Datasets:

---

Francesco-A

/

github-issues_huggingface-datasets

like 0

[ { "color": "a2eeef", "default": true, "description": "New feature or request", "id": 1935892871, "name": "enhancement", "node_id": "MDU6TGFiZWwxOTM1ODkyODcx", "url": "https://api.github.com/repos/huggingface/datasets/labels/enhancement" } ]

[ "Hi ! We're using the Arrow format for the datasets, and PyArrow tensors are not part of the Arrow format AFAIK:\r\n\r\n> There is no direct support in the arrow columnar format to store Tensors as column values.\r\n\r\nsource: https://github.com/apache/arrow/issues/4802#issuecomment-508494694", "@wesm @rok its been around three years. any updates, regarding dataset arrow tensor support? 🙏 I know you must be very busy, would appreciate to learn what is the state of art. I saw the PR is still open [#8510](https://github.com/apache/arrow/pull/8510)", "Hey @franz101 & @lhoestq!\r\nThere is a plan and a PR to create an [ExtensionArray of Tensors](https://github.com/apache/arrow/pull/8510) of equal sizes as well as a plan to do the same for Tensors of different sizes [ARROW-8714](https://issues.apache.org/jira/browse/ARROW-8714).", "The work stalled a little because it was not clear where TensorArray would live. However Arrow community recently agreed to make a [well-known-extension-type document](https://lists.apache.org/thread/sxd5fhc42hb6svs79t3fd79gkqj83pfh) and I would like https://github.com/apache/arrow/pull/8510 to land there and add an implementation to C++/Python + another language. Is that something you would find beneficial to you?", "that is a great update, thank you.\r\nit looks like this feature would benefit datasets implementation of [ArrayExtensionArray](https://github.com/huggingface/datasets/blob/9f2ff14673cac1f1ad56d80221a793f5938b68c7/src/datasets/features/features.py#L585-L641). Is that correct @eladsegal @lhoestq?\r\n\r\n", "TensorArray sounds great ! Looking forward to it :)\r\n\r\nWe've had our own ExtensionArray for fixed shape tensors for a while now, hoping to see something more standardized by the arrow community.\r\n\r\nAlso super interested in the extension array for tensors of different sizes cc @mariosasko ", "[FixedShapeTensor ExtensionType](https://github.com/apache/arrow/pull/8510) was merged and will be in Arrow 12.0.0 (release is planned mid April).\r\n", "@rok Thanks for keeping us updated! I think it's best to introduce a new feature type that would use this extension type under the hood. I'll create an issue to discuss the design with the community in the coming days.\r\n\r\nAlso, is there a tentative time frame for the variable-shape Tensor extension type?", "@mariosasko please tag me in the discussion, perhaps I can contribute.\r\n\r\nAs for the [variable shape tensor array](https://github.com/apache/arrow/issues/24868) - I'd be interested in working on it but didn't see much interest in community yet. Are you saying `huggingface/datasets` could use it?", "pyarrow 12 is out 🎉, will have a look if I can work on it for the ExtensionArray", "I think these two issues need to be fixed first on the Arrow side before adding the tensor feature type here: https://github.com/apache/arrow/issues/35573 and https://github.com/apache/arrow/issues/35599.\r\n\r\n@rok We've had a couple of requests for supporting variable-shape tensors on the forum/GH, but I did not manage to find the concrete issues using the search. TF/TFDS (and PyTorch with the `nested_tensor` API) support them, so it makes sense for us to do the same eventually (the Ray project has an [extension](https://github.com/ray-project/ray/blob/42a8d1489b37243f203120899a23d919dc85bf2a/python/ray/air/util/tensor_extensions/arrow.py#L634) type to support this case)", "> @rok We've had a couple of requests for supporting variable-shape tensors on the forum/GH, but I did not manage to find the concrete issues using the search. TF/TFDS (and PyTorch with the `nested_tensor` API) support them, so it makes sense for us to do the same eventually (the Ray project has an [extension](https://github.com/ray-project/ray/blob/42a8d1489b37243f203120899a23d919dc85bf2a/python/ray/air/util/tensor_extensions/arrow.py#L634) type to support this case)\r\n\r\nThat does make sense indeed. We should probably also be careful about memory layout to enable zero-copy interface to TF/PyTorch.", "So there is no way we can use [pyarrow.Tensor](https://arrow.apache.org/docs/python/generated/pyarrow.Tensor.html#pyarrow.Tensor) ?", "Not with with the Arrow format, and therefore not in `datasets`. But they released a new [FixedShapeTensorArray](https://arrow.apache.org/docs/python/extending_types.html#fixed-size-tensor) to store tensors in Arrow format. We plan to support this in `datasets` at one point !", "There is also an open issue to enable the conversion of `pyarrow.Tensor` to `pyarrow.FixedShapeTensorType`: https://github.com/apache/arrow/issues/35068. This way one could indirectly use `pyarrow.Tensor` in Arrow format." ]

[]

[]

[]

[]

[ { "color": "d73a4a", "default": true, "description": "Something isn't working", "id": 1935892857, "name": "bug", "node_id": "MDU6TGFiZWwxOTM1ODkyODU3", "url": "https://api.github.com/repos/huggingface/datasets/labels/bug" } ]

[ "Hi ! Thanks for reporting. The error seems to happen only if you use compressed files.\r\n\r\nThe second dataset is prepared in another dataset cache directory than the first - which is normal, since the source file is different. However, it doesn't uncompress the new data file because it finds the old uncompressed data in the extraction cache directory.\r\n\r\nIf we fix the extraction cache mechanism to uncompress a local file if it changed then it should fix the issue.\r\nCurrently the extraction cache mechanism only takes into account the path of the compressed file, which is an issue.", "Facing the same issue, is there any way to overtake this issue until it will be fixed? ", "You can clear your extraction cache in the meantime (by default at `~/.cache/huggingface/datasets/downloads/extracted`)" ]

[ { "color": "d73a4a", "default": true, "description": "Something isn't working", "id": 1935892857, "name": "bug", "node_id": "MDU6TGFiZWwxOTM1ODkyODU3", "url": "https://api.github.com/repos/huggingface/datasets/labels/bug" } ]

[ "Hi @courtneysprouse, thanks for reporting.\r\n\r\nYes, you are right: by default the \"csv\" loader loads all columns as strings. \r\n\r\nYou could tweak this behavior by passing the `feature` argument to `load_dataset`, but it is also true that currently it is not possible to perform some kind of casts, due to lacking of implementation in PyArrow. For example:\r\n```python\r\nimport datasets\r\n\r\nfeatures = datasets.Features(\r\n {\r\n \"tokens\": datasets.Sequence(datasets.Value(\"string\")),\r\n \"ner_tags\": datasets.Sequence(datasets.Value(\"int32\")),\r\n }\r\n)\r\n\r\nnew_conll = datasets.load_dataset(\"csv\", data_files=\"ner_conll.csv\", features=features)\r\n```\r\ngives `ArrowNotImplementedError` error:\r\n```\r\n/usr/local/lib/python3.7/dist-packages/pyarrow/error.pxi in pyarrow.lib.check_status()\r\n\r\nArrowNotImplementedError: Unsupported cast from string to list using function cast_list\r\n```\r\n\r\nOn the other hand, if you just would like to save and afterwards load your dataset, you could use `save_to_disk` and `load_from_disk` instead. These functions preserve all data types.\r\n```python\r\n>>> orig_conll.save_to_disk(\"ner_conll\")\r\n\r\n>>> from datasets import load_from_disk\r\n\r\n>>> new_conll = load_from_disk(\"ner_conll\")\r\n>>> new_conll\r\nDatasetDict({\r\n train: Dataset({\r\n features: ['id', 'tokens', 'pos_tags', 'chunk_tags', 'ner_tags'],\r\n num_rows: 14042\r\n })\r\n validation: Dataset({\r\n features: ['id', 'tokens', 'pos_tags', 'chunk_tags', 'ner_tags'],\r\n num_rows: 3251\r\n })\r\n test: Dataset({\r\n features: ['id', 'tokens', 'pos_tags', 'chunk_tags', 'ner_tags'],\r\n num_rows: 3454\r\n })\r\n})\r\n>>> new_conll[\"train\"][0]\r\n{'chunk_tags': [11, 21, 11, 12, 21, 22, 11, 12, 0],\r\n 'id': '0',\r\n 'ner_tags': [3, 0, 7, 0, 0, 0, 7, 0, 0],\r\n 'pos_tags': [22, 42, 16, 21, 35, 37, 16, 21, 7],\r\n 'tokens': ['EU',\r\n 'rejects',\r\n 'German',\r\n 'call',\r\n 'to',\r\n 'boycott',\r\n 'British',\r\n 'lamb',\r\n '.']}\r\n>>> new_conll[\"train\"].features\r\n{'chunk_tags': Sequence(feature=ClassLabel(num_classes=23, names=['O', 'B-ADJP', 'I-ADJP', 'B-ADVP', 'I-ADVP', 'B-CONJP', 'I-CONJP', 'B-INTJ', 'I-INTJ', 'B-LST', 'I-LST', 'B-NP', 'I-NP', 'B-PP', 'I-PP', 'B-PRT', 'I-PRT', 'B-SBAR', 'I-SBAR', 'B-UCP', 'I-UCP', 'B-VP', 'I-VP'], id=None), length=-1, id=None),\r\n 'id': Value(dtype='string', id=None),\r\n 'ner_tags': Sequence(feature=ClassLabel(num_classes=9, names=['O', 'B-PER', 'I-PER', 'B-ORG', 'I-ORG', 'B-LOC', 'I-LOC', 'B-MISC', 'I-MISC'], id=None), length=-1, id=None),\r\n 'pos_tags': Sequence(feature=ClassLabel(num_classes=47, names=['\"', \"''\", '#', '$', '(', ')', ',', '.', ':', '``', 'CC', 'CD', 'DT', 'EX', 'FW', 'IN', 'JJ', 'JJR', 'JJS', 'LS', 'MD', 'NN', 'NNP', 'NNPS', 'NNS', 'NN|SYM', 'PDT', 'POS', 'PRP', 'PRP$', 'RB', 'RBR', 'RBS', 'RP', 'SYM', 'TO', 'UH', 'VB', 'VBD', 'VBG', 'VBN', 'VBP', 'VBZ', 'WDT', 'WP', 'WP$', 'WRB'], id=None), length=-1, id=None),\r\n 'tokens': Sequence(feature=Value(dtype='string', id=None), length=-1, id=None)}\r\n```", "Hi @albertvillanova!\r\n\r\nThanks so much for your suggestions! That worked! " ]

