Datasets:

---

Francesco-A

/

github-issues_huggingface-datasets

like 0

[]

[ "Yes, this is intended." ]

[]

[ "We can look into removing some of the attributes from `GET /api/datasets` to make it smaller/faster, what do you think @lhoestq?", "> We can look into removing some of the attributes from `GET /api/datasets` to make it smaller/faster, what do you think @lhoestq?\r\n\r\nyes at least remove all the `dummy_data.zip`", "`GET /api/datasets` should now be much faster. @zhuzilin can you check if `list_datasets` is now faster for you?", "> `GET /api/datasets` should now be much faster. @zhuzilin can you check if `list_datasets` is now faster for you?\r\n\r\nYes, much faster! Thank you!" ]

[]

[ "`dummy_data` right now contains all article files, keeping only the required articles for dummy data fails the dummy data test.\r\nAny idea ?", "> `dummy_data` right now contains all article files, keeping only the required articles for dummy data fails the dummy data test.\r\n> Any idea ?\r\n\r\nWe should definitely find a way to make it work with only a few articles.\r\n\r\nIf it doesn't work right now for dummy data, I guess it's because it tries to load every single article file ?\r\n\r\nIf so, then maybe you can use `os.listdir` method to first check all the data files available in the path where the `articles.tgz` file is extracted. Then you can simply iter through the data files and depending on their ID, include them in the train or test set. With this method you should be able to have only a few articles files per split in the dummy data. Does that make sense ?", "fixed! so the issue was, `articles_ids` were prepared based on the number of files in articles dir, so for dummy data questions it was not able to load some articles due to incorrect ids and the test was failing" ]

[ { "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" } ]

[ "Why not adding `max_examples` as part of the config name?", "Yup it can also work, and maybe it's simpler this way. Opening a PR to fix bigbench instead of https://github.com/huggingface/datasets/pull/4463", "Hi @lhoestq,\r\n\r\nThank you for taking a look at this issue, and proposing a solution. \r\nUnfortunately, after trying the fix in #4465 I still see the same issue.\r\n\r\nI think there is some subtlety where the config name gets overwritten somewhere when `BUILDER_CONFIGS`[(link)](https://github.com/huggingface/datasets/blob/master/datasets/bigbench/bigbench.py#L126) is defined. \r\n\r\nIf I print out the `self.config.name` in the current version (with the fix in #4465), I see just the task name, but if I comment out `BUILDER_CONFIGS`, the `num_shots` and `max_examples` gets appended as was meant by #4465.\r\n\r\nI haven't managed to track down where this happens, but I thought you might know? \r\n\r\n(Another comment on your fix: the `name` variable is used to fetch the task from the bigbench API, so modifying it causes an error if it's actually called. This can easily be fixed by having `config_name` variable in addition to the `task_name`)\r\n\r\n\r\n" ]

[]

[]

[ { "color": "0075ca", "default": true, "description": "Improvements or additions to documentation", "id": 1935892861, "name": "documentation", "node_id": "MDU6TGFiZWwxOTM1ODkyODYx", "url": "https://api.github.com/repos/huggingface/datasets/labels/documentation" } ]

[ "_The documentation is not available anymore as the PR was closed or merged._", "> Let me know what you think, especially if we should include some code samples for training a model in the audio/vision sections. I left this out since we already showed it in the NLP section. I want to keep the focus on using Datasets to load and process a dataset, and not so much the training part. Maybe we can add links to the Transformers docs instead?\r\n\r\nWe plan to address this with end-to-end examples (for each modality) more focused on preprocessing than the ones in the Transformers docs." ]

[]

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

[]

[]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "Merging this one since mario is off, I took care of adding some tests to make sure everything is fine. Will do the release after it" ]

[]

[]

[]

[ "LGTM" ]

[]

[ "Do you want to convert a dataset script to the tfds format ?\r\nIf so, we currently have a comversion script nlp/commands/convert.py but it is a conversion script that goes from tfds to nlp.\r\nI think it shouldn't be too hard to do the changes in reverse (at some manual adjustments).\r\nIf you manage to make it work in reverse, feel free to open a PR to share it with the community :)", "In our docs: [Using a Dataset with PyTorch/Tensorflow](https://huggingface.co/docs/datasets/torch_tensorflow.html)." ]

