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" } ]

[ "I addressed your comments about the docstrings and the output validation :)", "I updated the bleurt mocking method and bleurt test is passing now.\r\nI also ran the slow tests and they are passing for bleurt.", "Thanks @lhoestq and @albertvillanova !" ]

[]

[]

[]

[]

[]

[ "Hi !\r\nThanks for reporting.\r\nIt looks like no object inherits from `datasets.GeneratorBasedBuilder` (or more generally from `datasets.DatasetBuilder`) in your script.\r\n\r\nCould you check that there exist at least one dataset builder class ?", "Hi @xixiaoyao did you manage to fix your issue ?", "No activity, closing", "It happened when try to change the old project which use 'nlp' to new project which use 'datasets'. You should check you old 'my_squad.py' file, change the inherit class from `nlp.xxx` to `datasets.xxx`. Otherwise datasets - load.py - import_main_class() `if inspect.isclass(obj) and issubclass(obj, main_cls_type):` can not find the main_cls." ]

[ { "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" } ]

[]

[ { "color": "fef2c0", "default": false, "description": "", "id": 3287858981, "name": "streaming", "node_id": "MDU6TGFiZWwzMjg3ODU4OTgx", "url": "https://api.github.com/repos/huggingface/datasets/labels/streaming" } ]

[ "Thanks for reporting @NitishkKarra.\r\n\r\nThe root source of the issue is that streaming mode is not supported out-of-the-box for that dataset, because it contains a TAR file.\r\n\r\nWe have opened a discussion in the corresponding Hub dataset page, pointing out this issue: https://huggingface.co/datasets/Theivaprakasham/wildreceipt/discussions/1\r\n\r\nI'm closing this issue here, so this discussion is transferred there instead." ]

[]

[ "Thanks @RachelKer for this PR.\r\n\r\nI think the dummy_data structure does not also match. In the `_split_generator` you have something like `os.path.join(downloaded_files[\"validation\"], lang+'_val.jsonl')` but in you dummy_data you have `os.path.join(downloaded_files[\"validation\"], lang+\"_val.zip\", lang+'_val.jsonl')`. I think ` jsonl` files should be directly in the `dummy_data` folder without the sub-folder \r\n\r\n@lhoestq ", "Hi @RachelKer :)\r\nThanks for adding MLSUM !\r\n\r\nTo fix the CI I think you just have to rebase from master", "Great, I think it is working now. Thanks :)", "It looks like your PR does tons of changes in other datasets. \r\nMaybe this is because of the merge from master ?", "Hmm, I see, sorry I messed up somewhere. Maybe it's easier if we close the pull request and I do another one ?", "Yea if it's easier for you feel free to re-open a PR" ]

[ { "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 @AnzorGozalishvili,\r\n\r\nMaybe their site was temporarily down, but it seems to work fine now.\r\n\r\nCould you please try again and confirm if the problem persists? ", "Hi @albertvillanova \r\nI checked and it works. \r\nIt seems that it was really temporarily down.\r\nThanks!" ]

[]

[ "Hi ! Thanks for the fix :)\r\nThe CI fail isn't related to your PR. I opened a PR #2286 to fix the CI.\r\nWe'll wait for #2286 to be merged to master first if you don't mind", "The PR has been merged ! Feel free to merge master into your branch to fix the CI" ]

[ { "color": "0e8a16", "default": false, "description": "Contribution to a dataset script", "id": 4564477500, "name": "dataset contribution", "node_id": "LA_kwDODunzps8AAAABEBBmPA", "url": "https://api.github.com/repos/huggingface/datasets/labels/dataset%20contribution" } ]

[ "_The documentation is not available anymore as the PR was closed or merged._", "Thanks @lhoestq, I had missed that... ", "thx for the super fast work @albertvillanova ! any estimate for when the relevant release will happen?\r\n\r\nThanks again ", "@thesofakillers after a recent change in our library (see #4059), now fixes in all datasets are immediately accessible. You can try it:\r\n```python\r\nfrench = datasets.load_dataset(\"xcsr\", \"X-CSQA-fr\")\r\n```\r\n\r\nPlease note there is an additional fix to that dataset in progress (to be merged today):\r\n- #5024" ]

[]

[]

[]

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

[]

[ "Hi ! It fails because the dataset is already being prepared by your first run. I'd encourage you to prepare your dataset before using it for multiple trainings.\r\n\r\nYou can also specify another cache directory by passing `cache_dir=` to `load_dataset()`.", "Thank you! What do you mean by prepare it beforehand? I am unclear how to conduct dataset preparation outside of using the `load_dataset` function.", "You can have a separate script that does load_dataset + map + save_to_disk to save your prepared dataset somewhere. Then in your training script you can reload the dataset with load_from_disk", "Thank you! I believe I was running additional map steps after loading, resulting in the cache conflict. " ]

[]

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

[]

[]

[ { "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" } ]

[ "This should be fixed with #645 ", "Feel free to re-open if it still occurs" ]

[ { "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" }, { "color": "d93f0b", "default": false, "description": "", "id": 2725241052, "name": "speech", "node_id": "MDU6TGFiZWwyNzI1MjQxMDUy", "url": "https://api.github.com/repos/huggingface/datasets/labels/speech" } ]

[ "I have started working on adding this dataset.", "Hey @BirgerMoell - awesome that you started working on Common Voice. Common Voice is a bit special since, there is no direct download link to download the data. In these cases we usually consider two options:\r\n\r\n1) Find a hacky solution to extract the download link somehow from the XLM tree of the website \r\n2) If this doesn't work we force the user to download the data himself and add a `\"data_dir\"` as an input parameter. E.g. you can take a look at how it is done for [this](https://github.com/huggingface/datasets/blob/66f2a7eece98d2778bd22bb5034cb7c2376032d4/datasets/arxiv_dataset/arxiv_dataset.py#L66) \r\n\r\nAlso the documentation here: https://huggingface.co/docs/datasets/add_dataset.html?highlight=data_dir#downloading-data-files-and-organizing-splits (especially the \"note\") might be helpful.", "Let me know if you have any other questions", "I added a Work in Progress pull request (hope that is ok). I've made a card for the dataset and filled out the common_voice.py file with information about the datset (not completely).\r\n\r\nI didn't manage to get the tagging tool working locally on my machine but will look into that later.\r\n\r\nLeft to do.\r\n\r\n- Tag the dataset\r\n- Add missing information and update common_voice.py\r\n\r\nhttps://github.com/huggingface/datasets/pull/1886", "Awesome! I left a longer comment on the PR :-)", "I saw that this current datasets package holds common voice version 6.1, how to add the new version 7.0 that is already available?", "Will me merged next week - we're working on it :-)", "Common voice still appears to be a 6.1. Is the plan still to upgrade to 7.0?", "We actually already have the code and everything ready to add Common Voice 7.0 to `datasets` but are still waiting for the common voice authors to give us the green light :-) \r\n\r\nAlso gently pinging @phirework and @milupo here", "Common Voice 7.0 is available here now: https://huggingface.co/datasets/mozilla-foundation/common_voice_7_0", "For anyone else stumbling upon this thread, the 8.0 version is also available now: https://huggingface.co/datasets/mozilla-foundation/common_voice_8_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! \r\n\r\nYes, this makes sense for numeric values, but first I have to finish https://github.com/huggingface/datasets/pull/3336 because currently ArrayXD only allows the first dimension to be dynamic." ]