[]

[ "Thanks for working on this. However, having thought about this issue a bit more, supporting this doesn't seem like a good idea - it's better to be explicit than implicit, according to the Zen of Python 🙂. Also, I don't think many users would use this, so this raises the question of whether this is something we want to maintain.\r\n\r\ncc @lhoestq for the 2nd opinion", "@mariosasko I was going to quote the Zen of Python in the other direction :) To me, this actually is much more explicit than the current behavior of guessing pyarrow types based on the raw dictionary return values. Explicit typehinting is increasingly the de facto way to deal with this dynamic type serialization - plus it feels like a clearer fit to me than separating out the mapper function from the feature column definition in the call to the actual `.map()`. Another benefit is providing typehinting support for clients that use mypy or other static typecheckers to detect return mismatches.\r\n\r\nBut will leave it to you and @lhoestq to see if it's something you'd like in core versus a support package.", "I meant that explicitly specifying the target features (the `features` param) is cleaner (easier to track) than relying on type hints.", "Passing features= to `map()` is richer and more explicit. Also I don't think users would guess that such API exist.\r\n\r\nOther libraries like dask also infer the type from the output or requires the typing to be specified using the `meta` argument", "Point about discoverability is a fair one, would certainly need some docs around it. All good! Will close this out and keep in our extension utilities." ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "Yes, as discussed in #4006 we should update facebook/multilingual_librispeech indeed before we do a release. @anton-l could you help taking care of updating facebook/multilingual_librispeech ? We need to update the task template\r\n```python\r\ntask_templates=[AutomaticSpeechRecognition(audio_column=\"audio\", transcription_column=\"text\")],\r\n```\r\nand write that `datasets>=2.1` is necessary to load it in the dataset card.\r\n\r\nOnce the change is done we can merge this PR and do the release I think", "@polinaeterna @lhoestq \r\nUpdated the script and the dataset card: https://huggingface.co/datasets/facebook/multilingual_librispeech ", "@anton-l @lhoestq now previewer doesn't work for this datasets as it cannot recognize new `audio_column` argument:\r\n\r\n\r\nI'm not an expert in previewer things, where should I look into the corresponding code?", "Yes, there are several datasets with the same error, eg https://github.com/huggingface/datasets-preview-backend/issues/188. I'm not sure what I should do to fix this? Upgrade datasets to master?\r\n", "@anton-l ended up removing the task template in facebook/multilingual_librispeech to make it work for the current version of `datasets` and fix the viewer :) thanks !", "@lhoestq can we merge now? ^^" ]

[]

[ "Awesome! ", "Hi, yes, this seems to fix #160 -- I cloned the branch locally and verified", "Perfect then :)" ]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_5992). All of your documentation changes will be reflected on that endpoint." ]

[]

[ "Hi @zphang,\r\n\r\nSo you can just give the local path to a dataset script file and it should work.\r\n\r\nHere is an example:\r\n- you can download one of the scripts in the `datasets` folder of the present repo (or clone the repo)\r\n- then you can load it with `load_dataset('PATH/TO/YOUR/LOCAL/SCRIPT.py')`\r\n\r\nDoes it make sense?", "Could you give a more concrete example, please? \r\n\r\nI looked up wikitext dataset script from the repo. Should I just overwrite the `data_file` on line 98 to point to the local dataset directory? Would it work for different configurations of wikitext (wikitext2, wikitext103 etc.)?\r\n\r\nOr maybe we can use DownloadManager to specify local dataset location? In that case, where do we use DownloadManager instance?\r\n\r\nThanks", "Hi @MaveriQ , although what I am doing is to commit a new dataset, but I think looking at imdb script might help.\r\nYou may want to use `dl_manager.download_custom`, give it a url(arbitrary string), a custom_download(arbitrary function) and return a path, and finally use _get sample to fetch a sample.", "The download manager supports local directories. You can specify a local directory instead of a url and it should work.", "Closing this one.\r\nFeel free to re-open if you have other questions :)" ]

[]

[]

[]

[ "BTW, is there a one-line command to make our building scripts pass flake8 test? (included code quality test), I got like trailing space or mixed space and tab warning and error, and fixed them manually.", "> BTW, is there a one-line command to make our building scripts pass flake8 test? (included code quality test), I got like trailing space or mixed space and tab warning and error, and fixed them manually.\r\n\r\nI don't think so.\r\nWe have a command for black and isort but not flake8 as far as I know.", "Thanks for your awesome work too.\r\nBTW a little reminder, this solves #132 " ]

[]

[]

[]

[ "> \r\n\r\nI did `make style` and it affected over 500 files\r\n\r\n```\r\nAll done! ✨ 🍰 ✨\r\n575 files reformatted, 372 files left unchanged.\r\nisort tests src benchmarks datasets/**/*.py metri\r\n```\r\n\r\n(result)\r\n\r\n", "Nvm I was using wrong black version" ]

[ { "color": "e99695", "default": false, "description": "Requesting to add a new dataset", "id": 2067376369, "name": "dataset request", "node_id": "MDU6TGFiZWwyMDY3Mzc2MzY5", "url": "https://api.github.com/repos/huggingface/datasets/labels/dataset%20request" } ]

[ "I'll take this one if no one hasn't yet!" ]

[]

[]

[ { "color": "d73a4a", "default": true, "description": "Something isn't working", "id": 1935892857, "name": "bug", "node_id": "MDU6TGFiZWwxOTM1ODkyODU3", "url": "https://api.github.com/repos/huggingface/datasets/labels/bug" } ]

[]

[]

