\Anaconda3\envs\huggingface\lib\site-packages\datasets\utils\file_utils.py", line 623, in get_from_cache
raise ConnectionError("Couldn't reach {}".format(url))
ConnectionError: Couldn't reach https://raw.githubusercontent.com/huggingface/datasets/1.7.0/datasets/glue/glue.py
Trying to do python run_glue.py --model_name_or_path
bert-base-cased
--task_name
mrpc
--do_train
--do_eval
--max_seq_length
128
--per_device_train_batch_size
32
--learning_rate
2e-5
--num_train_epochs
3
--output_dir
./tmp/mrpc/
Is this something on my end? From what I can tell, this was re-fixeded by @fullyz a few months ago.
Thank you!
|
closed
|
https://github.com/huggingface/datasets/issues/2787
| 2021-08-11T16:19:01
| 2023-10-03T12:39:25
| 2021-08-18T15:09:18
|
{
"login": "jinec",
"id": 39627475,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
966,282,934
| 2,786
|
Support streaming compressed files
|
Add support to stream compressed files (current options in fsspec):
- bz2
- lz4
- xz
- zstd
cc: @lewtun
|
closed
|
https://github.com/huggingface/datasets/pull/2786
| 2021-08-11T09:02:06
| 2021-08-17T05:28:39
| 2021-08-16T06:36:19
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
965,461,382
| 2,783
|
Add KS task to SUPERB
|
Add the KS (keyword spotting) task as described in the [SUPERB paper](https://arxiv.org/abs/2105.01051).
- [s3prl instructions](https://github.com/s3prl/s3prl/blob/master/s3prl/downstream/README.md#ks-keyword-spotting)
- [s3prl implementation](https://github.com/s3prl/s3prl/blob/master/s3prl/downstream/speech_commands/dataset.py)
- [TFDS implementation](https://github.com/tensorflow/datasets/blob/master/tensorflow_datasets/audio/speech_commands.py)
Some notable quirks:
- The dataset is originally single-archive (train+val+test all in one), but the test set has a "canonical" distribution in a separate archive, which is also used here (see `_split_ks_files()`).
- The `_background_noise_`/`_silence_` audio files are much longer than others, so they require some sort of slicing for downstream training. I decided to leave the implementation of that up to the users, since TFDS and s3prl take different approaches (either slicing wavs deterministically, or subsampling randomly at runtime)
Related to #2619.
|
closed
|
https://github.com/huggingface/datasets/pull/2783
| 2021-08-10T22:14:07
| 2021-08-12T16:45:01
| 2021-08-11T20:19:17
|
{
"login": "anton-l",
"id": 26864830,
"type": "User"
}
|
[] | true
|
[] |
964,858,439
| 2,782
|
Fix renaming of corpus_bleu args
|
Last `sacrebleu` release (v2.0.0) has renamed `sacrebleu.corpus_bleu` args from `(sys_stream, ref_streams)` to `(hipotheses, references)`: https://github.com/mjpost/sacrebleu/pull/152/files#diff-2553a315bb1f7e68c9c1b00d56eaeb74f5205aeb3a189bc3e527b122c6078795L17-R15
This PR passes the args without parameter names, so that it is valid for all versions of `sacrebleu`.
This is a partial hotfix of #2781.
Close #2781.