[]

[]

[]

[]

[]

[ "It looks like this PR now includes changes to many other files than the ones for CoNLLpp.\r\n\r\nTo fix that feel free to create another branch and another PR.\r\n\r\nThis was probably caused by a git rebase. You can avoid this issue by using git merge if you've already pushed your branch." ]

[]

[ "The script uses `Dataset.load_from_disk`, which as you can expect, doesn't work in streaming mode.\r\n\r\nIt would probably be more practical to load the dataset locally using `Dataset.load_from_disk` first and then `push_to_hub` to upload it in Parquet on the Hub", "I've transferred this issue to the Hub repo: https://huggingface.co/datasets/Jean-Baptiste/wikiner_fr/discussions/3\r\n\r\nI'm closing this." ]

[]

[]

[]

[]

[]

[]

[ { "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" } ]

[]

[]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_5001). All of your documentation changes will be reflected on that endpoint.", "> CC: @davanstrien\r\n\r\nI should have some time to look at this on Friday :) ", "@albertvillanova I've tried this with a few different XML datasets. One issue I've run into is getting a `KeyError` when the attributes of a field differ from the first parsed row. Unfortunately, this can come up in the ALTO XML format, for example, if you want to parse the 'string' field, which contains the text in the ALTO XML files. \r\n\r\nWhen parsing a file, this instance has no 'STYLE' attribute: \r\n\r\n```xml\r\n<TextLine HEIGHT=\"39\" WIDTH=\"295\" VPOS=\"926\" HPOS=\"247\"><String WC=\"0.4600000083\" CONTENT=\"jufqu’en\" HEIGHT=\"39\" WIDTH=\"117\" VPOS=\"926\" HPOS=\"247\"/><SP WIDTH=\"14\" VPOS=\"928\" HPOS=\"365\"/><String WC=\"0.6075000167\" CONTENT=\"l’an\" HEIGHT=\"26\" WIDTH=\"50\" VPOS=\"928\" HPOS=\"380\"/><SP WIDTH=\"24\" VPOS=\"936\" HPOS=\"431\"/><String WC=\"0.4300000072\" CONTENT=\"1\" HEIGHT=\"16\" WIDTH=\"9\" VPOS=\"936\" HPOS=\"456\"/><String STYLE=\"italics\" WC=\"0.5774999857\" CONTENT=\"361.\" HEIGHT=\"25\" WIDTH=\"68\" VPOS=\"933\" HPOS=\"474\"/></TextLine>\r\n```\r\n\r\nWhereas this one which appears later in the file, does have this field: \r\n\r\n```xml\r\n<TextLine HEIGHT=\"39\" WIDTH=\"712\" VPOS=\"966\" HPOS=\"297\"><String STYLE=\"italics\" WC=\"0.6999999881\" CONTENT=\"I\" HEIGHT=\"17\" WIDTH=\"9\" VPOS=\"977\" HPOS=\"297\"/><String WC=\"0.5\" CONTENT=\"I.\" HEIGHT=\"18\" WIDTH=\"25\" VPOS=\"976\" HPOS=\"318\"/><SP WIDTH=\"24\" VPOS=\"971\" HPOS=\"344\"/><String STYLE=\"italics\" WC=\"0.3359999955\" CONTENT=\"Crade\" HEIGHT=\"26\" WIDTH=\"91\" VPOS=\"967\" HPOS=\"369\"/><SP WIDTH=\"31\" VPOS=\"971\" HPOS=\"461\"/><String STYLE=\"italics\" WC=\"0.6060000062\" CONTENT=\"Pétri\" HEIGHT=\"26\" WIDTH=\"71\" VPOS=\"968\" HPOS=\"493\"/><SP WIDTH=\"23\" VPOS=\"968\" HPOS=\"565\"/><String STYLE=\"italics\" WC=\"0.612857163\" CONTENT=\"Candidi\" HEIGHT=\"27\" WIDTH=\"111\" VPOS=\"967\" HPOS=\"589\"/><SP WIDTH=\"19\" VPOS=\"967\" HPOS=\"701\"/><String STYLE=\"italics\" WC=\"0.4088888764\" CONTENT=\"Decembrii\" HEIGHT=\"28\" WIDTH=\"144\" VPOS=\"966\" HPOS=\"721\"/><SP WIDTH=\"10\" VPOS=\"968\" HPOS=\"866\"/><String STYLE=\"italics\" WC=\"0.4600000083\" CONTENT=\"in\" HEIGHT=\"25\" WIDTH=\"27\" VPOS=\"968\" HPOS=\"877\"/><SP WIDTH=\"9\" VPOS=\"967\" HPOS=\"905\"/><String STYLE=\"italics\" WC=\"0.5099999905\" CONTENT=\"funere\" HEIGHT=\"38\" WIDTH=\"94\" VPOS=\"967\" HPOS=\"915\"/></TextLine>\r\n```\r\n\r\nSince the first-seen fields define what is passed to `arrow_writer`, this causes a KeyError when the version with the extra attributes is encountered because it doesn't expect this column. \r\n\r\nSince it's important to support streaming, I'm not sure there is a nice way to detect attributes for the whole file easily in an automatic way. The two potential ways I can see of doing it.\r\n\r\n- Do an initial pass on a batch of data to have a higher chance of encountering variations in attributes before doing the arrow write. \r\n- Do a full pass on one file (and assume that this won't change across files) \r\n\r\nI think the other way of doing this would be to allow users to define expected/wanted attributes as another loading argument. This could then be used to extract the described attributes (and make them None if not found). This requires a bit more work from the user but could be helpful. For example, in the XML above, likely, most users will only want the `WC` and `CONTENT` attributes. So they could specify this upfront and avoid loading extra data they don't need or want. I suspect this option would make more sense than making this operation automatic for the case where attributes might change. WDYT? \r\n\r\n\r\n\r\n\r\n\r\n\r\n" ]