[ "Hi @faizankshaikh\r\n\r\nPlease note that we have recently launched the Community feature, specifically targeted to create Discussions (about issues/questions/asking-for-help) on each Dataset on the Hub:\r\n- Blog post: https://huggingface.co/blog/community-update\r\n- Docs: https://huggingface.co/docs/hub/repositories-pull-requests-discussions\r\n\r\nThe Discussion tab for your \"jalFaizy/detect_chess_pieces\" dataset is here: https://huggingface.co/datasets/jalFaizy/detect_chess_pieces/discussions\r\nYou can use it to ask for help by pinging the Datasets maintainers: see our docs here: https://huggingface.co/docs/datasets/master/en/share#ask-for-a-help-and-reviews\r\n\r\nI'm transferring this discussion to your Discussion tab and trying to address it: https://huggingface.co/datasets/jalFaizy/detect_chess_pieces/discussions/1", "Thank you @albertvillanova , I will keep that in mind.\r\n\r\nJust a quick note - I posted the issue on Github because the dataset viewer suggested me to \"open an issue for direct support\". Maybe it can be updated with your suggestion\r\n\r\n\r\n\r\n\r\n", "Thank you pointing this out: yes, definitely, we should fix the error message. We are working on this." ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "<details>\n<summary>Show benchmarks</summary>\n\nPyArrow==8.0.0\n\n<details>\n<summary>Show updated benchmarks!</summary>\n\n### Benchmark: benchmark_array_xd.json\n\n| metric | read_batch_formatted_as_numpy after write_array2d | read_batch_formatted_as_numpy after write_flattened_sequence | read_batch_formatted_as_numpy after write_nested_sequence | read_batch_unformated after write_array2d | read_batch_unformated after write_flattened_sequence | read_batch_unformated after write_nested_sequence | read_col_formatted_as_numpy after write_array2d | read_col_formatted_as_numpy after write_flattened_sequence | read_col_formatted_as_numpy after write_nested_sequence | read_col_unformated after write_array2d | read_col_unformated after write_flattened_sequence | read_col_unformated after write_nested_sequence | read_formatted_as_numpy after write_array2d | read_formatted_as_numpy after write_flattened_sequence | read_formatted_as_numpy after write_nested_sequence | read_unformated after write_array2d | read_unformated after write_flattened_sequence | read_unformated after write_nested_sequence | write_array2d | write_flattened_sequence | write_nested_sequence |\n|--------|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 0.008698 / 0.011353 (-0.002655) | 0.005250 / 0.011008 (-0.005758) | 0.104101 / 0.038508 (0.065593) | 0.085021 / 0.023109 (0.061912) | 0.426653 / 0.275898 (0.150755) | 0.460449 / 0.323480 (0.136969) | 0.005222 / 0.007986 (-0.002763) | 0.006280 / 0.004328 (0.001951) | 0.083458 / 0.004250 (0.079207) | 0.066132 / 0.037052 (0.029079) | 0.433416 / 0.258489 (0.174927) | 0.482718 / 0.293841 (0.188877) | 0.048872 / 0.128546 (-0.079675) | 0.013699 / 0.075646 (-0.061948) | 0.365660 / 0.419271 (-0.053611) | 0.071008 / 0.043533 (0.027475) | 0.428688 / 0.255139 (0.173549) | 0.443554 / 0.283200 (0.160354) | 0.035901 / 0.141683 (-0.105782) | 1.829296 / 1.452155 (0.377141) | 1.862351 / 1.492716 (0.369635) |\n\n### Benchmark: benchmark_getitem\\_100B.json\n\n| metric | get_batch_of\\_1024\\_random_rows | get_batch_of\\_1024\\_rows | get_first_row | get_last_row |\n|--------|---|---|---|---|\n| new / old (diff) | 0.236284 / 0.018006 (0.218278) | 0.584075 / 0.000490 (0.583585) | 0.004634 / 0.000200 (0.004434) | 0.000125 / 0.000054 (0.000070) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.034723 / 0.037411 (-0.002688) | 0.100989 / 0.014526 (0.086464) | 0.113722 / 0.176557 (-0.062834) | 0.187659 / 0.737135 (-0.549477) | 0.113937 / 0.296338 (-0.182401) |\n\n### Benchmark: benchmark_iterating.json\n\n| metric | read 5000 | read 50000 | read_batch 50000 10 | read_batch 50000 100 | read_batch 50000 1000 | read_formatted numpy 5000 | read_formatted pandas 5000 | read_formatted tensorflow 5000 | read_formatted torch 5000 | read_formatted_batch numpy 5000 10 | read_formatted_batch numpy 5000 1000 | shuffled read 5000 | shuffled read 50000 | shuffled read_batch 50000 10 | shuffled read_batch 50000 100 | shuffled read_batch 50000 1000 | shuffled read_formatted numpy 5000 | shuffled read_formatted_batch numpy 5000 10 | shuffled read_formatted_batch numpy 5000 1000 |\n|--------|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 0.587500 / 0.215209 (0.372291) | 5.847371 / 2.077655 (3.769716) | 2.599691 / 1.504120 (1.095571) | 2.246187 / 1.541195 (0.704992) | 2.419126 / 1.468490 (0.950636) | 0.847327 / 4.584777 (-3.737450) | 5.230438 / 3.745712 (1.484726) | 7.539021 / 5.269862 (2.269160) | 4.617473 / 4.565676 (0.051797) | 0.103620 / 0.424275 (-0.320655) | 0.009195 / 0.007607 (0.001588) | 0.714247 / 0.226044 (0.488203) | 7.331621 / 2.268929 (5.062693) | 3.416575 / 55.444624 (-52.028049) | 2.649467 / 6.876477 (-4.227009) | 2.928091 / 2.142072 (0.786018) | 1.002155 / 4.805227 (-3.803072) | 0.210790 / 6.500664 (-6.289874) | 0.081303 / 0.075469 (0.005834) |\n\n### Benchmark: benchmark_map_filter.json\n\n| metric | filter | map fast-tokenizer batched | map identity | map identity batched | map no-op batched | map no-op batched numpy | map no-op batched pandas | map no-op batched pytorch | map no-op batched tensorflow |\n|--------|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 1.655431 / 1.841788 (-0.186357) | 24.069595 / 8.074308 (15.995287) | 20.923766 / 10.191392 (10.732374) | 0.232021 / 0.680424 (-0.448403) | 0.026355 / 0.534201 (-0.507846) | 0.496830 / 0.579283 (-0.082453) | 0.582620 / 0.434364 (0.148257) | 0.551227 / 0.540337 (0.010890) | 0.756389 / 1.386936 (-0.630547) |\n\n</details>\nPyArrow==latest\n\n<details>\n<summary>Show updated benchmarks!</summary>\n\n### Benchmark: benchmark_array_xd.json\n\n| metric | read_batch_formatted_as_numpy after write_array2d | read_batch_formatted_as_numpy after write_flattened_sequence | read_batch_formatted_as_numpy after write_nested_sequence | read_batch_unformated after write_array2d | read_batch_unformated after write_flattened_sequence | read_batch_unformated after write_nested_sequence | read_col_formatted_as_numpy after write_array2d | read_col_formatted_as_numpy after write_flattened_sequence | read_col_formatted_as_numpy after write_nested_sequence | read_col_unformated after write_array2d | read_col_unformated after write_flattened_sequence | read_col_unformated after write_nested_sequence | read_formatted_as_numpy after write_array2d | read_formatted_as_numpy after write_flattened_sequence | read_formatted_as_numpy after write_nested_sequence | read_unformated after write_array2d | read_unformated after write_flattened_sequence | read_unformated after write_nested_sequence | write_array2d | write_flattened_sequence | write_nested_sequence |\n|--------|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 0.009329 / 0.011353 (-0.002024) | 0.005045 / 0.011008 (-0.005964) | 0.082116 / 0.038508 (0.043608) | 0.082420 / 0.023109 (0.059311) | 0.502513 / 0.275898 (0.226615) | 0.526098 / 0.323480 (0.202618) | 0.007468 / 0.007986 (-0.000517) | 0.005477 / 0.004328 (0.001148) | 0.082617 / 0.004250 (0.078367) | 0.070292 / 0.037052 (0.033239) | 0.503290 / 0.258489 (0.244801) | 0.541631 / 0.293841 (0.247790) | 0.050826 / 0.128546 (-0.077721) | 0.014699 / 0.075646 (-0.060948) | 0.094441 / 0.419271 (-0.324830) | 0.065034 / 0.043533 (0.021501) | 0.486778 / 0.255139 (0.231639) | 0.516907 / 0.283200 (0.233707) | 0.045140 / 0.141683 (-0.096543) | 1.831676 / 1.452155 (0.379521) | 1.910865 / 1.492716 (0.418149) |\n\n### Benchmark: benchmark_getitem\\_100B.json\n\n| metric | get_batch_of\\_1024\\_random_rows | get_batch_of\\_1024\\_rows | get_first_row | get_last_row |\n|--------|---|---|---|---|\n| new / old (diff) | 0.286818 / 0.018006 (0.268812) | 0.558621 / 0.000490 (0.558131) | 0.002830 / 0.000200 (0.002630) | 0.000148 / 0.000054 (0.000094) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.036716 / 0.037411 (-0.000696) | 0.107830 / 0.014526 (0.093305) | 0.116368 / 0.176557 (-0.060188) | 0.178401 / 0.737135 (-0.558734) | 0.124729 / 0.296338 (-0.171609) |\n\n### Benchmark: benchmark_iterating.json\n\n| metric | read 5000 | read 50000 | read_batch 50000 10 | read_batch 50000 100 | read_batch 50000 1000 | read_formatted numpy 5000 | read_formatted pandas 5000 | read_formatted tensorflow 5000 | read_formatted torch 5000 | read_formatted_batch numpy 5000 10 | read_formatted_batch numpy 5000 1000 | shuffled read 5000 | shuffled read 50000 | shuffled read_batch 50000 10 | shuffled read_batch 50000 100 | shuffled read_batch 50000 1000 | shuffled read_formatted numpy 5000 | shuffled read_formatted_batch numpy 5000 10 | shuffled read_formatted_batch numpy 5000 1000 |\n|--------|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 0.633557 / 0.215209 (0.418348) | 6.423135 / 2.077655 (4.345480) | 2.981883 / 1.504120 (1.477763) | 2.755592 / 1.541195 (1.214398) | 2.769337 / 1.468490 (1.300847) | 0.836219 / 4.584777 (-3.748558) | 5.302030 / 3.745712 (1.556318) | 7.463960 / 5.269862 (2.194098) | 4.427254 / 4.565676 (-0.138422) | 0.095990 / 0.424275 (-0.328285) | 0.009264 / 0.007607 (0.001657) | 0.770642 / 0.226044 (0.544597) | 7.779667 / 2.268929 (5.510739) | 3.799115 / 55.444624 (-51.645509) | 3.212560 / 6.876477 (-3.663917) | 3.281657 / 2.142072 (1.139584) | 1.044981 / 4.805227 (-3.760246) | 0.210693 / 6.500664 (-6.289971) | 0.079466 / 0.075469 (0.003997) |\n\n### Benchmark: benchmark_map_filter.json\n\n| metric | filter | map fast-tokenizer batched | map identity | map identity batched | map no-op batched | map no-op batched numpy | map no-op batched pandas | map no-op batched pytorch | map no-op batched tensorflow |\n|--------|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 1.793155 / 1.841788 (-0.048632) | 24.691127 / 8.074308 (16.616819) | 22.083150 / 10.191392 (11.891758) | 0.242246 / 0.680424 (-0.438178) | 0.028001 / 0.534201 (-0.506200) | 0.494061 / 0.579283 (-0.085222) | 0.599288 / 0.434364 (0.164924) | 0.552101 / 0.540337 (0.011764) | 0.784093 / 1.386936 (-0.602843) |\n\n</details>\n</details>\n\n![](https://cml.dev/watermark.png#cd429c39604af34bc3a3ba1f463329b23fcbc1e3 \"CML watermark\")\n", "<details>\n<summary>Show benchmarks</summary>\n\nPyArrow==8.0.0\n\n<details>\n<summary>Show updated benchmarks!