|
closed
|
https://github.com/huggingface/datasets/pull/2782
| 2021-08-10T11:02:34
| 2021-08-10T11:16:07
| 2021-08-10T11:16:07
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
964,805,351
| 2,781
|
Latest v2.0.0 release of sacrebleu has broken some metrics
|
## Describe the bug
After `sacrebleu` v2.0.0 release (see changes here: https://github.com/mjpost/sacrebleu/pull/152/files#diff-2553a315bb1f7e68c9c1b00d56eaeb74f5205aeb3a189bc3e527b122c6078795L17-R15), some of `datasets` metrics are broken:
- Default tokenizer `sacrebleu.DEFAULT_TOKENIZER` no longer exists:
- #2739
- #2778
- Bleu tokenizers are no longer accessible with `sacrebleu.TOKENIZERS`:
- #2779
- `corpus_bleu` args have been renamed from `(sys_stream, ref_streams)` to `(hipotheses, references)`:
- #2782
|
closed
|
https://github.com/huggingface/datasets/issues/2781
| 2021-08-10T09:59:41
| 2021-08-10T11:16:07
| 2021-08-10T11:16:07
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
964,794,764
| 2,780
|
VIVOS dataset for Vietnamese ASR
| null |
closed
|
https://github.com/huggingface/datasets/pull/2780
| 2021-08-10T09:47:36
| 2021-08-12T11:09:30
| 2021-08-12T11:09:30
|
{
"login": "binh234",
"id": 57580923,
"type": "User"
}
|
[] | true
|
[] |
964,775,085
| 2,779
|
Fix sacrebleu tokenizers
|
Last `sacrebleu` release (v2.0.0) has removed `sacrebleu.TOKENIZERS`: https://github.com/mjpost/sacrebleu/pull/152/files#diff-2553a315bb1f7e68c9c1b00d56eaeb74f5205aeb3a189bc3e527b122c6078795L17-R15
This PR makes a hot fix of the bug by using a private function in `sacrebleu`: `sacrebleu.metrics.bleu._get_tokenizer()`.
Eventually, this should be further fixed in order to use only public functions.
This is a partial hotfix of #2781.
|
closed
|
https://github.com/huggingface/datasets/pull/2779
| 2021-08-10T09:24:27
| 2021-08-10T11:03:08
| 2021-08-10T10:57:54
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
964,737,422
| 2,778
|
Do not pass tokenize to sacrebleu
|
Last `sacrebleu` release (v2.0.0) has removed `sacrebleu.DEFAULT_TOKENIZER`: https://github.com/mjpost/sacrebleu/pull/152/files#diff-2553a315bb1f7e68c9c1b00d56eaeb74f5205aeb3a189bc3e527b122c6078795L17-R15
This PR does not pass `tokenize` to `sacrebleu` (note that the user cannot pass it anyway) and `sacrebleu` will use its default, no matter where it is and how it is called.
Related to #2739.
This is a partial hotfix of #2781.
|
closed
|
https://github.com/huggingface/datasets/pull/2778
| 2021-08-10T08:40:37
| 2021-08-10T10:03:37
| 2021-08-10T10:03:37
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
964,696,380
| 2,777
|
Use packaging to handle versions
|
Use packaging module to handle/validate/check versions of Python packages.
Related to #2769.
|
closed
|
https://github.com/huggingface/datasets/pull/2777
| 2021-08-10T07:51:39
| 2021-08-18T13:56:27
| 2021-08-18T13:56:27
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
964,400,596
| 2,776
|
document `config.HF_DATASETS_OFFLINE` and precedence
|
https://github.com/huggingface/datasets/pull/1976 implemented `HF_DATASETS_OFFLINE`, but:
1. `config.HF_DATASETS_OFFLINE` is not documented
2. the precedence is not documented (env, config)
I'm thinking it probably should be similar to what it says https://huggingface.co/docs/datasets/loading_datasets.html#from-the-huggingface-hub about `datasets.config.IN_MEMORY_MAX_SIZE`:
Quote:
> The default in 🤗 Datasets is to memory-map the dataset on disk unless you set datasets.config.IN_MEMORY_MAX_SIZE different from 0 bytes (default). In that case, the dataset will be copied in-memory if its size is smaller than datasets.config.IN_MEMORY_MAX_SIZE bytes, and memory-mapped otherwise. This behavior can be enabled by setting either the configuration option datasets.config.IN_MEMORY_MAX_SIZE (higher precedence) or the environment variable HF_DATASETS_IN_MEMORY_MAX_SIZE (lower precedence) to nonzero.
Context: trying to use `config.HF_DATASETS_OFFLINE` here:
https://github.com/bigscience-workshop/Megatron-DeepSpeed/pull/48
but are uncertain if it's safe, since it's not documented as a public API.
Thank you!