[]

[ "@lhoestq thanks for the comments! Should be fixed in the latest commit, I assume the CI errors are unrelated." ]

[]

[ "Thanks!" ]

[ { "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 @BramVanroy, thanks for reporting.\r\n\r\nFirst, note that `\"min\"` is not an allowed value for `average`. According to scikit-learn [documentation](https://scikit-learn.org/stable/modules/generated/sklearn.metrics.f1_score.html), `average` can only take the values: `{\"micro\", \"macro\", \"samples\", \"weighted\", \"binary\"} or None, default=\"binary\"`.\r\n\r\nSecond, you should take into account that all additional metric-specific argument should be passed in the method `compute` (and not in the method `load_metric`). You can find more information in our documentation: https://huggingface.co/docs/datasets/using_metrics.html#computing-the-metric-scores\r\n\r\nSo for example, if you would like to calculate the macro-averaged F1 score, you should use:\r\n```python\r\nimport datasets\r\n\r\nf1 = datasets.load_metric(\"f1\", keep_in_memory=True)\r\nf1.add_batch(predictions=[0,2,3], references=[1, 2, 3])\r\nf1.compute(average=\"macro\")\r\n```", "Thanks, that was it. A bit strange though, since `load_metric` had an argument `metric_init_kwargs`. I assume that that's for specific initialisation arguments whereas `average` is for the function itself." ]

[ { "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" } ]

[ "Fixed, thanks\r\n<img width=\"661\" alt=\"Capture d’écran 2021-12-22 à 12 02 30\" src=\"https://user-images.githubusercontent.com/1676121/147082881-cf700e8d-0511-4431-b214-d6cf8137db10.png\">\r\n" ]

[]

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

[]

[]

[]

[ "Hi ! The huggingface/dataset-viewer code was not maintained anymore because we switched to a new dataset viewer that is deployed available for each dataset the Hugging Face website.\r\n\r\nWhat are you using this old repository for ?", "I think these parts are outdated:\r\n\r\n* https://github.com/huggingface/datasets-viewer/blob/8efad8eae313a891f713469983bf4c744786f26e/run.py#L126-L131\r\n* https://github.com/huggingface/datasets-viewer/blob/8efad8eae313a891f713469983bf4c744786f26e/run.py#L145-L150\r\n\r\nTo make the viewer work, the first one should be replaced with the following:\r\n```python\r\ndataset_module = datasets.load.dataset_module_factory(path)\r\nbuilder_cls = datasets.load.import_main_class(dataset_module.module_path)\r\nconfs = builder_cls.BUILDER_CONFIGS\r\n```\r\nAnd the second one:\r\n```python\r\ndataset_module = datasets.load.dataset_module_factory(path)\r\nbuilder_cls = datasets.load.import_main_class(dataset_module.module_path)\r\nif conf:\r\n builder_instance = builder_cls(name=conf, cache_dir=path if path_to_datasets is not None else None)\r\nelse:\r\n builder_instance = builder_cls(cache_dir=path if path_to_datasets is not None else None)\r\n```\r\n\r\nBut as @lhoestq suggested, it's better to use the `datasets-server` API nowadays to [fetch the rows](https://huggingface.co/docs/datasets-server/rows).", "> The dataset viewer on the Hugging Face website is incredibly useful\r\n\r\n@mariosasko i think @lilyorlilypad wants to run the new dataset-viewer, not the old one", "> wants to run the new dataset-viewer, not the old one\r\n\r\nThanks for the clarification for me. I do want to run the new dataset-viewer. ", "It should be possible to run it locally using the HF datasets-server API (docs [here](https://huggingface.co/docs/datasets-server)) but the front end part is not open source (yet ?)\r\n\r\nThe back-end is open source though if you're interested: https://github.com/huggingface/datasets-server\r\nIt automatically converts datasets on HF to Parquet, which is the format we use to power the viewer.", "the new frontend would probably be hard to open source, as is, as it's quite intertwined with the Hub's code.\r\n\r\nHowever, at some point it would be amazing to have a community-driven open source implementation of a frontend to datasets-server! " ]

[]

[]

[]

[ "Can you explain your use case for `max_shard_size`? \r\n\r\nOn some systems, there is a limit to the size of a memory-mapped file, so we could consider exposing this parameter in `load_dataset`.", "In my use case, users may choose a proper size to balance the cost and benefit of using large shard size. (On azure blob or hdfs which may automatically download the shard from background)", "But `load_dataset` doesn't support caching (and reading) Arrow datasets from remote storage. \r\n\r\n`load_datset_builder` + `download_and_prepare` is not equal to `load_dataset`. The latter has one more step, `builder.as_dataset`, that memory-maps Arrow files, which only works for local files.", "Thanks. So if I want to use `IterableDataset` and control the size of single arrow file, how should I organize the data loader? Maybe `load_dataset_build` + `download_and_prepare` + `builder.as_dataset` + `dataset.to_iterable_dataset`?", "Yes, this should work.\r\n\r\nI think we can expose `max_shard_size` in `load_dataset`, so feel free to open a PR." ]

[]