[]

[]

[ { "color": "d876e3", "default": true, "description": "Further information is requested", "id": 1935892912, "name": "question", "node_id": "MDU6TGFiZWwxOTM1ODkyOTEy", "url": "https://api.github.com/repos/huggingface/datasets/labels/question" } ]

[ "Hi @jordiae,\r\n\r\nInstead of the `Sequence` feature, you can use just a regular list: put the dict between `[` and `]`:\r\n```python\r\n\"list_of_dict_feature\": [\r\n {\r\n \"key1_in_dict\": datasets.Value(\"string\"),\r\n \"key2_in_dict\": datasets.Value(\"int32\"),\r\n ...\r\n }\r\n],\r\n```\r\n\r\nFeel free to re-open this issue if that does not work for your use case.", "> Hi @jordiae,\r\n> \r\n> Instead of the `Sequence` feature, you can use just a regular list: put the dict between `[` and `]`:\r\n> \r\n> ```python\r\n> \"list_of_dict_feature\": [\r\n> {\r\n> \"key1_in_dict\": datasets.Value(\"string\"),\r\n> \"key2_in_dict\": datasets.Value(\"int32\"),\r\n> ...\r\n> }\r\n> ],\r\n> ```\r\n> \r\n> Feel free to re-open this issue if that does not work for your use case.\r\n\r\nThank you" ]

[ { "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": "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" } ]

[ "Upgrading datasets to version 1.6 fixes the issue", "This bug was fixed in #1995. Upgrading the `datasets` should work! ", "Thanks @ekeleshian for having reported.\r\n\r\nI am closing this issue once that you updated `datasets`. Feel free to reopen it if the problem persists." ]

[]

[ "done." ]

[]

[ "- I updated the previous dummy data: I just had to rename the file and its directory\r\n - the dummy data zip contains only a single file: `pubmed22n0001.xml.gz`\r\n\r\nThen I discover it fails: https://app.circleci.com/pipelines/github/huggingface/datasets/9800/workflows/173a4433-8feb-4fc6-ab9e-59762084e3e1/jobs/60437\r\n```\r\nNo such file or directory: '.../dummy_data/pubmed22n0002.xml.gz'\r\n```\r\n- it needs dummy data for all the 1114 files: \r\n `_URLs = [f\"ftp://ftp.ncbi.nlm.nih.gov/pubmed/baseline/pubmed22n{i:04d}.xml.gz\" for i in range(1, 1115)]`\r\n- this confirms me that it never passed the test: these dummy data files were not present before my PR\r\n- therefore, is it really useful the data test if we just ignore it when it does not pass?\r\n\r\nIn relation with JSON metadata, I was generating the file for `pubmed` (see above) in a GCP instance: after running during ~12h without finishing, I decided to stop the process.", "Hi ! Yes I remembered we hardcoded an exception for this one:\r\nhttps://github.com/huggingface/datasets/blob/36db39c75179a0a491c69a4491f7ae7e4615e66f/src/datasets/utils/mock_download_manager.py#L174-L176\r\n\r\nThe exception was used to only require one dummy data file, feel free to update it if you want" ]

[]

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

[]