</summary>\n\n### Benchmark: benchmark_array_xd.json\n\n| metric | read_batch_formatted_as_numpy after write_array2d | read_batch_formatted_as_numpy after write_flattened_sequence | read_batch_formatted_as_numpy after write_nested_sequence | read_batch_unformated after write_array2d | read_batch_unformated after write_flattened_sequence | read_batch_unformated after write_nested_sequence | read_col_formatted_as_numpy after write_array2d | read_col_formatted_as_numpy after write_flattened_sequence | read_col_formatted_as_numpy after write_nested_sequence | read_col_unformated after write_array2d | read_col_unformated after write_flattened_sequence | read_col_unformated after write_nested_sequence | read_formatted_as_numpy after write_array2d | read_formatted_as_numpy after write_flattened_sequence | read_formatted_as_numpy after write_nested_sequence | read_unformated after write_array2d | read_unformated after write_flattened_sequence | read_unformated after write_nested_sequence | write_array2d | write_flattened_sequence | write_nested_sequence |\n|--------|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 0.006658 / 0.011353 (-0.004695) | 0.004044 / 0.011008 (-0.006965) | 0.085844 / 0.038508 (0.047336) | 0.077147 / 0.023109 (0.054038) | 0.344387 / 0.275898 (0.068489) | 0.376718 / 0.323480 (0.053238) | 0.005537 / 0.007986 (-0.002448) | 0.003452 / 0.004328 (-0.000876) | 0.065326 / 0.004250 (0.061076) | 0.057639 / 0.037052 (0.020587) | 0.352363 / 0.258489 (0.093873) | 0.378939 / 0.293841 (0.085098) | 0.031259 / 0.128546 (-0.097287) | 0.008464 / 0.075646 (-0.067183) | 0.289076 / 0.419271 (-0.130195) | 0.052991 / 0.043533 (0.009459) | 0.346053 / 0.255139 (0.090914) | 0.362761 / 0.283200 (0.079561) | 0.023501 / 0.141683 (-0.118182) | 1.478312 / 1.452155 (0.026157) | 1.545437 / 1.492716 (0.052721) |\n\n### Benchmark: benchmark_getitem\\_100B.json\n\n| metric | get_batch_of\\_1024\\_random_rows | get_batch_of\\_1024\\_rows | get_first_row | get_last_row |\n|--------|---|---|---|---|\n| new / old (diff) | 0.202964 / 0.018006 (0.184957) | 0.534793 / 0.000490 (0.534303) | 0.006025 / 0.000200 (0.005825) | 0.000225 / 0.000054 (0.000171) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.029418 / 0.037411 (-0.007993) | 0.084297 / 0.014526 (0.069771) | 0.096702 / 0.176557 (-0.079855) | 0.157355 / 0.737135 (-0.579781) | 0.097858 / 0.296338 (-0.198480) |\n\n### Benchmark: benchmark_iterating.json\n\n| metric | read 5000 | read 50000 | read_batch 50000 10 | read_batch 50000 100 | read_batch 50000 1000 | read_formatted numpy 5000 | read_formatted pandas 5000 | read_formatted tensorflow 5000 | read_formatted torch 5000 | read_formatted_batch numpy 5000 10 | read_formatted_batch numpy 5000 1000 | shuffled read 5000 | shuffled read 50000 | shuffled read_batch 50000 10 | shuffled read_batch 50000 100 | shuffled read_batch 50000 1000 | shuffled read_formatted numpy 5000 | shuffled read_formatted_batch numpy 5000 10 | shuffled read_formatted_batch numpy 5000 1000 |\n|--------|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 0.380728 / 0.215209 (0.165519) | 3.787712 / 2.077655 (1.710057) | 1.836393 / 1.504120 (0.332273) | 1.678415 / 1.541195 (0.137220) | 1.781800 / 1.468490 (0.313310) | 0.478677 / 4.584777 (-4.106100) | 3.614080 / 3.745712 (-0.131632) | 3.255637 / 5.269862 (-2.014225) | 2.063642 / 4.565676 (-2.502035) | 0.056470 / 0.424275 (-0.367805) | 0.007408 / 0.007607 (-0.000199) | 0.459155 / 0.226044 (0.233111) | 4.586679 / 2.268929 (2.317750) | 2.305737 / 55.444624 (-53.138888) | 1.954755 / 6.876477 (-4.921721) | 2.190809 / 2.142072 (0.048737) | 0.572426 / 4.805227 (-4.232802) | 0.130349 / 6.500664 (-6.370315) | 0.059346 / 0.075469 (-0.016124) |\n\n### Benchmark: benchmark_map_filter.json\n\n| metric | filter | map fast-tokenizer batched | map identity | map identity batched | map no-op batched | map no-op batched numpy | map no-op batched pandas | map no-op batched pytorch | map no-op batched tensorflow |\n|--------|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 1.253671 / 1.841788 (-0.588117) | 19.509015 / 8.074308 (11.434706) | 13.951349 / 10.191392 (3.759957) | 0.171038 / 0.680424 (-0.509386) | 0.018826 / 0.534201 (-0.515375) | 0.394642 / 0.579283 (-0.184642) | 0.419614 / 0.434364 (-0.014750) | 0.470931 / 0.540337 (-0.069406) | 0.643858 / 1.386936 (-0.743078) |\n\n</details>\nPyArrow==latest\n\n<details>\n<summary>Show updated benchmarks!</summary>\n\n### Benchmark: benchmark_array_xd.json\n\n| metric | read_batch_formatted_as_numpy after write_array2d | read_batch_formatted_as_numpy after write_flattened_sequence | read_batch_formatted_as_numpy after write_nested_sequence | read_batch_unformated after write_array2d | read_batch_unformated after write_flattened_sequence | read_batch_unformated after write_nested_sequence | read_col_formatted_as_numpy after write_array2d | read_col_formatted_as_numpy after write_flattened_sequence | read_col_formatted_as_numpy after write_nested_sequence | read_col_unformated after write_array2d | read_col_unformated after write_flattened_sequence | read_col_unformated after write_nested_sequence | read_formatted_as_numpy after write_array2d | read_formatted_as_numpy after write_flattened_sequence | read_formatted_as_numpy after write_nested_sequence | read_unformated after write_array2d | read_unformated after write_flattened_sequence | read_unformated after write_nested_sequence | write_array2d | write_flattened_sequence | write_nested_sequence |\n|--------|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 0.006765 / 0.011353 (-0.004587) | 0.003955 / 0.011008 (-0.007053) | 0.064377 / 0.038508 (0.025869) | 0.076980 / 0.023109 (0.053871) | 0.368675 / 0.275898 (0.092777) | 0.403746 / 0.323480 (0.080267) | 0.005303 / 0.007986 (-0.002683) | 0.003257 / 0.004328 (-0.001072) | 0.064154 / 0.004250 (0.059903) | 0.056975 / 0.037052 (0.019923) | 0.376718 / 0.258489 (0.118229) | 0.416291 / 0.293841 (0.122450) | 0.031444 / 0.128546 (-0.097102) | 0.008532 / 0.075646 (-0.067115) | 0.070455 / 0.419271 (-0.348816) | 0.049032 / 0.043533 (0.005499) | 0.361413 / 0.255139 (0.106274) | 0.384648 / 0.283200 (0.101448) | 0.024050 / 0.141683 (-0.117633) | 1.514330 / 1.452155 (0.062176) | 1.585424 / 1.492716 (0.092708) |\n\n### Benchmark: benchmark_getitem\\_100B.json\n\n| metric | get_batch_of\\_1024\\_random_rows | get_batch_of\\_1024\\_rows | get_first_row | get_last_row |\n|--------|---|---|---|---|\n| new / old (diff) | 0.214701 / 0.018006 (0.196695) | 0.447706 / 0.000490 (0.447216) | 0.000373 / 0.000200 (0.000173) | 0.000058 / 0.000054 (0.000003) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.031007 / 0.037411 (-0.006404) | 0.090545 / 0.014526 (0.076019) | 0.100611 / 0.176557 (-0.075945) | 0.154847 / 0.737135 (-0.582289) | 0.102864 / 0.296338 (-0.193475) |\n\n### Benchmark: benchmark_iterating.json\n\n| metric | read 5000 | read 50000 | read_batch 50000 10 | read_batch 50000 100 | read_batch 50000 1000 | read_formatted numpy 5000 | read_formatted pandas 5000 | read_formatted tensorflow 5000 | read_formatted torch 5000 | read_formatted_batch numpy 5000 10 | read_formatted_batch numpy 5000 1000 | shuffled read 5000 | shuffled read 50000 | shuffled read_batch 50000 10 | shuffled read_batch 50000 100 | shuffled read_batch 50000 1000 | shuffled read_formatted numpy 5000 | shuffled read_formatted_batch numpy 5000 10 | shuffled read_formatted_batch numpy 5000 1000 |\n|--------|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 0.427740 / 0.215209 (0.212531) | 4.273143 / 2.077655 (2.195488) | 2.294906 / 1.504120 (0.790786) | 2.138460 / 1.541195 (0.597265) | 2.274126 / 1.468490 (0.805636) | 0.486559 / 4.584777 (-4.098218) | 3.565554 / 3.745712 (-0.180158) | 3.377659 / 5.269862 (-1.892202) | 2.029883 / 4.565676 (-2.535793) | 0.057303 / 0.424275 (-0.366972) | 0.007314 / 0.007607 (-0.000293) | 0.504263 / 0.226044 (0.278219) | 5.041196 / 2.268929 (2.772268) | 2.819273 / 55.444624 (-52.625351) | 2.421479 / 6.876477 (-4.454998) | 2.503063 / 2.142072 (0.360991) | 0.581467 / 4.805227 (-4.223760) | 0.133532 / 6.500664 (-6.367132) | 0.062504 / 0.075469 (-0.012965) |\n\n### Benchmark: benchmark_map_filter.json\n\n| metric | filter | map fast-tokenizer batched | map identity | map identity batched | map no-op batched | map no-op batched numpy | map no-op batched pandas | map no-op batched pytorch | map no-op batched tensorflow |\n|--------|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 1.328765 / 1.841788 (-0.513022) | 20.131672 / 8.074308 (12.057363) | 14.312895 / 10.191392 (4.121503) | 0.191199 / 0.680424 (-0.489225) | 0.018522 / 0.534201 (-0.515679) | 0.393121 / 0.579283 (-0.186162) | 0.413122 / 0.434364 (-0.021242) | 0.469312 / 0.540337 (-0.071026) | 0.633140 / 1.386936 (-0.753796) |\n\n</details>\n</details>\n\n![](https://cml.dev/watermark.png#dbf6c103f5844de40431478e7e4a64fbf2c2c067 \"CML watermark\")\n" ]