[ "cc @polinaeterna @lhoestq ", "@sanchit-gandhi can you please also post the logs of `pip install soundfile==0.12.1`? To check what wheel is being installed or if it's being built from source (I think it's the latter case). \r\nRequired `libsndfile` binary **should** be bundeled with `soundfile` wheel but I assume it **might not** be the case for some non standard Linux distributions. \r\nThe only solution for using `soundfile` here is to build [`libsndfile`](https://github.com/libsndfile/libsndfile) from source:\r\n\r\n```bash\r\ngit clone https://github.com/libsndfile/libsndfile.git\r\ncd libsndfile/\r\nautoreconf -vif\r\n./configure --enable-werror \r\nmake\r\nmake install\r\n```\r\nfor this, some building libraries should be installed, for Debian/Ubuntu it's like:\r\n```bash\r\napt install autoconf autogen automake build-essential libasound2-dev \\\r\n libflac-dev libogg-dev libtool libvorbis-dev libopus-dev libmp3lame-dev \\\r\n libmpg123-dev pkg-config python\r\n```\r\nbut for other Linux distributions it might be different.\r\n\r\nWhen the binary is compiled, it should be put into location where `soundfile` would search for it (the directory is named `_soundfile_data`), it depends on where`libsdfile` (from the previous step) and `soundfile` were installed, might be something like this:\r\n\r\n```bash\r\ncp /usr/local/lib/libsndfile.so /usr/local/lib/python3.7/dist-packages/_soundfile_data/\r\ncp /usr/local/lib/libsndfile.la /usr/local/lib/python3.7/dist-packages/_soundfile_data/\r\n```\r\n\r\nAnother solution is to not use `soundfile` and apply custom processing function with `torchaudio` while setting `decode=False` in `Audio` feature and passing custom function to `.map`. ", "Not sure if it may help, but you could also try updating `pip` before installing soundfile", "@lhoestq @sanchit-gandhi. I encountered the same error (also on the TPU v4) when trying to run `datasets` from source.\r\n\r\nDowngrading soundfile with `pip install soundfile==0.12.0` seems to fix the issue for me.", "Maybe let's open an issue at https://github.com/bastibe/python-soundfile/issues in case they might know why you get `OSError: cannot load library 'libsndfile.so'` ?", "> @sanchit-gandhi can you please also post the logs of `pip install soundfile==0.12.1`? To check what wheel is being installed or if it's being built from source (I think it's the latter case). Required `libsndfile` binary **should** be bundeled with `soundfile` wheel but I assume it **might not** be the case for some non standard Linux distributions. The only solution for using `soundfile` here is to build [`libsndfile`](https://github.com/libsndfile/libsndfile) from source:\r\n> \r\n> ```shell\r\n> git clone https://github.com/libsndfile/libsndfile.git\r\n> cd libsndfile/\r\n> autoreconf -vif\r\n> ./configure --enable-werror \r\n> make\r\n> make install\r\n> ```\r\n\r\nThis fixed the issue for me. After installing libsndfile as described above, I had to uninstall soundfile and re-install it with this command. `pip install \"soundfile>=0.12.1\"`", "Thank you so much for the comprehensive instructions @polinaeterna! Also confirming that they worked for me 🤗 In my case, I had to run several of these commands under \"sudo\" for privileges, but otherwise this workaround gave a successful `libsndfile` install:\r\n\r\n1. Grab source code:\r\n```\r\ngit clone https://github.com/libsndfile/libsndfile.git\r\n```\r\n\r\n2. Set up a build environment:\r\n```\r\nsudo apt install autoconf autogen automake build-essential libasound2-dev \\\r\n libflac-dev libogg-dev libtool libvorbis-dev libopus-dev libmp3lame-dev \\\r\n libmpg123-dev pkg-config python\r\n```\r\n\r\n3. Build and test `libsndfile`:\r\n\r\n```\r\nautoreconf -vif\r\n./configure --enable-werror\r\nsudo make\r\nsudo make check\r\n```\r\n\r\n4. Create `_soundfile_data` submodule (if it does not exist already):\r\n```\r\nsudo mkdir /usr/local/lib/python3.8/dist-packages/_soundfile_data/\r\n```\r\n\r\n5. Copy `libsndfile` files into submodule:\r\n```\r\nsudo cp /usr/local/lib/libsndfile.* /usr/local/lib/python3.8/dist-packages/_soundfile_data/\r\n```", "On a different machine, I also tried separately by first upgrading pip, then installing soundfile. This worked too! Thanks @lhoestq 🙌", "> @sanchit-gandhi can you please also post the logs of `pip install soundfile==0.12.1`? To check what wheel is being installed or if it's being built from source (I think it's the latter case). Required `libsndfile` binary **should** be bundeled with `soundfile` wheel but I assume it **might not** be the case for some non standard Linux distributions. The only solution for using `soundfile` here is to build [`libsndfile`](https://github.com/libsndfile/libsndfile) from source:\r\n> \r\n> ```shell\r\n> git clone https://github.com/libsndfile/libsndfile.git\r\n> cd libsndfile/\r\n> autoreconf -vif\r\n> ./configure --enable-werror \r\n> make\r\n> make install\r\n> ```\r\n> \r\n> for this, some building libraries should be installed, for Debian/Ubuntu it's like:\r\n> \r\n> ```shell\r\n> apt install autoconf autogen automake build-essential libasound2-dev \\\r\n> libflac-dev libogg-dev libtool libvorbis-dev libopus-dev libmp3lame-dev \\\r\n> libmpg123-dev pkg-config python\r\n> ```\r\n> \r\n> but for other Linux distributions it might be different.\r\n> \r\n> When the binary is compiled, it should be put into location where `soundfile` would search for it (the directory is named `_soundfile_data`), it depends on where`libsdfile` (from the previous step) and `soundfile` were installed, might be something like this:\r\n> \r\n> ```shell\r\n> cp /usr/local/lib/libsndfile.so /usr/local/lib/python3.7/dist-packages/_soundfile_data/\r\n> cp /usr/local/lib/libsndfile.la /usr/local/lib/python3.7/dist-packages/_soundfile_data/\r\n> ```\r\n> \r\n> Another solution is to not use `soundfile` and apply custom processing function with `torchaudio` while setting `decode=False` in `Audio` feature and passing custom function to `.map`.\r\n\r\nThanks, the solution solved my problem. \r\n\r\n1. Purge uninstall libsndfile, uninstall python-soundfile.\r\n2. Build libsndfile from source code and install.\r\n3. Build python-soundfile from source code and install\r\n4. Well done." ]

[ { "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" }, { "color": "cfd3d7", "default": true, "description": "This issue or pull request already exists", "id": 1935892865, "name": "duplicate", "node_id": "MDU6TGFiZWwxOTM1ODkyODY1", "url": "https://api.github.com/repos/huggingface/datasets/labels/duplicate" } ]

[ "Hi @Arij-Aladel, thanks for reporting.\r\n\r\nThis issue was already reported \r\n- #3784\r\n\r\nand its root cause is a change in the Google Drive service. See:\r\n- #3786 \r\n\r\nWe have already fixed it in our 2.0.0 release. See:\r\n- #3787 \r\n\r\nPlease, update your `datasets` version:\r\n```\r\npip install -U datasets\r\n```\r\nand retry loading the dataset by forcing its redownload:\r\n```python\r\ndataset = load_dataset(\"cnn_dailymail\", \"3.0.0\", download_mode=\"force_redownload\")\r\n```" ]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_6016). All of your documentation changes will be reflected on that endpoint.", "It we could have something similar to Polars, that would be great.\r\n\r\nThis is what Polars outputs: \r\n* `__repr__`/`__str__` :\r\n```\r\nshape: (67_349, 3)\r\n┌───────┬───────────────────────────────────┬───────┐\r\n│ idx ┆ sentence ┆ label │\r\n│ --- ┆ --- ┆ --- │\r\n│ i32 ┆ str ┆ i64 │\r\n╞═══════╪═══════════════════════════════════╪═══════╡\r\n│ 0 ┆ hide new secretions from the par… ┆ 0 │\r\n│ 1 ┆ contains no wit , only labored g… ┆ 0 │\r\n│ 2 ┆ that loves its characters and co… ┆ 1 │\r\n│ 3 ┆ remains utterly satisfied to rem… ┆ 0 │\r\n│ … ┆ … ┆ … │\r\n│ 67345 ┆ anguish , anger and frustration ┆ 0 │\r\n│ 67346 ┆ at achieving the modest , crowd-… ┆ 1 │\r\n│ 67347 ┆ a patient viewer ┆ 1 │\r\n│ 67348 ┆ this new jangle of noise , mayhe… ┆ 0 │\r\n└───────┴───────────────────────────────────┴───────┘\r\n```\r\n\r\n* `_repr_html_`:\r\n<img width=\"251\" alt=\"Screenshot 2023-07-12 at 18 25 58\" src=\"https://github.com/huggingface/datasets/assets/47462742/5d04519d-f302-4411-9fbc-7445bdf53b23\">\r\n\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" } ]