[ "Possibly not related to the dataset viewer in itself. cc @huggingface/datasets.\r\n\r\nIn particular, I think that the import of bigbiohub is not working here: https://huggingface.co/datasets/bigscience-biomedical/biosses/blob/main/biosses.py#L29 (requires a relative path?)\r\n\r\n```python\r\n>>> from datasets import get_dataset_config_names\r\n>>> get_dataset_config_names('bigscience-biomedical/biosses')\r\nDownloading builder script: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 8.00k/8.00k [00:00<00:00, 7.47MB/s]\r\nTraceback (most recent call last):\r\n File \"<stdin>\", line 1, in <module>\r\n File \"/home/slesage/hf/datasets-server/services/worker/.venv/lib/python3.9/site-packages/datasets/inspect.py\", line 289, in get_dataset_config_names\r\n dataset_module = dataset_module_factory(\r\n File \"/home/slesage/hf/datasets-server/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py\", line 1247, in dataset_module_factory\r\n raise e1 from None\r\n File \"/home/slesage/hf/datasets-server/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py\", line 1220, in dataset_module_factory\r\n return HubDatasetModuleFactoryWithScript(\r\n File \"/home/slesage/hf/datasets-server/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py\", line 931, in get_module\r\n local_imports = _download_additional_modules(\r\n File \"/home/slesage/hf/datasets-server/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py\", line 215, in _download_additional_modules\r\n raise ImportError(\r\nImportError: To be able to use bigscience-biomedical/biosses, you need to install the following dependency: bigbiohub.\r\nPlease install it using 'pip install bigbiohub' for instance'\r\n```", "Opened a PR here to (hopefully) fix the dataset script: https://huggingface.co/datasets/bigscience-biomedical/biosses/discussions/1/files", "thanks for taking a look @severo . agree this isn't related to dataset viewer (sorry just clicked on the auto issue creator). also thanks @lhoestq , I see the format to use for relative imports. was a bit confused b/c it seems to be working here \r\n\r\nhttps://huggingface.co/datasets/bigscience-biomedical/scitail/blob/main/scitail.py#L31\r\n\r\nI'll try this PR a see what happens. ", "closing as I think the issue is relative imports and attempting to read json files directly in the repo (thanks again @lhoestq ) " ]

[ { "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": "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, can I work on this?", "Hi, sure! Let us know if you need some pointers/help." ]

[]

[]

[]

[ "Hi ! It's been fixed by https://github.com/huggingface/datasets/pull/5598. We're doing a new release tomorrow with the fix and you'll be able to push your 100k images ;)\r\n\r\nBasically `push_to_hub` used to fail if the remote repository already exists and has a README.md without dataset_info in the YAML tags.\r\n\r\nIn the meantime you can install datasets from source", "Hi @lhoestq ,\r\n\r\nWhat version of datasets library fix this case? I am using the last `v2.10.1` and I get the same error.", "We just released 2.11 which includes a fix :)" ]

[]

[ "Thanks :) merging it" ]

[]

[]

[ { "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" } ]

[ "uploaded dataset [here](https://huggingface.co/datasets/embedding-data/Amazon-QA)." ]

[ { "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 ! In the `transformers` issue the object to not hash is a `Pool` - I think you can instantiate it inside your function instead of passing it as a parameter. It's good practice that your function and all its fn_kwargs are picklable, in case you want to parallelize `map` using `num_proc>1`\r\n\r\nFor the other case `def fn(example, verbose=False):` however, I agree it would be nice to let the user specify that \"verbose\" needs to be ignored.\r\n\r\nDo you think providing a decorator could help ? Maybe\r\n```python\r\[email protected](ignore_kwargs=[\"verbose\"])\r\ndef func(example, verbose=False):\r\n ...\r\n```", "Hi @lhoestq! Thanks for your response.\r\n\r\nA `Pool` shouldn't be instantiated within the function, because there's a huge overhead in doing so. The main idea is that the same `Pool` should be used across all function calls. Parallel `map` is not helpful/desired in that specific scenario, because the heavy parallel computation is done by another lib (`pyctcdecode`, called within `transformer`'s model inference code).\r\n\r\nBut yes, it makes sense to be able to leverage parallel processing by just doing `num_proc>1` when possible.\r\n\r\nYour decorator suggestions seems like a pretty clean API to me. I didn't find a `datasets.hashing` module though. Would it be created for this specific purpose? Any downsides in just using `datasets.fingerprint`?\r\n\r\nAnd would `datasets.hashing.register` just add some metadata to `func` in your approach (so it could be inspected from `fingerprint_transform`)?\r\n\r\nAnd looking to the `datasets.Dataset` API, `.filter` would also benefited from this.", "> Would it be created for this specific purpose? Any downsides in just using datasets.fingerprint?\r\n\r\nThis can also go in datasets.fingerprint indeed - but maybe datasets.hashing tells more about what the register function does (i.e. register this function to have a custom hashing) ?\r\n\r\n> And would datasets.hashing.register just add some metadata to func in your approach (so it could be inspected from fingerprint_transform)?\r\n\r\nYup that's the idea :)\r\n\r\n> And looking to the datasets.Dataset API, .filter would also benefited from this.\r\n\r\nIndeed !\r\n\r\n-----\r\n\r\nIf you would like to contribute this you can assign yourself to this issue by posting #self-assign\r\nAnd of course if you have questions or if I can help, feel free to ping me !", "> This can also go in datasets.fingerprint indeed - but maybe datasets.hashing tells more about what the register function does (i.e. register this function to have a custom hashing) ?\r\n\r\nSure, it makes sense.\r\n\r\n---\r\n\r\nI don't plan to work on it right now, so I'll let it unassigned in case somebody wants to join. I'll get back at it as soon as possible though.\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" } ]