[]

[ "merging since the CI is fixed on master" ]

[ { "color": "d73a4a", "default": true, "description": "Something isn't working", "id": 1935892857, "name": "bug", "node_id": "MDU6TGFiZWwxOTM1ODkyODU3", "url": "https://api.github.com/repos/huggingface/datasets/labels/bug" } ]

[]

[]

[]

[]

[ "Looks like this PR includes changes about many other files than the ones for clickbait_news_bg\r\n\r\nCan you create another branch and another PR please ?", "I created a new branch with the dataset code and submitted a new PR for it: https://github.com/huggingface/datasets/pull/1568" ]

[]

[]

[ { "color": "d4c5f9", "default": false, "description": "Maintenance tasks", "id": 4296013012, "name": "maintenance", "node_id": "LA_kwDODunzps8AAAABAA_01A", "url": "https://api.github.com/repos/huggingface/datasets/labels/maintenance" } ]

[ "It seems all the tasks have been addressed, meaning this issue can be closed, no?" ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "Looks like your dummy_data.zip file is not in the right location ;)\r\ndatasets/datasets/conceptual_12m/dummy/default/0.0.0/dummy_data.zip\r\n->\r\ndatasets/conceptual_12m/dummy/default/0.0.0/dummy_data.zip" ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "This looks like a breaking change no ?\r\nAlso could you explain why it would be better this way ?", "The might be only there to help type checkers, but I am not too familiar with the code base to know for sure. I think this might be useful:\n\nhttps://docs.python.org/3/library/typing.html#typing.TYPE_CHECKING", "> This looks like a breaking change no ?\r\n> Also could you explain why it would be better this way ?\r\n\r\nSorry, @lhoestq, I naively thought it was obvious. I have tried to give some arguments in the motivation of this PR (see above). I can give additional arguments if needed. " ]

[]

[]

[]

[ "Sorry for asking a duplicate question about `num_proc`, I searched the forum and find the solution.\r\n\r\nBut I still can't make the trick with `torch.distributed.barrier()` to only map at the main process work. The [post on forum]( https://discuss.huggingface.co/t/slow-processing-with-map-when-using-deepspeed-or-fairscale/7229/7) didn't help.", "If it does the `map` twice then it means the hash of your map function is not some same between your two processes.\r\n\r\nCan you make sure your map functions have the same hash in different processes ?\r\n\r\n```python\r\nfrom datasets.fingerprint import Hasher\r\n\r\nprint(Hasher.hash(lambda x: cut_reorder_keys(x, num_stations_list=args.num_stations_list, is_pad=True, is_train=True)))\r\nprint(Hasher.hash(lambda x: random_shift(x, shift_range=(-160, 0), feature_scale=16)))\r\n```\r\n\r\nYou can also set the fingerprint used to reload the resulting dataset by passing `new_finegrprint=` in `map`, see https://huggingface.co/docs/datasets/v2.13.1/en/about_cache#the-cache. This will force the different processes to use the same fingerprint used to locate the resulting dataset in the cache.", "Thanks for help! I find the fingerprint between processes don't have same hash:\r\n```\r\nRank 0: Gpu 0 cut_reorder_keys fingerprint c7f47f40e9a67657\r\nRank 0: Gpu 0 random_shift fingerprint 240a0ce79831e7d4\r\n\r\nRank 1: Gpu 1 cut_reorder_keys fingerprint 20edd3d9cf284001\r\nRank 1: Gpu 1 random_shift fingerprint 819f7c1c18e7733f\r\n```\r\nBut my functions only process the example one by one and don't need rank or other arguments. After all it can work in the test for dataset and dataloader.\r\nI'll try to set `new_fingerprint` to see if it works and figure out the reason of different hash." ]

[]

[ "@lhoestq Thanks for the quick review.", "I merged master to this branch to re-run the CI before merging :)" ]

[]

[]

[]

[]

[]

[]

[]

[ "Yes! might do it with @srush one of these days. Hopefully it won't break too many links (we can always redirect from old url to new)", "This was fixed but forgot to close the issue. cc @lhoestq @yjernite \r\n\r\nThanks @jarednielsen!" ]

[ { "color": "d73a4a", "default": true, "description": "Something isn't working", "id": 1935892857, "name": "bug", "node_id": "MDU6TGFiZWwxOTM1ODkyODU3", "url": "https://api.github.com/repos/huggingface/datasets/labels/bug" } ]

[ "Thanks for reporting ! We're looking into it", "I'm not able to reproduce this, do you think you can provide a code that creates a DatasetDict that has this issue when saving and reloading ?", "Hi, I just ran into a similar error. Here is the minimal code to reproduce:\r\n```python\r\nfrom datasets import load_dataset, DatasetDict\r\nds = load_dataset('super_glue', 'multirc')\r\n\r\nds.save_to_disk('tempds')\r\n\r\nds = DatasetDict.load_from_disk('tempds')\r\n\r\n```\r\n\r\n```bash\r\nReusing dataset super_glue (/home/idahl/.cache/huggingface/datasets/super_glue/multirc/1.0.2/2fb163bca9085c1deb906aff20f00c242227ff704a4e8c9cfdfe820be3abfc83)\r\nTraceback (most recent call last):\r\n File \"/home/idahl/eval-util-expl/multirc/tmp.py\", line 7, in <module>\r\n ds = DatasetDict.load_from_disk('tempds')\r\n File \"/home/idahl/miniconda3/envs/eval-util-expl/lib/python3.9/site-packages/datasets/dataset_dict.py\", line 710, in load_from_disk\r\n dataset_dict[k] = Dataset.load_from_disk(dataset_dict_split_path, fs, keep_in_memory=keep_in_memory)\r\n File \"/home/idahl/miniconda3/envs/eval-util-expl/lib/python3.9/site-packages/datasets/arrow_dataset.py\", line 687, in load_from_disk\r\n return Dataset(\r\n File \"/home/idahl/miniconda3/envs/eval-util-expl/lib/python3.9/site-packages/datasets/arrow_dataset.py\", line 274, in __init__\r\n raise ValueError(\r\nValueError: External features info don't match the dataset:\r\nGot\r\n{'answer': Value(dtype='string', id=None), 'idx': {'answer': Value(dtype='int32', id=None), 'paragraph': Value(dtype='int32', id=None), 'question': Value(dtype='int32', id=None)}, 'label': ClassLabel(num_classes=2, names=['False', 'True'], names_file=None, id=None), 'paragraph': Value(dtype='string', id=None), 'question': Value(dtype='string', id=None)}\r\nwith type\r\nstruct<answer: string, idx: struct<answer: int32, paragraph: int32, question: int32>, label: int64, paragraph: string, question: string>\r\n\r\nbut expected something like\r\n{'answer': Value(dtype='string', id=None), 'idx': {'paragraph': Value(dtype='int32', id=None), 'question': Value(dtype='int32', id=None), 'answer': Value(dtype='int32', id=None)}, 'label': Value(dtype='int64', id=None), 'paragraph': Value(dtype='string', id=None), 'question': Value(dtype='string', id=None)}\r\nwith type\r\nstruct<answer: string, idx: struct<paragraph: int32, question: int32, answer: int32>, label: int64, paragraph: string, question: string>\r\n\r\n```\r\n\r\nThe non-matching part seems to be\r\n`'label': ClassLabel(num_classes=2, names=['False', 'True'], names_file=None, id=None),`\r\nvs \r\n`'label': Value(dtype='int64', id=None),`\r\n\r\nAnd the order in the `<struct...` being different, which might cause the [features.type != inferred_features.type](https://github.com/huggingface/datasets/blob/master/src/datasets/arrow_dataset.py#L274) condition to become true and raise this ValueError.\r\n\r\n\r\nI am using datasets version 1.6.2.\r\n\r\nEdit: can confirm, this works without error in version 1.5.0", "My current workaround is to remove the idx feature:\r\n\r\n```\r\n\r\nfrom datasets import load_dataset, DatasetDict, Value\r\nds = load_dataset('super_glue', 'multirc')\r\nds = ds.remove_columns('idx')\r\n\r\nds.save_to_disk('tempds')\r\n\r\nds = DatasetDict.load_from_disk('tempds')\r\n\r\n```\r\n\r\nworks.", "It looks like this issue comes from the order of the fields in the 'idx' struct that is different for some reason.\r\nI'm looking into it. Note that as a workaround you can also flatten the nested features with `ds = ds.flatten()`", "I just pushed a fix on `master`. We'll do a new release soon !\r\n\r\nThanks for reporting" ]

[]

[ "merging since the CI is fixed on master" ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "Ok recomputed the json file, this should be ready to review now! @lhoestq ", "Note: the generated JSON from `hf/hub-docs` can be found in the output of a GitHub Action run on that repo, for instance in https://github.com/huggingface/hub-docs/runs/6006686983?check_suite_focus=true\r\n\r\n(click on \"Run export-tasks script\")", "Should we not add the tasks with hideInDatasets?", "yes, probably true – i'll change that in a PR in `hub-docs`", "Yes that's good :) feel free to merge", "thanks to the both of you!" ]