[ "Hi ! I just tested and it worked fine today for me.\r\n\r\nI think this is because the dataset is stored on Google Drive which has a quota limit for the number of downloads per day, see this similar issue https://github.com/huggingface/datasets/issues/996 \r\n\r\nFeel free to try again today, now that the quota was reset", "Fixed once data URL was replaced:\r\n- #4023" ]

[]

[ "Hi @thomwolf, for simplification purposes, I think you could remove the \"`pip install ...`\" steps from this commit, 'cause these deps (black, isort, flake8) are already installed on `pip install -e \".[dev]\"` on the [Start by preparing your environment](https://github.com/huggingface/datasets/blob/704107f924e74445f6f0fbde69a218b72178b588/ADD_NEW_DATASET.md#start-by-preparing-your-environment)\r\n" ]

[]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_5891). All of your documentation changes will be reflected on that endpoint.", "<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) | 0.006916 / 0.011353 (-0.004437) | 0.004749 / 0.011008 (-0.006259) | 0.096086 / 0.038508 (0.057578) | 0.035448 / 0.023109 (0.012338) | 0.299645 / 0.275898 (0.023747) | 0.331279 / 0.323480 (0.007799) | 0.006018 / 0.007986 (-0.001968) | 0.004210 / 0.004328 (-0.000118) | 0.072998 / 0.004250 (0.068747) | 0.050082 / 0.037052 (0.013030) | 0.297714 / 0.258489 (0.039225) | 0.365523 / 0.293841 (0.071682) | 0.028081 / 0.128546 (-0.100465) | 0.009072 / 0.075646 (-0.066574) | 0.327628 / 0.419271 (-0.091643) | 0.051165 / 0.043533 (0.007633) | 0.295091 / 0.255139 (0.039952) | 0.320052 / 0.283200 (0.036852) | 0.109841 / 0.141683 (-0.031842) | 1.467867 / 1.452155 (0.015712) | 1.572600 / 1.492716 (0.079884) |\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) | 0.281490 / 0.018006 (0.263484) | 0.499259 / 0.000490 (0.498770) | 0.000691 / 0.000200 (0.000491) | 0.000062 / 0.000054 (0.000008) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.027548 / 0.037411 (-0.009863) | 0.106592 / 0.014526 (0.092066) | 0.118654 / 0.176557 (-0.057902) | 0.174313 / 0.737135 (-0.562822) | 0.124491 / 0.296338 (-0.171848) |\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) | 0.399674 / 0.215209 (0.184465) | 3.984092 / 2.077655 (1.906437) | 1.790935 / 1.504120 (0.286815) | 1.593612 / 1.541195 (0.052417) | 1.694595 / 1.468490 (0.226105) | 0.517588 / 4.584777 (-4.067189) | 3.724353 / 3.745712 (-0.021359) | 3.244807 / 5.269862 (-2.025054) | 1.602929 / 4.565676 (-2.962748) | 0.065334 / 0.424275 (-0.358941) | 0.012259 / 0.007607 (0.004652) | 0.501355 / 0.226044 (0.275311) | 4.996546 / 2.268929 (2.727618) | 2.279333 / 55.444624 (-53.165291) | 1.940126 / 6.876477 (-4.936351) | 2.122945 / 2.142072 (-0.019128) | 0.626104 / 4.805227 (-4.179123) | 0.141278 / 6.500664 (-6.359386) | 0.064522 / 0.075469 (-0.010947) |\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) | 1.195351 / 1.841788 (-0.646436) | 15.258932 / 8.074308 (7.184624) | 14.627623 / 10.191392 (4.436231) | 0.266897 / 0.680424 (-0.413527) | 0.017557 / 0.534201 (-0.516644) | 0.392932 / 0.579283 (-0.186351) | 0.416409 / 0.434364 (-0.017955) | 0.469100 / 0.540337 (-0.071237) | 0.556247 / 1.386936 (-0.830689) |\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) | 0.006880 / 0.011353 (-0.004473) | 0.004837 / 0.011008 (-0.006171) | 0.074518 / 0.038508 (0.036010) | 0.034204 / 0.023109 (0.011095) | 0.365100 / 0.275898 (0.089202) | 0.394976 / 0.323480 (0.071496) | 0.006364 / 0.007986 (-0.001621) | 0.004269 / 0.004328 (-0.000060) | 0.073531 / 0.004250 (0.069281) | 0.051334 / 0.037052 (0.014281) | 0.373904 / 0.258489 (0.115415) | 0.413662 / 0.293841 (0.119821) | 0.028779 / 0.128546 (-0.099767) | 0.009292 / 0.075646 (-0.066354) | 0.081574 / 0.419271 (-0.337698) | 0.046531 / 0.043533 (0.002998) | 0.368995 / 0.255139 (0.113856) | 0.376938 / 0.283200 (0.093739) | 0.112576 / 0.141683 (-0.029107) | 1.458880 / 1.452155 (0.006725) | 1.550918 / 1.492716 (0.058202) |\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) | 0.319521 / 0.018006 (0.301515) | 0.510146 / 0.000490 (0.509656) | 0.000438 / 0.000200 (0.000238) | 0.000059 / 0.000054 (0.000004) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.033082 / 0.037411 (-0.004329) | 0.118009 / 0.014526 (0.103483) | 0.127108 / 0.176557 (-0.049448) | 0.176600 / 0.737135 (-0.560535) | 0.133790 / 0.296338 (-0.162549) |\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) | 0.437360 / 0.215209 (0.222151) | 4.367426 / 2.077655 (2.289771) | 2.193646 / 1.504120 (0.689526) | 2.025002 / 1.541195 (0.483808) | 2.142347 / 1.468490 (0.673856) | 0.525497 / 4.584777 (-4.059280) | 3.751275 / 3.745712 (0.005563) | 1.912271 / 5.269862 (-3.357590) | 1.087286 / 4.565676 (-3.478390) | 0.066328 / 0.424275 (-0.357947) | 0.011904 / 0.007607 (0.004297) | 0.545870 / 0.226044 (0.319825) | 5.434481 / 2.268929 (3.165552) | 2.719745 / 55.444624 (-52.724880) | 2.445001 / 6.876477 (-4.431476) | 2.500205 / 2.142072 (0.358133) | 0.645735 / 4.805227 (-4.159492) | 0.144210 / 6.500664 (-6.356455) | 0.065688 / 0.075469 (-0.009781) |\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) | 1.273522 / 1.841788 (-0.568265) | 15.771778 / 8.074308 (7.697470) | 14.685261 / 10.191392 (4.493869) | 0.176523 / 0.680424 (-0.503900) | 0.017877 / 0.534201 (-0.516324) | 0.392687 / 0.579283 (-0.186596) | 0.449992 / 0.434364 (0.015628) | 0.462851 / 0.540337 (-0.077487) | 0.560178 / 1.386936 (-0.826758) |\n\n</details>\n</details>\n\n![](https://cml.dev/watermark.png#0fa3ef6eba906ee1214e0596d15a78fc358909f4 \"CML watermark\")\n" ]