[]

[]

[]

[]

[ "I manage to remove the mentions of ellipsis in the code by launching the command as follows:\r\n\r\n```\r\npython -m doctest -v docs/source/load_hub.rst -o=ELLIPSIS\r\n```\r\n\r\nThe way you put your ellipsis will only work on mac, I've adapted it for linux as well with the following:\r\n\r\n```diff\r\n >>> from datasets import load_dataset_builder\r\n >>> dataset_builder = load_dataset_builder('imdb')\r\n- >>> print(dataset_builder.cache_dir) #doctest: +ELLIPSIS\r\n- /Users/.../.cache/huggingface/datasets/imdb/plain_text/1.0.0/...\r\n+ >>> print(dataset_builder.cache_dir)\r\n+ /.../.cache/huggingface/datasets/imdb/plain_text/1.0.0/...\r\n```\r\n\r\nThis passes on my machine:\r\n\r\n```\r\nTrying:\r\n print(dataset_builder.cache_dir)\r\nExpecting:\r\n /.../.cache/huggingface/datasets/imdb/plain_text/1.0.0/...\r\nok\r\n```\r\n\r\nI'm getting a last error:\r\n\r\n```py\r\nExpected:\r\n DatasetDict({\r\n train: Dataset({\r\n features: ['sentence1', 'sentence2', 'label', 'idx'],\r\n num_rows: 3668\r\n })\r\n validation: Dataset({\r\n features: ['sentence1', 'sentence2', 'label', 'idx'],\r\n num_rows: 408\r\n })\r\n test: Dataset({\r\n features: ['sentence1', 'sentence2', 'label', 'idx'],\r\n num_rows: 1725\r\n })\r\n })\r\nGot:\r\n DatasetDict({\r\n train: Dataset({\r\n features: ['idx', 'label', 'sentence1', 'sentence2'],\r\n num_rows: 3668\r\n })\r\n validation: Dataset({\r\n features: ['idx', 'label', 'sentence1', 'sentence2'],\r\n num_rows: 408\r\n })\r\n test: Dataset({\r\n features: ['idx', 'label', 'sentence1', 'sentence2'],\r\n num_rows: 1725\r\n })\r\n })\r\n```\r\n\r\nBut this is due to `doctest` looking for an exact match and the list having an unordered print order. I wish `doctest` would be a bit more flexible with that." ]

[]

[]

[]

[ "I have seen that somebody is hard working on easierly loadable wikipedia. #129 \r\nMaybe I should wait a few days for that version ?", "Yes we (well @lhoestq) are very actively working on this." ]

[]