@lhoestq, @albertvillanova
|
open
|
https://github.com/huggingface/datasets/issues/2776
| 2021-08-09T21:23:17
| 2021-08-09T21:23:17
| null |
{
"login": "stas00",
"id": 10676103,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
964,303,626
| 2,775
|
`generate_random_fingerprint()` deterministic with 🤗Transformers' `set_seed()`
|
## Describe the bug
**Update:** I dug into this to try to reproduce the underlying issue, and I believe it's that `set_seed()` from the `transformers` library makes the "random" fingerprint identical each time. I believe this is still a bug, because `datasets` is used exactly this way in `transformers` after `set_seed()` has been called, and I think that using `set_seed()` is a standard procedure to aid reproducibility. I've added more details to reproduce this below.
Hi there! I'm using my own local dataset and custom preprocessing function. My preprocessing function seems to be unpickle-able, perhaps because it is from a closure (will debug this separately). I get this warning, which is expected:
https://github.com/huggingface/datasets/blob/450b9174765374111e5c6daab0ed294bc3d9b639/src/datasets/fingerprint.py#L260-L265
However, what's not expected is that the `datasets` actually _does_ seem to cache and reuse this dataset between runs! After that line, the next thing that's logged looks like:
```text
Loading cached processed dataset at /home/xxx/.cache/huggingface/datasets/csv/default-xxx/0.0.0/xxx/cache-xxx.arrow
```
The path is exactly the same each run (e.g., last 26 runs).
This becomes a problem because I'll pass in the `--max_eval_samples` flag to the HuggingFace example script I'm running off of ([run_swag.py](https://github.com/huggingface/transformers/blob/master/examples/pytorch/multiple-choice/run_swag.py)). The fact that the cached dataset is reused means this flag gets ignored. I'll try to load 100 examples, and it will load the full cached 1,000,000.
I think that
https://github.com/huggingface/datasets/blob/450b9174765374111e5c6daab0ed294bc3d9b639/src/datasets/fingerprint.py#L248
... is actually consistent because randomness is being controlled in HuggingFace/Transformers for reproducibility. I've added a demo of this below.
## Steps to reproduce the bug
```python
# Contents of print_fingerprint.py
from transformers import set_seed
from datasets.fingerprint import generate_random_fingerprint
set_seed(42)
print(generate_random_fingerprint())
```
```bash
for i in {0..10}; do
python print_fingerprint.py
done
1c80317fa3b1799d
1c80317fa3b1799d
1c80317fa3b1799d
1c80317fa3b1799d
1c80317fa3b1799d
1c80317fa3b1799d
1c80317fa3b1799d
1c80317fa3b1799d
1c80317fa3b1799d
1c80317fa3b1799d
1c80317fa3b1799d
```
## Expected results
After the "random hash" warning is emitted, a random hash is generated, and no outdated cached datasets are reused.
## Actual results
After the "random hash" warning is emitted, an identical hash is generated each time, and an outdated cached dataset is reused each run.
## Environment info
<!-- You can run the command `datasets-cli env` and copy-and-paste its output below. -->
- `datasets` version: 1.9.0
- Platform: Linux-5.8.0-1038-gcp-x86_64-with-glibc2.31
- Python version: 3.9.6
- PyArrow version: 4.0.1
|
closed
|
https://github.com/huggingface/datasets/issues/2775
| 2021-08-09T19:28:51
| 2024-01-26T15:05:36
| 2024-01-26T15:05:35
|
{
"login": "mbforbes",
"id": 1170062,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
963,932,199
| 2,774
|
Prevent .map from using multiprocessing when loading from cache
|
## Context
On our setup, we use different setup to train vs proprocessing datasets. Usually we are able to obtain a high number of cpus to preprocess, which allows us to use `num_proc` however we can't use as many during training phase. Currently if we use `num_proc={whatever the preprocessing value was}` we load from cache, but we get:
```
Traceback (most recent call last):
File "lib/python3.8/site-packages/multiprocess/pool.py", line 131, in worker
put((job, i, result))
File "lib/python3.8/site-packages/multiprocess/queues.py", line 371, in put
self._writer.send_bytes(obj)
File "lib/python3.8/site-packages/multiprocess/connection.py", line 203, in send_bytes
self._send_bytes(m[offset:offset + size])
File "lib/python3.8/site-packages/multiprocess/connection.py", line 414, in _send_bytes
self._send(header + buf)
File "lib/python3.8/site-packages/multiprocess/connection.py", line 371, in _send
n = write(self._handle, buf)
BrokenPipeError: [Errno 32] Broken pipe
```
Our current guess, is that we're spawning too many processes compared to the number of cpus available, and it's running OOM. Also we're loading this in DDP setting which means that for each gpu, I need to spawn a high number of processes to match the preprocessing fingerprint.