[]

[ "So I was curious why the issue reported at #1919 wasn't caught in [this test](https://github.com/huggingface/datasets/blob/248104c4bdb2e01c036b7578867199191fbff181/tests/test_arrow_dataset.py#L209), so I did some digging.\r\nI tried to save to a temporary directory (just like in the test), like this:\r\n```python\r\nwith tempfile.TemporaryDirectory() as requested_tempdir:\r\n squad.save_to_disk(requested_tempdir) # no error\r\n```\r\nand it executes succesfuly without problems.\r\nSo why does it work, but this doesn't?\r\n```python\r\nsquad.save_to_disk(\"./squad\") # error\r\n```\r\nIt's because `save_to_disk` also creates a temporary directory (let's call it `tempdir`), and since `tempdir` and `requested_tempdir` share the same parents, the `Path.joinpath` method [(here)](https://github.com/huggingface/datasets/blob/248104c4bdb2e01c036b7578867199191fbff181/src/datasets/arrow_dataset.py#L469) will keep `requested_tempdir` as it is and the *train* directory will be created under `requested_tempdir` and hence no errors will arise.\r\n\r\nBut in the second case (where we are saving to a local dir), the *train* directory is created under *squad* which in turn is created under `tempdir`, not under `.` (current dir).\r\n\r\nSo, all of this probably doesn't help solving the issue but it might help creating a better test, and it also makes me wonder why are we saving to a temporary dir in `save_to_disk` anyway? I mean, won't it be removed with all its contents upon execution completion? what's the point then? ", "CLosing in favor of #1923" ]

[ { "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 @smeyerhot, thanks for reporting.\r\n\r\nYou are right: there is an issue with the dataset metadata. I'm fixing it.\r\n\r\nIn the meantime, you can circumvent the issue by passing `ignore_verifications=True`:\r\n```python\r\nraw_datasets = load_dataset(\"medical_dialog\", \"en\", split=\"train\", download_mode=\"force_redownload\", data_dir=\"./Medical-Dialogue-Dataset-English\", ignore_verifications=True)\r\n```" ]

[]

[]

[]

[ "looks like this PR includes changes to many files other than the ones related to bprec\r\nCan you create another branch and another PR please ?", "> looks like this PR includes changes to many files other than the ones related to bprec\r\n> Can you create another branch and another PR please ?\r\n\r\nYes, I realized I messed this one up, learning my way :) I'll close this one and open another hopefully clean PR :) Thanks!" ]

[ { "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" }, { "color": "cfd3d7", "default": true, "description": "This issue or pull request already exists", "id": 1935892865, "name": "duplicate", "node_id": "MDU6TGFiZWwxOTM1ODkyODY1", "url": "https://api.github.com/repos/huggingface/datasets/labels/duplicate" } ]

[ "Hi @jon-tow, thanks for reporting.\r\n\r\nThis issue was already reported and its root cause is a change in the Google Drive service. See:\r\n- #3786 \r\n\r\nWe have already fixed it. See:\r\n- #3787 \r\n\r\nWe are planning to make a patch release today.\r\n\r\nIn the meantime, you can get this fix by installing our library from the GitHub master branch:\r\n```shell\r\npip install git+https://github.com/huggingface/datasets#egg=datasets\r\n```\r\nThen, if you had previously tried to load the data and got the checksum error, you should force the redownload of the data (before the fix, you just downloaded and cached the virus scan warning page, instead of the data file):\r\n```shell\r\nload_dataset(\"...\", download_mode=\"force_redownload\")\r\n```" ]

[]