[ { "color": "a2eeef", "default": true, "description": "New feature or request", "id": 1935892871, "name": "enhancement", "node_id": "MDU6TGFiZWwxOTM1ODkyODcx", "url": "https://api.github.com/repos/huggingface/datasets/labels/enhancement" }, { "color": "7057ff", "default": true, "description": "Good for newcomers", "id": 1935892877, "name": "good first issue", "node_id": "MDU6TGFiZWwxOTM1ODkyODc3", "url": "https://api.github.com/repos/huggingface/datasets/labels/good%20first%20issue" } ]

[ "Hi! Linking a comment to provide more info on the issue: https://stackoverflow.com/a/39458225. This means we should replace all (most of) the occurrences of `List` with `Sequence` in function signatures.\r\n\r\n@tranhd95 Would you be interested in submitting a PR?", "Hi all! I tried to reproduce this issue and didn't work for me. Also in your example i noticed that the variables have different names: `list_of_filenames` and `list_of_files`, could this be related to that?\r\n```python\r\n#I found random data in parquet format:\r\n!wget \"https://github.com/Teradata/kylo/raw/master/samples/sample-data/parquet/userdata1.parquet\"\r\n!wget \"https://github.com/Teradata/kylo/raw/master/samples/sample-data/parquet/userdata2.parquet\"\r\n\r\n#Then i try reproduce\r\nlist_of_files = [\"userdata1.parquet\", \"userdata2.parquet\"]\r\nds = Dataset.from_parquet(list_of_files)\r\n```\r\n**My output:**\r\n```python\r\nWARNING:datasets.builder:Using custom data configuration default-e287d097dc54e046\r\nDownloading and preparing dataset parquet/default to /root/.cache/huggingface/datasets/parquet/default-e287d097dc54e046/0.0.0/2a3b91fbd88a2c90d1dbbb32b460cf621d31bd5b05b934492fdef7d8d6f236ec...\r\nDownloading data files: 100%\r\n1/1 [00:00<00:00, 40.38it/s]\r\nExtracting data files: 100%\r\n1/1 [00:00<00:00, 23.43it/s]\r\nDataset parquet downloaded and prepared to /root/.cache/huggingface/datasets/parquet/default-e287d097dc54e046/0.0.0/2a3b91fbd88a2c90d1dbbb32b460cf621d31bd5b05b934492fdef7d8d6f236ec. Subsequent calls will reuse this data.\r\n```\r\nP.S. This is my first experience with open source. So do not judge strictly if I do not understand something)", "@dantema There is indeed a typo in variable names. Nevertheless, I'm sorry if I was not clear but the output is from `mypy` type checker. You can run the code snippet without issues. The problem is with the type checking.", "However, I found out that the type annotation is actually misleading. The [`from_parquet`](https://github.com/huggingface/datasets/blob/5ef1ab1cc06c2b7a574bf2df454cd9fcb071ccb2/src/datasets/arrow_dataset.py#L1039) method should also accept list of [`PathLike`](https://github.com/huggingface/datasets/blob/main/src/datasets/utils/typing.py#L8) objects which includes [`os.PathLike`](https://docs.python.org/3/library/os.html#os.PathLike). But if I would ran the code snippet below, an exception is thrown.\r\n\r\n**Code**\r\n```py\r\nfrom pathlib import Path\r\n\r\nlist_of_filenames = [Path(\"foo.parquet\"), Path(\"bar.parquet\")]\r\nds = Dataset.from_parquet(list_of_filenames)\r\n```\r\n**Output**\r\n```py\r\n[/usr/local/lib/python3.8/dist-packages/datasets/arrow_dataset.py](https://localhost:8080/#) in from_parquet(path_or_paths, split, features, cache_dir, keep_in_memory, columns, **kwargs)\r\n 1071 from .io.parquet import ParquetDatasetReader\r\n 1072 \r\n-> 1073 return ParquetDatasetReader(\r\n 1074 path_or_paths,\r\n 1075 split=split,\r\n\r\n[/usr/local/lib/python3.8/dist-packages/datasets/io/parquet.py](https://localhost:8080/#) in __init__(self, path_or_paths, split, features, cache_dir, keep_in_memory, streaming, **kwargs)\r\n 35 path_or_paths = path_or_paths if isinstance(path_or_paths, dict) else {self.split: path_or_paths}\r\n 36 hash = _PACKAGED_DATASETS_MODULES[\"parquet\"][1]\r\n---> 37 self.builder = Parquet(\r\n 38 cache_dir=cache_dir,\r\n 39 data_files=path_or_paths,\r\n\r\n[/usr/local/lib/python3.8/dist-packages/datasets/builder.py](https://localhost:8080/#) in __init__(self, cache_dir, config_name, hash, base_path, info, features, use_auth_token, repo_id, data_files, data_dir, name, **config_kwargs)\r\n 298 \r\n 299 if data_files is not None and not isinstance(data_files, DataFilesDict):\r\n--> 300 data_files = DataFilesDict.from_local_or_remote(\r\n 301 sanitize_patterns(data_files), base_path=base_path, use_auth_token=use_auth_token\r\n 302 )\r\n\r\n[/usr/local/lib/python3.8/dist-packages/datasets/data_files.py](https://localhost:8080/#) in from_local_or_remote(cls, patterns, base_path, allowed_extensions, use_auth_token)\r\n 794 for key, patterns_for_key in patterns.items():\r\n 795 out[key] = (\r\n--> 796 DataFilesList.from_local_or_remote(\r\n 797 patterns_for_key,\r\n 798 base_path=base_path,\r\n\r\n[/usr/local/lib/python3.8/dist-packages/datasets/data_files.py](https://localhost:8080/#) in from_local_or_remote(cls, patterns, base_path, allowed_extensions, use_auth_token)\r\n 762 ) -> \"DataFilesList\":\r\n 763 base_path = base_path if base_path is not None else str(Path().resolve())\r\n--> 764 data_files = resolve_patterns_locally_or_by_urls(base_path, patterns, allowed_extensions)\r\n 765 origin_metadata = _get_origin_metadata_locally_or_by_urls(data_files, use_auth_token=use_auth_token)\r\n 766 return cls(data_files, origin_metadata)\r\n\r\n[/usr/local/lib/python3.8/dist-packages/datasets/data_files.py](https://localhost:8080/#) in resolve_patterns_locally_or_by_urls(base_path, patterns, allowed_extensions)\r\n 357 data_files = []\r\n 358 for pattern in patterns:\r\n--> 359 if is_remote_url(pattern):\r\n 360 data_files.append(Url(pattern))\r\n 361 else:\r\n\r\n[/usr/local/lib/python3.8/dist-packages/datasets/utils/file_utils.py](https://localhost:8080/#) in is_remote_url(url_or_filename)\r\n 62 \r\n 63 def is_remote_url(url_or_filename: str) -> bool:\r\n---> 64 parsed = urlparse(url_or_filename)\r\n 65 return parsed.scheme in (\"http\", \"https\", \"s3\", \"gs\", \"hdfs\", \"ftp\")\r\n 66 \r\n\r\n[/usr/lib/python3.8/urllib/parse.py](https://localhost:8080/#) in urlparse(url, scheme, allow_fragments)\r\n 373 Note that we don't break the components up in smaller bits\r\n 374 (e.g. netloc is a single string) and we don't expand % escapes.\"\"\"\r\n--> 375 url, scheme, _coerce_result = _coerce_args(url, scheme)\r\n 376 splitresult = urlsplit(url, scheme, allow_fragments)\r\n 377 scheme, netloc, url, query, fragment = splitresult\r\n\r\n[/usr/lib/python3.8/urllib/parse.py](https://localhost:8080/#) in _coerce_args(*args)\r\n 125 if str_input:\r\n 126 return args + (_noop,)\r\n--> 127 return _decode_args(args) + (_encode_result,)\r\n 128 \r\n 129 # Result objects are more helpful than simple tuples\r\n\r\n[/usr/lib/python3.8/urllib/parse.py](https://localhost:8080/#) in _decode_args(args, encoding, errors)\r\n 109 def _decode_args(args, encoding=_implicit_encoding,\r\n 110 errors=_implicit_errors):\r\n--> 111 return tuple(x.decode(encoding, errors) if x else '' for x in args)\r\n 112 \r\n 113 def _coerce_args(*args):\r\n\r\n[/usr/lib/python3.8/urllib/parse.py](https://localhost:8080/#) in <genexpr>(.0)\r\n 109 def _decode_args(args, encoding=_implicit_encoding,\r\n 110 errors=_implicit_errors):\r\n--> 111 return tuple(x.decode(encoding, errors) if x else '' for x in args)\r\n 112 \r\n 113 def _coerce_args(*args):\r\n\r\nAttributeError: 'PosixPath' object has no attribute 'decode'\r\n```\r\n\r\n@mariosasko Should I create a new issue? ", "@mariosasko I would like to take this issue up. ", "@avinashsai Hi, I've assigned you the issue.\r\n\r\n@tranhd95 Yes, feel free to report this in a new issue.", "@avinashsai Are you still working on this? If not I would like to give it a try." ]

[]

[]

[]

[ "Closing this one in favor of #670 \r\n\r\nThanks again for reporting the issue and proposing this fix !\r\nLet me know if you have other remarks" ]

[]

[ "The checks failed again even if I didn't make any changes.", "you just need to rebase from master to fix the CI :)", "Sorry for the mess, I'm confused by the rebase and thus created a new branch." ]

[]