Instead what we suggest:
- Allow loading shard sequentially, sharing the same fingerprint as the multiprocessed one, in order to leverage multiprocessing when we actually generate the cache, and remove it when loading from cache.
## Current issues
~I'm having a hard time making fingerprints match. For some reason, the multiprocessing and the sequential version generate two different hash.~
**EDIT**: Turns out multiprocessing and sequential have different `transform` value for fingerprinting (check `fingerprint_transform`) when running `_map_single`:
- sequential : `datasets.arrow_dataset.Dataset._map_single`
- multiprocessing: `datasets.arrow_dataset._map_single`
This discrepancy is caused by multiprocessing pickling the transformer function, it doesn't seem to keep the `Dataset` hierarchy. I'm still unclear on why `func.__qual_name__` isn't handled correctly in multiprocessing. But replacing `__qualname__` by `__name__` fixes the issue.
## What was done
~We try to prevent the usage of multiprocessing when loading a dataset. Instead we load all cached shards sequentially.~
I couldn't find a nice way to obtain the cached_file_name and check they all exist before deciding to use the multiprocessing flow or not. Instead I expose an optional boolean `sequential` in `map` method.
## TODO
- [x] Check that the multiprocessed version and the sequential version output the same output
- [x] Check that sequential can load multiprocessed
- [x] Check that multiprocessed can load sequential
## Test
```python
from datasets import load_dataset
from multiprocessing import Pool
import random
def process(batch, rng):
length = len(batch["text"])
return {**batch, "processed_text": [f"PROCESSED {rng.random()}" for _ in range(length)]}
dataset = load_dataset("stas/openwebtext-10k", split="train")
print(dataset.column_names)
print(type(dataset))
rng = random.Random(42)
dataset1 = dataset.map(process, batched=True, batch_size=50, num_proc=4, fn_kwargs={"rng": rng})
# This one should be loaded from cache
rng = random.Random(42)
dataset2 = dataset.map(process, batched=True, batch_size=50, num_proc=4, fn_kwargs={"rng": rng}, sequential=True)
# Just to check that the random generator was correct
print(dataset1[-1]["processed_text"])
print(dataset2[-1]["processed_text"])
```
## Other solutions
I chose to load everything sequentially, but we can probably find a way to load shards in parallel using another number of workers (essentially this would be an argument not used for fingerprinting, allowing to allow `m` shards using `n` processes, which would be very useful when same dataset have to be loaded on two different setup, and we still want to leverage cache).
Also we can use a env variable similarly to `TOKENIZERS_PARALLELISM` as this seems generally setup related (though this changes slightly if we use multiprocessing).
cc @lhoestq (since I had asked you previously on `num_proc` being used for fingerprinting). Don't know if this is acceptable.
|
closed
|
https://github.com/huggingface/datasets/pull/2774
| 2021-08-09T12:11:38
| 2021-09-09T10:20:28
| 2021-09-09T10:20:28
|
{
"login": "thomasw21",
"id": 24695242,
"type": "User"
}
|
[] | true
|
[] |
963,730,497
| 2,773
|
Remove dataset_infos.json
|
**Is your feature request related to a problem? Please describe.**
As discussed, there are infos in the `dataset_infos.json` which are redundant and we could have them only in the README file.
Others could be migrated to the README, like: "dataset_size", "size_in_bytes", "download_size", "splits.split_name.[num_bytes, num_examples]",...
However, there are others that do not seem too meaningful in the README, like the checksums.