[ "I think that's not related to the config name but the filenames in the dummy data. Mostly it occurs with files downloaded from drive. In that case the dummy file name is extracted from the google drive link and it corresponds to what comes after `https://drive.google.com/`\r\n\r\n", "> I think that's not related to the config name but the filenames in the dummy data. Mostly it occurs with files downloaded from drive. In that case the dummy file name is extracted from the google drive link and it corresponds to what comes after `https://drive.google.com/`\r\n\r\nThe filenames of the dummy data are now encoded (see #173). So this is not a problem anymore.\r\n\r\nThe problem here is different and comes from the directory names where we save the arrow files (basically `dataset_name/config_name/version`). In this case we could have invalid directory names because of the config name\r\n", "Okay great then.", "I like the method, but I'm wondering whether it should just be a test method instead of a `__post_init__` function. From a logical point of view the only reason this error would be thrown is because of an invalid config name introduced when creating the dataset script / adding a new dataset => so I think it might be better to write a simple test for this in `test_dataset_common.py`...what do you think @lhoestq ?", "`test_dataset_common.py` only tests canonical datasets no ? What if users wants to create their own script ?", "> `test_dataset_common.py` only tests canonical datasets no ? What if users wants to create their own script ?\r\n\r\nIt tests all dataset that can be loaded either locally or on AWS (which includes all non-canonical datasets as well)...by their own script you mean like a private dataset script that they don't want to be public? I guess even then they could locally run the test functions to check...", "We could have a bunch of simple consistency tests that run before uploading with the CLI (without loading data if we don't want to force the user to have dummy data)?", "Let's say someone want to create his own private script. As the script is not meant to be shared, it's not going to be placed in `/datasets` right ? Maybe the script is going to be inside another project. If I'm not wrong in this case the `test_dataset_common.py` is not going to test his script.\r\n\r\nRaising an error in the post init is a sanity check that would tell the user immediately what's wrong.\r\nThe error is raised if he tried to load the script or if he uses `nlp-cli test`", "> Let's say someone want to create his own private script. As the script is not meant to be shared, it's not going to be placed in `/datasets` right ? Maybe the script is going to be inside another project. If I'm not wrong in this case the `test_dataset_common.py` is not going to test his script.\r\n> \r\n> Raising an error in the post init is a sanity check that would tell the user immediately what's wrong.\r\n> The error is raised if he tried to load the script or if he uses `nlp-cli test`\r\n\r\nOK, fair point! I'm good with this then :-) ", "I'm fine with this as well (even though I understand what you meant @patrickvonplaten, we can still change it later if needed)", "> We could have a bunch of simple consistency tests that run before uploading with the CLI (without loading data if we don't want to force the user to have dummy data)?\r\n\r\nYes! I guess that's a big question whether we should force the user to add dummy data. It's probably too tedious for the user...so when uploading to circle ci should we just check \r\n- 1) All configs can be instantiated (if there are any)\r\n- 2) The BuilderClass can be instantiated ... \r\n- 3) ... maybe some more\r\n\r\nand maybe suggest to the user to add dummy data using the dummy data command?", "I really like that we have a test with dummy data for canonical datasets. This is insurance that they'll keep working in the long run. \r\n\r\nOn the other hand I understand that we will probably not force this practice for scripts uploaded on S3 by a user under his namespace (non-canonical), as it is tedious. As I understand right now the test is done for all the datasets on aws, even the non-canonical ? We should think about different tests for non-canonical datasets.\r\n\r\nI also like the idea of a simple consistency test !", "Merging this one for now, we can think about the test for non-canonical datasets later" ]

[ { "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" } ]

[ "The pile dataset would be very nice.\r\nBenchmarks show that pile trained models achieve better results than most of actually trained models", "The pile can very easily be added and adapted using this [tfds implementation](https://github.com/EleutherAI/The-Pile/blob/master/the_pile/tfds_pile.py) from the repo. \r\n\r\nHowever, the question is whether you'd be ok with 800GB+ cached in your local disk, since the tfds implementation was designed to offload the storage to Google Cloud Storage.", "With the dataset streaming feature (see #2375) it will be more convenient to play with such big datasets :)\r\nI'm currently adding C4 (see #2511 ) but I can probably start working on this afterwards", "Hi folks! Just wanted to follow up on this -- would be really nice to get the Pile on HF Datasets... unclear if it would be easy to also add partitions of the Pile subject to the original 22 datasets used, but that would be nice too!", "Hi folks, thanks to some awesome work by @lhoestq and @albertvillanova you can now stream the Pile as follows:\r\n\r\n```python\r\n# Install master branch of `datasets`\r\npip install git+https://github.com/huggingface/datasets.git#egg=datasets[streaming]\r\npip install zstandard\r\n\r\nfrom datasets import load_dataset\r\n\r\ndset = load_dataset(\"json\", data_files=\"https://the-eye.eu/public/AI/pile/train/00.jsonl.zst\", streaming=True, split=\"train\")\r\nnext(iter(dset))\r\n# {'meta': {'pile_set_name': 'Pile-CC'},\r\n# 'text': 'It is done, and submitted. You can play “Survival of the Tastiest” on Android, and on the web ... '}\r\n```\r\n\r\nNext step is to add the Pile as a \"canonical\" dataset that can be streamed without specifying the file names explicitly :)", "> Hi folks! Just wanted to follow up on this -- would be really nice to get the Pile on HF Datasets... unclear if it would be easy to also add partitions of the Pile subject to the original 22 datasets used, but that would be nice too!\r\n\r\nHi @siddk thanks to a tip from @richarddwang it seems we can access some of the partitions that EleutherAI created for the Pile [here](https://the-eye.eu/public/AI/pile_preliminary_components/). What's missing are links to the preprocessed versions of pre-existing datasets like DeepMind Mathematics and OpenSubtitles, but worst case we do the processing ourselves and host these components on the Hub.\r\n\r\nMy current idea is that we could provide 23 configs: one for each of the 22 datasets and an `all` config that links to the train / dev / test splits that EleutherAI released [here](https://the-eye.eu/public/AI/pile/), e.g.\r\n\r\n```python\r\nfrom datasets import load_dataset\r\n\r\n# Load a single component\r\nyoutube_subtitles = load_dataset(\"the_pile\", \"youtube_subtitles\")\r\n# Load the train / dev / test splits of the whole corpus\r\ndset = load_dataset(\"the_pile\", \"all\")\r\n```\r\n\r\nIdeally we'd like everything to be compatible with the streaming API and there's ongoing work by @albertvillanova to make this happen for the various compression algorithms.\r\n\r\ncc @lhoestq ", "Ah I just saw that @lhoestq is already thinking about the specifying of one or more subsets in [this PR](https://github.com/huggingface/datasets/pull/2817#issuecomment-901874049) :)" ]

[]

[ "@SBrandeis all checks passing" ]

[]

[ "Hi ! Thanks for reporting, indeed `disable_nullable` doesn't seem to be supported in this case. Maybe at one point we can have `disable_nullable` as a parameter of certain feature types" ]

[ { "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" } ]

[ "found a solution here https://discuss.pytorch.org/t/train-simultaneously-on-two-datasets/649/35, closed for now, thanks " ]

[]

[]

[]

[]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "hahaha what a timing, I added my comment right after you merged x)\r\n\r\nyou can ignore my (nit), it's fine", "Sorry, just too sync... :sweat_smile: " ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "Hi @mariosasko , I have added a test for the `keep_in_memory` version. I have also removed the `Compatible with temp_seed` part in the scope of `dset_shuffled`, please verify if that makes sense." ]