[ "\r\n> Also, maybe calling the dataset as \"bnl_historical_newspapers\" and setting \"processed\" as one configuration name?\r\n\r\nThis sounds like a good idea but my only question around this is how easy it would be to use the same approach for processing the other newspaper collections [https://data.bnl.lu/data/historical-newspapers/](). \r\n\r\nFor example, the \"BIG DATA PACK\" is `257GB` of ALTO XML. This format is slightly more annoying to process because the metadata and text are contained in different files but the bigger issue might be that processing this XML using the Python XML libraries will probably be quite slow? I had thought for those larger datasets it might be more appropriate to use the Beam datasets? I don't have any experience using Beam so I'm not sure what that would involve and there is a reason to not include it in a dataset script alongside a non Beam dataset? \r\n\r\nIf there isn't an issue with potentially later adding other datasets (which may require Beam) into the same script I'll add one config for the processed version now which leaves open the option for later adding the other datasets. If this makes sense I'll also change the name as you suggest. \r\n\r\nThere is another dataset that could be a good candidate for inclusion here is the \"Monograph Text pack\" which is also processed into a simpler XML format however as the name suggests this isn't newspapers so might be confusing to include under a 'newspapers' script. One option would be to put everything under a `BNL` collection but it might be better to keep the monographs separate if they are added as a dataset so a single script doesn't end up including too much variety of content types? \r\n\r\n\r\n\r\n", "> My initial idea was to contribute the script also as \"community\" datasets (instead of canonical), i.e. in this case, pushing the script to the repo [huggingface.co/datasets/bigscience-catalogue-data/bnl_historical_newspapers](https://huggingface.co/datasets/bigscience-catalogue-data/bnl_historical_newspapers)\r\n\r\nSorry to respond to this late - happy for this to go in the community datasets. I think it would be nice to include in the canonical datasets at some point but since there is less urgency with this I could try and first work on improving the Datacard before doing that (i.e. make this a draft PR) - let me know if you think that makes more sense? \r\n\r\n\r\n", "> My initial idea was to contribute the script also as \"community\" datasets (instead of canonical), i.e. in this case, pushing the script to the repo https://huggingface.co/datasets/bigscience-catalogue-data/bnl_historical_newspapers\r\n> One of the advantages is that no dummy data is required, so the addition can be made faster\r\n> On the other hand, one disadvantage is that contributions cannot be made through PRs\r\n> Therefore, we should use the Issue page for discussions, reviews, decisions,...\r\n\r\nSure we can use the issues to discuss/review community datasets. Maybe let's have an issue template for that ?\r\nFor this dataset in particular I'll let @albertvillanova decide whether it's best as community dataset or not. IMO both are fine :)\r\n\r\n> I had thought for those larger datasets it might be more appropriate to use the Beam datasets? I don't have any experience using Beam so I'm not sure what that would involve and there is a reason to not include it in a dataset script alongside a non Beam dataset?\r\n\r\nBeam is nice to process a dataset once and for all and store the resulting processed data on the Hugging Face Hub or elsewhere. However for big datasets it must run on a distributed processing runtime like Google DataFlow, which is often inconvenient for many users. We've been using it though for datasets like Wikipedia and sharing the processed data in a GCS bucket.\r\n\r\nSo feel free to use the tools you like to process the datasets, but in the end I think we just need to host the processed data in a convenient format on the Hugging Face Hub to share it with the community. The processing script you used can also be shared with the community for reproducibility and documentation. But maybe @albertvillanova already has something in mind", "> > My initial idea was to contribute the script also as \"community\" datasets (instead of canonical), i.e. in this case, pushing the script to the repo [huggingface.co/datasets/bigscience-catalogue-data/bnl_historical_newspapers](https://huggingface.co/datasets/bigscience-catalogue-data/bnl_historical_newspapers)\r\n> > One of the advantages is that no dummy data is required, so the addition can be made faster\r\n> > On the other hand, one disadvantage is that contributions cannot be made through PRs\r\n> > Therefore, we should use the Issue page for discussions, reviews, decisions,...\r\n> \r\n> Sure we can use the issues to discuss/review community datasets. Maybe let's have an issue template for that ? For this dataset in particular I'll let @albertvillanova decide whether it's best as community dataset or not. IMO both are fine :)\r\n\r\nThanks, I'll hold off and let @albertvillanova decide best place for this. \r\n\r\n> > I had thought for those larger datasets it might be more appropriate to use the Beam datasets? I don't have any experience using Beam so I'm not sure what that would involve and there is a reason to not include it in a dataset script alongside a non Beam dataset?\r\n> \r\n> Beam is nice to process a dataset once and for all and store the resulting processed data on the Hugging Face Hub or elsewhere. However for big datasets it must run on a distributed processing runtime like Google DataFlow, which is often inconvenient for many users. We've been using it though for datasets like Wikipedia and sharing the processed data in a GCS bucket.\r\n> \r\n> So feel free to use the tools you like to process the datasets, but in the end I think we just need to host the processed data in a convenient format on the Hugging Face Hub to share it with the community. The processing script you used can also be shared with the community for reproducibility and documentation. But maybe @albertvillanova already has something in mind\r\n\r\nThat's useful, my own 2 cents are that it would make sense to do as @albertvillanova suggested and:-\r\n\r\n- rename the dataset to 'bnl_newspapers' \r\n- make the 'processed dataset' a config \r\n\r\nI won't try and include all the other datasets now but this leaves open the option of adding those later. The actual ALTO processing should be okay to do but I think it makes sense to do this as a one-off process and make the plain text + some associated metadata available elsewere so the dataset script can be kept simple and the processing doesn't get done multiple times. \r\n\r\n@albertvillanova if that sounds okay I'll update pull request to include those changes. \r\n", "@albertvillanova I've now created a config (currently with only one option) and renamed the dataset. This should keep the option to add other configs based on different bnl newspapers in the future. \r\n", "@mariosasko thanks for those suggestions ", "I just merged `master` into your branch to fix the CI :)", "@albertvillanova do you have additional comments ? Otherwise I think this PR is ready to merge :)", "> @davanstrien you did an awsome job!!! Thanks a lot!\r\n> \r\n> Just some very minor comments (mainly about the README documentation), and we merge this to master!\r\n\r\nThanks! Hopefully all addressed now. Thanks again for all the support with this pull request! " ]

[]

[]

[]

[ "Hi! The times match when I run this code locally or on Colab.\r\n\r\nAlso, we use `multiprocess`, not `multiprocessing`, for parallelization, and torch's `__init__.py` (executed on `import torch` ) slightly modifies the latter.", "Hey Mariosasko,\r\n\r\nThanks for looking into it. We further did some investigations after your comment and figured out it's only affecting some hardware/software configurations with the `pytorch` installation of `conda-forge`. Based on this we found the following issue in PyTorch: https://github.com/pytorch/pytorch/issues/102269 with a quick fix for now.\r\n\r\nSince it seems to be a deeper issue with forking processes, the difference between`multiprocess` and `multiprocessing` didn't make a difference.\r\n\r\nClosing this, since the issue comes from `pytorch` not `dataset`. \r\n" ]

[ { "color": "d73a4a", "default": true, "description": "Something isn't working", "id": 1935892857, "name": "bug", "node_id": "MDU6TGFiZWwxOTM1ODkyODU3", "url": "https://api.github.com/repos/huggingface/datasets/labels/bug" } ]

[ "The hash used for caching is computed by pickling recursively the function passed to `map`. Maybe some objects don't have the same hash across sessions. In particular you can check the hash of your model using\r\n```python\r\nfrom datasets.fingerprint import Hasher\r\nobj = AutoModel.from_config(config=config, add_pooling_layer=False)\r\nprint(Hasher.hash(obj))\r\n```\r\n\r\nYou can find mode info here: https://huggingface.co/docs/datasets/about_cache\r\n\r\nYou can also provide your own unique hash in `map` if you want, with the `new_fingerprint` argument", "Indeed, the hash is changing. The `dumps` function serialize the model object in different ways because the model object is not deterministic\r\n```python\r\nfrom datasets.utils.py_utils import dumps\r\nobj1 = AutoModel.from_config(config=config, add_pooling_layer=False)\r\nobj2 = AutoModel.from_config(config=config, add_pooling_layer=False)\r\n\r\ndumps(bert) == dumps(bert2). # False\r\n```\r\n\r\n> You can find mode info here: https://huggingface.co/docs/datasets/about_cache\r\n> \r\n> You can also provide your own unique hash in map if you want, with the new_fingerprint argument\r\n\r\n\r\nThanks, the doc is so helpful. Indeed, we can fix the hash and get cache hit using `new_fingerprint`. Closing the issue." ]

[ { "color": "d73a4a", "default": true, "description": "Something isn't working", "id": 1935892857, "name": "bug", "node_id": "MDU6TGFiZWwxOTM1ODkyODU3", "url": "https://api.github.com/repos/huggingface/datasets/labels/bug" } ]

[ "Hi,\r\n\r\nthis issue is similar to https://github.com/huggingface/datasets/issues/3093, so you can either use the solution provided there or try to load the data in one chunk (you can control the chunk size by specifying the `chunksize` parameter (`int`) in `load_dataset`).\r\n\r\n@lhoestq Is this worth opening an issue on Jira? Basically, PyArrow doesn't allow casts that change the order of the struct fields because they treat `pa.struct` as an ordered sequence. Reordering fields manually in Python is probably too slow, so I think this needs to be fixed by them to be usable on our side.", "I agree I would expect PyArrow to be able to handle this, do you want to open the issue @mariosasko ?\r\nAlthough maybe it's possible to fix struct casting on our side without hurting performance too much, if it's simply a matter of reordering the arrays in the StructArray", "Fixed in #3575, so I'm closing this issue." ]

[]