**Describe the solution you'd like**
Open a discussion to decide what to do with the `dataset_infos.json` files: which information to be migrated and/or which information to be kept.
cc: @julien-c @lhoestq
|
closed
|
https://github.com/huggingface/datasets/issues/2773
| 2021-08-09T07:43:19
| 2024-05-04T14:52:10
| 2024-05-04T14:52:10
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
},
{
"name": "generic discussion",
"color": "c5def5"
}
] | false
|
[] |
963,348,834
| 2,772
|
Remove returned feature constrain
|
In the current version, the returned value of the map function has to be list or ndarray. However, this makes it unsuitable for many tasks. In NLP, many features are sparse like verb words, noun chunks, if we want to assign different values to different words, which will result in a large sparse matrix if we only score useful words like verb words.
Mostly, when using it on large scale, saving it as a whole takes a lot of disk storage and making it hard to read, the normal method is saving it in sparse form. However, the NumPy does not support sparse, therefore I have to use PyTorch or scipy to transform a matrix into special sparse form, which is not a form that can be transformed into list or ndarry. This violates the feature constraints of the map function.
I do appreciate the convenience of Datasets package, but I do not think the compulsory datatype constrain is necessary, in some cases, we just cannot transform it into a list or ndarray due to some reasons. Any way to fix this? Or what I can do to disable the compulsory datatype constrain?
|
open
|
https://github.com/huggingface/datasets/issues/2772
| 2021-08-08T04:01:30
| 2021-08-08T08:48:01
| null |
{
"login": "PosoSAgapo",
"id": 33200481,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
963,257,036
| 2,771
|
[WIP][Common Voice 7] Add common voice 7.0
|
This PR allows to load the new common voice dataset manually as explained when doing:
```python
from datasets import load_dataset
ds = load_dataset("./datasets/datasets/common_voice_7", "ab")
```
=>
```
Please follow the manual download instructions:
You need to manually the dataset from `https://commonvoice.mozilla.org/en/datasets`.
Make sure you choose the version `Common Voice Corpus 7.0`.
Choose a language of your choice and find the corresponding language-id, *e.g.*, `Abkhaz` with language-id `ab`. The following language-ids are available:
['ab', 'ar', 'as', 'az', 'ba', 'bas', 'be', 'bg', 'br', 'ca', 'cnh', 'cs', 'cv', 'cy', 'de', 'dv', 'el', 'en', 'eo', 'es', 'et', 'eu', 'fa', 'fi', 'fr', 'fy-NL', 'ga-IE', 'gl', 'gn', 'ha', 'hi', 'hsb', 'hu', 'hy-AM', 'ia', 'id', 'it', 'ja', 'ka', 'kab', 'kk', 'kmr', 'ky', 'lg', 'lt', 'lv', 'mn', 'mt', 'nl', 'or', 'pa-IN', 'pl', 'pt', 'rm-sursilv', 'rm-vallader', 'ro', 'ru', 'rw', 'sah', 'sk', 'sl', 'sr', 'sv-SE', 'ta', 'th', 'tr', 'tt', 'ug', 'uk', 'ur', 'uz', 'vi', 'vot', 'zh-CN', 'zh-HK', 'zh-TW']
Next, you will have to enter your email address to download the dataset in the `tar.gz` format. Save the file under <path-to-file>.
The file should then be extracted with: ``tar -xvzf <path-to-file>`` which will extract a folder called ``cv-corpus-7.0-2021-07-21``.
The dataset can then be loaded with `datasets.load_dataset("common_voice", <language-id>, data_dir="<path-to-'cv-corpus-7.0-2021-07-21'-folder>", ignore_verifications=True).