[ { "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 @qqaatw.\r\n\r\n@mishig25 @sgugger do you think this can be tweaked in the new doc framework?\r\n- From: https://github.com/huggingface/datasets/blob/v2.0.0/\r\n- To: https://github.com/huggingface/datasets/blob/2.0.0/", "@qqaatw thanks a lot for notifying about this issue!\r\n\r\nin comparison, transformers tags start with `v` like [this one](https://github.com/huggingface/transformers/blob/v4.17.0/src/transformers/models/bert/configuration_bert.py#L54).\r\n\r\nTherefore, we have to do one of 2 options below:\r\n1. Make necessary changes on doc-builder side\r\nOR\r\n2. Make [datasets tags](https://github.com/huggingface/datasets/tags) start with `v`, just like [transformers](https://github.com/huggingface/transformers/tags) (so that tag naming can be consistent amongst hf repos)\r\n\r\nI'll let you decide @albertvillanova @lhoestq @sgugger ", "I think option 2 is the easiest and would provide harmony in the HF ecosystem but we can also add a doc config parameter to decide whether the default version has a v or not if `datasets` folks prefer their tags without a v :-)", "For me it is OK to conform to the rest of libraries and tag/release with a preceding \"v\", rather than adding an extra argument to the doc builder just for `datasets`.\r\n\r\nLet me know if it is also OK for you @lhoestq. ", "https://github.com/huggingface/doc-build/commit/f41c1e8ff900724213af4c75d287d8b61ecf6141\r\n\r\nhotfix so that `datasets` docs source button works correctly on hf.co/docs/datasets", "We could add a tag for each release without a 'v' but it could be confusing on github to see both tags `v2.0.0` and `2.0.0` IMO (not sure if many users check them though). Removing the tags without 'v' would break our versioning for github datasets: the library looks for dataset scripts at the URLs like `https://raw.githubusercontent.com/huggingface/datasets/{revision}/datasets/{path}/{name}` where `revision` is equal to `datasets.__version__` (which doesn't start with a 'v') for all released versions of `datasets`.\r\n\r\nI think we could just have a parameter for the documentation - and having different URLs schemes for the source links that the users don't even see (they simply click on a button) is probably fine", "This is done in #3943 to go along with [doc-builder#146](https://github.com/huggingface/doc-builder/pull/146).\r\n\r\nNote that this will only work for future versions, so once those two are merged, the actual v2.0.0 doc should be fixed. The easiest is to cherry-pick this commit on the v2.0.0 release branch (or on a new branch created from the 2.0.0 tag, with a name that triggers the doc building job, for instance v2.0.0-release)", "Thanks for fixing @sgugger." ]

[]

[]

[]

[ "Thanks for reporting! \r\n\r\nSpecifying the target features explicitly should avoid this error:\r\n```python\r\ndataset = dataset.map(\r\n batch_process,\r\n batched=True,\r\n batch_size=1,\r\n num_proc=1,\r\n remove_columns=dataset.column_names,\r\n features=datasets.Features({\"texts\": datasets.Sequence(datasets.Value(\"string\"))})\r\n)\r\n```\r\n\r\nThis error stems from our type promotion not handling the nested case. But this promotion/casting allocates memory in most scenarios, which can be problematic for large datasets, so explicitly passing the features is the optimal solution.", "Hi @mariosasko thanks for the context, this is helpful to know. Would it be worth having some logic to generate this explicit feature specification automatically if a type annotation for a .map returns a dataclass that can be inferred?\r\n\r\nFeels like something that would be easy to implement and could save memory / deal with this case in a standardized way.", "> . Would it be worth having some logic to generate this explicit feature specification automatically if a type annotation for a .map returns a dataclass that can be inferred?\r\n\r\nInteresting proposal! Yes, we could consider doing this if the (return) type hint is `TypedDict`, and raise an error that type hints are incorrect if the cast using the inferred types fails.", "@mariosasko Put up an initial PR to implement this proposal. Let me know your thoughts on direction and what else should be in-scope here." ]

[]

[]

[]

[ "This PR can be merged after #3163 is merged (this PR is pretty big because I was working on the forked branch).\r\n\r\n@lhoestq @albertvillanova Could you please take a look at the changes in `src/datasets/utils/streaming_download_manager.py`? These changes were required to support streaming of the `cats_vs_dogs` and the `beans` datasets.", "The CI failures are due to the missing fields in the README files.", "and thanks for adding support for Path.name and Path.parent for streaming :)" ]

[]

[ "Not yet but it’s on the short/mid-term roadmap (requested by many indeed).", "oh, great, really awesome feature to have, thank you very much for the great, fabulous work", "We'll work on dataset streaming soon. This should allow you to only load the examples you need ;)", "thanks a lot Quentin, this would be really really a great feature to have\n\nOn Wed, Apr 7, 2021 at 12:14 PM Quentin Lhoest ***@***.***>\nwrote:\n\n> We'll work on dataset streaming soon. This should allow you to only load\n> the examples you need ;)\n>\n> —\n> You are receiving this because you authored the thread.\n> Reply to this email directly, view it on GitHub\n> <https://github.com/huggingface/datasets/issues/2161#issuecomment-814791922>,\n> or unsubscribe\n> <https://github.com/notifications/unsubscribe-auth/AS37NMROD62QAKIJMAKWISTTHQWBVANCNFSM42IUI5JQ>\n> .\n>\n", "Is streaming completed? On the 1.8.0 docs it is mentioned (https://huggingface.co/docs/datasets/dataset_streaming.html), but when following the example I get the following error:\r\n\r\n```\r\n>>> dataset2 = load_dataset(\"amazon_us_reviews\", \"Pet_Products_v1_00\", split='train', streaming=True)\r\n\r\n---------------------------------------------------------------------------\r\nValueError Traceback (most recent call last)\r\n<ipython-input-21-1eedab26cff1> in <module>()\r\n----> 1 en_dataset = load_dataset('oscar', \"unshuffled_deduplicated_en\", split='train', streaming=True)\r\n\r\n3 frames\r\n/usr/local/lib/python3.7/dist-packages/datasets/builder.py in _create_builder_config(self, name, custom_features, **config_kwargs)\r\n 339 if value is not None:\r\n 340 if not hasattr(builder_config, key):\r\n--> 341 raise ValueError(f\"BuilderConfig {builder_config} doesn't have a '{key}' key.\")\r\n 342 setattr(builder_config, key, value)\r\n 343 \r\n\r\nValueError: BuilderConfig OscarConfig(name='unshuffled_deduplicated_en', version=1.0.0, data_dir=None, data_files=None, description='Unshuffled and deduplicated, English OSCAR dataset') doesn't have a 'streaming' key.\r\n```\r\n\r\nUPDATE: Managed to get streaming working by building from source and installing the additional `datasets[streaming]` package:\r\n\r\n```\r\n!pip install git+https://github.com/huggingface/datasets.git\r\n!pip install datasets[streaming]\r\n```", "Hi ! Streaming is available on `master` only right now. We'll make a new release 1.9.0 on Monday :)" ]

[]