[ "Hi ! `load_dataset` downloads the full dataset once and caches it, so that subsequent calls to `load_dataset` just reloads the dataset from your disk.\r\nThen when you specify a `split` in `load_dataset`, it will just load the requested split from the disk. If your specified split is a sliced split (e.g. `\"train[:10]\"`), then it will load the 10 first rows of the train split that you have on disk.\r\n\r\nTherefore, as long as you don't delete your cache, all your calls to `load_dataset` will be very fast. Except the first call that downloads the dataset of course ^^", "That’s a use case for the new streaming feature, no?", "Hi @reachtarunhere.\r\n\r\nBesides the above insights provided by @lhoestq and @thomwolf, there is also a Dataset feature in progress (I plan to finish it this week): #2249, which will allow you, when calling `load_dataset`, to pass the option to download/preprocess/cache only some specific split(s), which will definitely speed up your workflow.\r\n\r\nIf this feature is interesting for you, I can ping you once it will be merged into the master branch.", "Thanks all for responding.\r\n\r\nHey @albertvillanova \r\n\r\nThanks. Yes, I would be interested.\r\n\r\n@lhoestq I think even if a small split is specified it loads up the full dataset from the disk (please correct me if this is not the case). Because it does seem to be slow to me even on subsequent calls. There is no repeated downloading so it seems that the cache is working.\r\n\r\nI am not aware of the streaming feature @thomwolf mentioned. So I might need to read up on it.", "@reshinthadithyan I use the .select function to have a fraction of indices.", "If I want to create a dataset, containing only the 10 elements of a given dataset (slice it), how do I do that?", "```python \r\nsmall_ds = ds.select(range(10))\r\n```", "\r\n\r\n> ```python\r\n> small_ds = ds.select(range(10))\r\n> ```\r\n\r\nThanks, but this doesn't help me to save time during initial loading, right?", "Indeed by default load_dataset would download and prepare everything as Arrow files. And passing `split=train[:10]` memory maps only the beginning of the full dataset that has been prepared on disk.\r\n\r\nIf you don't want to download everything, you can use streaming : \r\n```python \r\nids = load_dataset(..., streaming=True)\r\nfirst_samples = list(ids[\"train\"].take(10))\r\n```\r\n\r\nTo get a Dataset you can use \r\n```python \r\nds = Dataset.from_generator(ids.take(10).__iter__)\r\n```\r\n\r\nedit: fixed small bug", "Thanks @lhoestq, but I don't think it is 100% accurate, as it doesn't keep the dataset structure exactly the same.\r\nTo load the full dataset, I do:\r\n```\r\ndata = load_dataset(\"json\", data_files=\"a.json\")\r\ntrain_data = data[\"train\"].shuffle()\r\n```\r\n\r\nBut when I am changing it as per your instructions: \r\n```\r\nids = load_dataset(\"json\", data_files=\"a.json\", streaming=True)\r\ndata = Dataset.from_generator(ids[\"train\"].take(1).__iter__)\r\ntrain_data = data[\"train\"].shuffle()\r\n```\r\nIt throws KeyError.\r\nI need a simple way, like you suggested, to have a subset of a Dataset, which exactly the same attributes.\r\n", "Whoops I fixed my code sorry\r\n```diff\r\n- ds = Dataset.from_generator(ids[\"train\"].take(10).__iter__)\r\n+ ds = Dataset.from_generator(ids.take(10).__iter__)\r\n```\r\n\r\nin your case that means running\r\n```python\r\ntrain_data = data.shuffle()\r\n```\r\n\r\nwithout `[\"train\"]`" ]

[]

[ "Hi !\r\n\r\nGood catch, thanks for creating this PR :)\r\n\r\nCould you also regenerate the metadata for this dataset using \r\n```\r\ndatasets-cli test ./datasets/gap --save_infos --all_configs\r\n```\r\n\r\nThat'd be awesome", "@lhoestq Thank you for your revieing!!!\r\n\r\nI've performed it and have read CONTRIBUTING.md now!" ]

[]

[]

[]

[]

[]

[ "I validated that this fixed the problem, thank you, @albertvillanova!\r\n", "still facing the same issue or similar:\r\nfrom datasets import load_dataset\r\nwtm14_test = load_dataset('wmt14',\"de-en\",cache_dir='./datasets')\r\n\r\n~\\.cache\\huggingface\\modules\\datasets_modules\\datasets\\wmt14\\43e717d978d2261502b0194999583acb874ba73b0f4aed0ada2889d1bb00f36e\\wmt_utils.py in _split_generators(self, dl_manager)\r\n 758 # Extract manually downloaded files.\r\n 759 manual_files = dl_manager.extract(manual_paths_dict)\r\n--> 760 extraction_map = dict(downloaded_files, **manual_files)\r\n 761 \r\n 762 for language in self.config.language_pair:\r\n\r\nTypeError: type object argument after ** must be a mapping, not list", "Hi @sabania \r\nWe released a patch version that fixes this issue (1.4.1), can you try with the new version please ?\r\n```\r\npip install --upgrade datasets\r\n```", "I re-validated with the hotfix and the problem is no more.", "It's working. thanks a lot." ]