```
Having followed those instructions one can then download the data as follows:
```python
from datasets import load_dataset
ds = load_dataset("./datasets/datasets/common_voice_7", "ab", data_dir="./cv-corpus-7.0-2021-07-21/", ignore_verifications=True)
```
## TODO
- [ ] Discuss naming. Is the name ok here "common_voice_7"? The dataset script differs only really in one point from `common_voice.py` in that all the metadata is different (more hours etc...) and that it has to use manual data dir for now
- [ ] Ideally we should get a bundled download link. For `common_voice.py` there is a bundled download link: `https://voice-prod-bundler-ee1969a6ce8178826482b88e843c335139bd3fb4.s3.amazonaws.com/cv-corpus-6.1-2020-12-11/{}.tar.gz` that allows one to directly download the data. However such a link is missing for Common Voice 7. I guess we should try to contact common voice about it and ask whether we could host the data or help otherwise somehow. See: https://github.com/common-voice/common-voice-bundler/issues/15 cc @yjernite
- [ ] I did not compute the dataset.json and it would mean that I'd have to download 76 datasets totalling around 1TB manually before running the checksum command. This just takes too much time. For now the user will have to add a `ignore_verifications=True` to download the data. This step would also be much easier if we could get a bundled link
- [ ] Add dummy data
|
closed
|
https://github.com/huggingface/datasets/pull/2771
| 2021-08-07T16:01:10
| 2021-12-06T23:24:02
| 2021-12-06T23:24:02
|
{
"login": "patrickvonplaten",
"id": 23423619,
"type": "User"
}
|
[] | true
|
[] |
963,246,512
| 2,770
|
Add support for fast tokenizer in BertScore
|
This PR adds support for a fast tokenizer in BertScore, which has been added recently to the lib.
Fixes #2765
|
closed
|
https://github.com/huggingface/datasets/pull/2770
| 2021-08-07T15:00:03
| 2021-08-09T12:34:43
| 2021-08-09T11:16:25
|
{
"login": "mariosasko",
"id": 47462742,
"type": "User"
}
|
[] | true
|
[] |
963,240,802
| 2,769
|
Allow PyArrow from source
|
When installing pyarrow from source the version is:
```python
>>> import pyarrow; pyarrow.__version__
'2.1.0.dev612'
```
-> however this breaks the install check at init of `datasets`. This PR makes sure that everything coming after the last `'.'` is removed.
|
closed
|
https://github.com/huggingface/datasets/pull/2769
| 2021-08-07T14:26:44
| 2021-08-09T15:38:39
| 2021-08-09T15:38:39
|
{
"login": "patrickvonplaten",
"id": 23423619,
"type": "User"
}
|
[] | true
|
[] |
963,229,173
| 2,768
|
`ArrowInvalid: Added column's length must match table's length.` after using `select`
|
## Describe the bug
I would like to add a column to a downsampled dataset. However I get an error message saying the length don't match with the length of the unsampled dataset indicated. I suspect that the dataset size is not updated when calling `select`.
## Steps to reproduce the bug
```python
from datasets import load_dataset
ds = load_dataset("tweets_hate_speech_detection")['train'].select(range(128))
ds = ds.add_column('ones', [1]*128)
```
## Expected results
I would expect a new column named `ones` filled with `1`. When I check the length of `ds` it says `128`. Interestingly, it works when calling `ds = ds.map(lambda x: x)` before adding the column.
## Actual results
Specify the actual results or traceback.
```python
---------------------------------------------------------------------------
ArrowInvalid Traceback (most recent call last)
/var/folders/l4/2905jygx4tx5jv8_kn03vxsw0000gn/T/ipykernel_6301/868709636.py in <module>
1 from datasets import load_dataset
2 ds = load_dataset("tweets_hate_speech_detection")['train'].select(range(128))
----> 3 ds = ds.add_column('ones', [0]*128)
~/git/semantic-clustering/env/lib/python3.8/site-packages/datasets/arrow_dataset.py in wrapper(*args, **kwargs)
183 }
184 # apply actual function
--> 185 out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
186 datasets: List["Dataset"] = list(out.values()) if isinstance(out, dict) else [out]
187 # re-apply format to the output
~/git/semantic-clustering/env/lib/python3.8/site-packages/datasets/fingerprint.py in wrapper(*args, **kwargs)
395 # Call actual function
396
--> 397 out = func(self, *args, **kwargs)