[ "Hi @lhoestq, thanks for the review. I will have a look and update it accordingly.", "Strange error message :-)\r\n\r\n```\r\n> tf_context = tf.python.context.context() # eager mode context\r\nE AttributeError: module 'tensorflow' has no attribute 'python'\r\n```\r\n" ]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_3868). All of your documentation changes will be reflected on that endpoint.", "Cool thanks ! Could you add a test please ?" ]

[]

[ "@elronbandel it looks like something went wrong with the renaming, as the old files are still in the PR. Can you `git rm datasets/hebrew-sentiment` ?", "merging since the CI is fixed on master", "This is the old version of the data.\r\nHere is the fixed version.\r\nhttps://github.com/OnlpLab/Hebrew-Sentiment-Data\r\n\r\nI hope I would find time to open a PR. I think it supposed to be only to change the data path ", "Cool ! Sure feel free to open a PR if you have some time :) and feel free to ping me for review or if you have questions" ]

[]

[ "Hi @kradonneoh, thanks for reporting.\r\n\r\nPlease note that to work with audio datasets (and specifically with MP3 files) we have detailed installation instructions in our docs: https://huggingface.co/docs/datasets/installation#audio\r\n- one of the requirements is torchaudio<0.12.0\r\n\r\nLet us know if the problem persists after having followed them.", "I saw that and have followed it (hence the Expected Behavior section of the bug report). \r\n\r\nIs there no intention of updating to the latest version? It does limit the version of `torch` I can use, which isn’t ideal.", "@kradonneoh hey! actually with `ffmpeg4` loading of mp3 files should work, so this is a not expected behavior and we need to investigate it. It works on my side with `torchaudio==0.13` and `ffmpeg==4.2.7`. Which `torchaudio` version do you use?\r\n\r\n`datasets` should support decoding of mp3 files with `torchaudio` when its version is `>0.12` but as you noted it requires `ffmpeg>4`, we need to fix this in the documentation, thank you for pointing to this! \r\n\r\nBut according to your traceback it seems that it tries to use [`libsndfile`](https://github.com/libsndfile/libsndfile) backend for mp3 decoding. And `libsndfile` library supports mp3 decoding starting from version 1.1.0 which on Linux has to be compiled from source for now afaik. \r\n\r\nfyi - we are aiming at getting rid of `torchaudio` dependency at all by the next major library release in favor of `libsndfile` too.", "We now decode MP3 with `soundfile`, so I'm closing this issue" ]

[]

[]

[]

[]

[]

[ "I believe we don't use PyArrow >= 4.0.0 because of some segfault issues:\r\nhttps://github.com/huggingface/datasets/blob/1206ffbcd42dda415f6bfb3d5040708f50413c93/setup.py#L78\r\nCan you confirm @lhoestq ?", "@SBrandeis pyarrow version 4.0.1 has fixed that issue: #2489 😉 " ]

[]

[ "Solved!" ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "I thought of just mentioning the benefits I got. Here's the code that @lhoestq provided:\r\n\r\n```py\r\nimport os\r\nfrom datasets import load_dataset\r\nfrom tqdm.auto import tqdm\r\n\r\nds = load_dataset(\"squad\", split=\"train\")\r\nos.makedirs(\"tmp\")\r\n\r\nnum_shards = 5\r\nfor index in tqdm(range(num_shards)):\r\n size = len(ds) // num_shards\r\n shard = Dataset(ds.data.slice(size * index, size), fingerprint=f\"{ds._fingerprint}_{index}\")\r\n shard.to_json(f\"tmp/data_{index}.jsonl\")\r\n```\r\n\r\nIt is 1.64s. Previously the code was:\r\n\r\n```py\r\nnum_shards = 5\r\nfor index in tqdm(range(num_shards)):\r\n shard = ds.shard(num_shards=num_shards, index=index, contiguous=True)\r\n shard.to_json(f\"tmp/data_{index}.jsonl\")\r\n # upload_to_gcs(f\"tmp/data_{index}.jsonl\")\r\n```\r\n\r\nIt was 2min31s. \r\n\r\nI ran it on my humble MacBook Pro:\r\n\r\n<img width=\"574\" alt=\"image\" src=\"https://user-images.githubusercontent.com/22957388/172864881-f1db489a-2305-47f2-a07f-7d3df610b1b8.png\">\r\n", "I addressed your comments @albertvillanova , let me know what you think :)" ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._" ]

[ { "color": "a2eeef", "default": true, "description": "New feature or request", "id": 1935892871, "name": "enhancement", "node_id": "MDU6TGFiZWwxOTM1ODkyODcx", "url": "https://api.github.com/repos/huggingface/datasets/labels/enhancement" } ]

[ "Hi !\r\nIf we want something more general, we could either\r\n1. delete the extracted files after the arrow data generation automatically, or \r\n2. delete each extracted file during the arrow generation right after it has been closed.\r\n\r\nSolution 2 is better to save disk space during the arrow generation. Is it what you had in mind ?\r\n\r\nThe API could look like\r\n```python\r\nload_dataset(..., delete_extracted_files_after_usage=True)\r\n```\r\n\r\nIn terms of implementation, here are some directions we could take for each solution:\r\n1. get the list of the extracted files from the DownloadManager and then delete them after the dataset is processed. This can be implemented in `download_and_prepare` I guess\r\n2. maybe wrap and mock `open` in the builder to make it delete the file when the file is closed.", "Also, if I delete the extracted files they need to be re-extracted again instead of loading from the Arrow cache files", "I think we already opened an issue about this topic (suggested by @stas00): duplicated of #2481?\r\n\r\nThis is in our TODO list... 😅 ", "I think the deletion of each extracted file could be implemented in our CacheManager and ExtractManager (once merged to master: #2295, #2277). 😉 ", "Oh yes sorry, I didn't check if this was a duplicate", "Nevermind @thomwolf, I just mentioned the other issue so that both appear linked in GitHub and we do not forget to close both once we make the corresponding Pull Request... That was the main reason! 😄 ", "Ok yes. I think this is an important feature to be able to use large datasets which are pretty much always compressed files.\r\n\r\nIn particular now this requires to keep the extracted file on the drive if you want to avoid reprocessing the dataset so in my case, this require using always ~400GB of drive instead of just 200GB (which is already significant). \r\n\r\nTwo nice features would be to:\r\n- allow to delete the extracted files without loosing the ability to load the dataset from the cached arrow-file\r\n- streamlined decompression when only the currently read file is extracted - this might require to read the list of files from the extracted archives before processing them?", "Here is a sample dataset with 2 such large compressed JSON files for debugging: https://huggingface.co/datasets/thomwolf/github-python", "Note that I'm confirming that with the current master branch of dataset, deleting extracted files (without deleting the arrow cache file) lead to **re-extracting** these files when reloading the dataset instead of directly loading the arrow cache file.", "Hi ! That's weird, it doesn't do that on my side (tested on master on my laptop by deleting the `extracted` folder in the download cache directory). You tested with one of the files at https://huggingface.co/datasets/thomwolf/github-python that you have locally ?", "Yes it’s when I load local compressed JSON line files with load_dataset(‘json’, data_files=…) ", "@thomwolf I'm sorry but I can't reproduce this problem. I'm also using: \r\n```python\r\nds = load_dataset(\"json\", split=\"train\", data_files=data_files, cache_dir=cache_dir)\r\n```\r\nafter having removed the extracted files:\r\n```python\r\nassert sorted((cache_dir / \"downloads\" / \"extracted\").iterdir()) == []\r\n```\r\n\r\nI get the logging message:\r\n```shell\r\nWARNING datasets.builder:builder.py:531 Reusing dataset json ...\r\n```", "Do you confirm the extracted folder stays empty after reloading?", "> \r\n> \r\n> Do you confirm the extracted folder stays empty after reloading?\r\n\r\nYes, I have the above mentioned assertion on the emptiness of the extracted folder:\r\n```python\r\nassert sorted((cache_dir / \"downloads\" / \"extracted\").iterdir()) == []\r\n```\r\n" ]

[ { "color": "E5583E", "default": false, "description": "Related to the dataset viewer on huggingface.co", "id": 3470211881, "name": "dataset-viewer", "node_id": "LA_kwDODunzps7O1zsp", "url": "https://api.github.com/repos/huggingface/datasets/labels/dataset-viewer" } ]

[ "It's a bug! Thanks for reporting, I'm looking at it", "Fixed." ]

[]

[ "Hi! Let's not do this - renaming it would be a breaking change, and going through the deprecation cycle is only worth it if it improves user experience.", "Hi @mariosasko, ok it makes sense. Anyway, don't you think it's worth it at some point to start a deprecation cycle e.g. `fs` in `load_from_disk`? It doesn't affect user experience but it's for sure a bad practice IMO, but's up to you 😄 Feel free to close this issue otherwise!", "I don't think deprecating a param name in this particular instance is worth the hassle, so I'm closing the issue 🙂.", "Sure, makes sense @mariosasko thanks!" ]

[ { "color": "a2eeef", "default": true, "description": "New feature or request", "id": 1935892871, "name": "enhancement", "node_id": "MDU6TGFiZWwxOTM1ODkyODcx", "url": "https://api.github.com/repos/huggingface/datasets/labels/enhancement" }, { "color": "BDE59C", "default": false, "description": "Issues a bit more difficult than \"Good First\" issues", "id": 3761482852, "name": "good second issue", "node_id": "LA_kwDODunzps7gM6xk", "url": "https://api.github.com/repos/huggingface/datasets/labels/good%20second%20issue" } ]

[]

[]

[ "merging since the CI is fixed on master" ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._" ]

[]

[]

[]

[]

[]