398
399 # Update fingerprint of in-place transforms + update in-place history of transforms
~/git/semantic-clustering/env/lib/python3.8/site-packages/datasets/arrow_dataset.py in add_column(self, name, column, new_fingerprint)
2965 column_table = InMemoryTable.from_pydict({name: column})
2966 # Concatenate tables horizontally
-> 2967 table = ConcatenationTable.from_tables([self._data, column_table], axis=1)
2968 # Update features
2969 info = self.info.copy()
~/git/semantic-clustering/env/lib/python3.8/site-packages/datasets/table.py in from_tables(cls, tables, axis)
715 table_blocks = to_blocks(table)
716 blocks = _extend_blocks(blocks, table_blocks, axis=axis)
--> 717 return cls.from_blocks(blocks)
718
719 @property
~/git/semantic-clustering/env/lib/python3.8/site-packages/datasets/table.py in from_blocks(cls, blocks)
663 return cls(table, blocks)
664 else:
--> 665 table = cls._concat_blocks_horizontally_and_vertically(blocks)
666 return cls(table, blocks)
667
~/git/semantic-clustering/env/lib/python3.8/site-packages/datasets/table.py in _concat_blocks_horizontally_and_vertically(cls, blocks)
623 if not tables:
624 continue
--> 625 pa_table_horizontally_concatenated = cls._concat_blocks(tables, axis=1)
626 pa_tables_to_concat_vertically.append(pa_table_horizontally_concatenated)
627 return cls._concat_blocks(pa_tables_to_concat_vertically, axis=0)
~/git/semantic-clustering/env/lib/python3.8/site-packages/datasets/table.py in _concat_blocks(blocks, axis)
612 else:
613 for name, col in zip(table.column_names, table.columns):
--> 614 pa_table = pa_table.append_column(name, col)
615 return pa_table
616 else:
~/git/semantic-clustering/env/lib/python3.8/site-packages/pyarrow/table.pxi in pyarrow.lib.Table.append_column()
~/git/semantic-clustering/env/lib/python3.8/site-packages/pyarrow/table.pxi in pyarrow.lib.Table.add_column()
~/git/semantic-clustering/env/lib/python3.8/site-packages/pyarrow/error.pxi in pyarrow.lib.pyarrow_internal_check_status()
~/git/semantic-clustering/env/lib/python3.8/site-packages/pyarrow/error.pxi in pyarrow.lib.check_status()
ArrowInvalid: Added column's length must match table's length. Expected length 31962 but got length 128
```
## Environment info
- `datasets` version: 1.11.0
- Platform: macOS-10.16-x86_64-i386-64bit
- Python version: 3.8.5
- PyArrow version: 5.0.0
|
closed
|
https://github.com/huggingface/datasets/issues/2768
| 2021-08-07T13:17:29
| 2021-08-09T11:26:43
| 2021-08-09T11:26:43
|
{
"login": "lvwerra",
"id": 8264887,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
963,002,120
| 2,767
|
equal operation to perform unbatch for huggingface datasets
|
Hi
I need to use "unbatch" operation in tensorflow on a huggingface dataset, I could not find this operation, could you kindly direct me how I can do it, here is the problem I am trying to solve:
I am considering "record" dataset in SuperGlue and I need to replicate each entery of the dataset for each answer, to make it similar to what T5 originally did:
https://github.com/google-research/text-to-text-transfer-transformer/blob/3c58859b8fe72c2dbca6a43bc775aa510ba7e706/t5/data/preprocessors.py#L925
Here please find an example:
For example, a typical example from ReCoRD might look like
{
'passsage': 'This is the passage.',
'query': 'A @placeholder is a bird.',
'entities': ['penguin', 'potato', 'pigeon'],
'answers': ['penguin', 'pigeon'],
}
and I need a prosessor which would turn this example into the following two examples:
{
'inputs': 'record query: A @placeholder is a bird. entities: penguin, '
'potato, pigeon passage: This is the passage.',
'targets': 'penguin',
}
and
{
'inputs': 'record query: A @placeholder is a bird. entities: penguin, '
'potato, pigeon passage: This is the passage.',
'targets': 'pigeon',
}
For doing this, one need unbatch, as each entry can map to multiple samples depending on the number of answers, I am not sure how to perform this operation with huggingface datasets library and greatly appreciate your help
@lhoestq
Thank you very much.
|
closed
|
https://github.com/huggingface/datasets/issues/2767
| 2021-08-06T19:45:52
| 2022-03-07T13:58:00
| 2022-03-07T13:58:00
|
{
"login": "dorooddorood606",
"id": 79288051,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.