[]

[ "Update: release is planed mid-next week." ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "The failing test has nothing to do with this PR:\r\n```\r\nFAILED tests/test_upstream_hub.py::TestPushToHub::test_push_dataset_dict_to_hub_multiple_files\r\n```" ]

[]

[ "https://huggingface.co/datasets/fgrezes/WIESP2022-NER\r\n\r\nThe error:\r\n\r\n```\r\nMessage: Couldn't find a dataset script at /src/services/worker/fgrezes/WIESP2022-NER/WIESP2022-NER.py or any data file in the same directory. Couldn't find 'fgrezes/WIESP2022-NER' on the Hugging Face Hub either: FileNotFoundError: Unable to resolve any data file that matches ['**test*', '**eval*'] in dataset repository fgrezes/WIESP2022-NER with any supported extension ['csv', 'tsv', 'json', 'jsonl', 'parquet', 'txt', 'blp', 'bmp', 'dib', 'bufr', 'cur', 'pcx', 'dcx', 'dds', 'ps', 'eps', 'fit', 'fits', 'fli', 'flc', 'ftc', 'ftu', 'gbr', 'gif', 'grib', 'h5', 'hdf', 'png', 'apng', 'jp2', 'j2k', 'jpc', 'jpf', 'jpx', 'j2c', 'icns', 'ico', 'im', 'iim', 'tif', 'tiff', 'jfif', 'jpe', 'jpg', 'jpeg', 'mpg', 'mpeg', 'msp', 'pcd', 'pxr', 'pbm', 'pgm', 'ppm', 'pnm', 'psd', 'bw', 'rgb', 'rgba', 'sgi', 'ras', 'tga', 'icb', 'vda', 'vst', 'webp', 'wmf', 'emf', 'xbm', 'xpm', 'zip']\r\n```\r\n\r\nI understand the issue is not related to the dataset viewer in itself, but with the autodetection of the data files without a loading script in the datasets library. cc @lhoestq @albertvillanova @mariosasko ", "Apparently it finds `scoring-scripts/compute_seqeval.py` which matches `**eval*`, a regex that detects a test split. We should probably improve the regex because it's not supposed to catch this kind of files. It must also only check for files with supported extensions: txt, csv, png etc." ]

[ { "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\ndid you fill out the env info section manually or by copy-pasting the output of the `datasets-cli env` command?\r\n\r\nThis code should work without issues on 1.6.2 version (I'm working on master (1.6.2.dev0 version) and can't reproduce this error).", "@mariosasko you are right I was still on `1.5.0`. " ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "Thank you thank you! Let's merge and pray? 😱 ", "I just need to add `license_details` to the validator and yup we can merge" ]

[]

[ "Good catch, thanks!", "Thank you for merging!" ]

[]

[]

[]

[ "We now have the `align_labels_with_mapping` method in the API for this purpose." ]

[ { "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\n200+ data loaders is a lot. Have you tried to reduce the number of datasets by concatenating/interleaving the ones with the same structure/task (the API is `{concatenate_datasets/interleave_datasets}([dset1, ..., dset_N])`)?", "Hi @mariosasko, thank you for your reply. I tried pre-concatenating different datasets into one, but one key need is to keep each batch the same data type. Considering that the concatenate-then-segment operation for prefetched samples may span across different data types after concatenating/interleaving (cuz different data sources are mixed), any solution to remain the same data source for each batch?", "@cyk1337 have you found any solutions to it?\r\n@mariosasko I tried with interleave_datasets to sample batches from two large datasets (wikipedia alike) and it results in out-of-memory error during data loading (16gpus, >1TB physical memory). Do you have any idea about it?" ]

[]

[ "Hi! The link pointing to the code that generated the dataset is broken. Can you please fix it to make debugging easier?", "> Hi! The link pointing to the code that generated the dataset is broken. Can you please fix it to make debugging easier?\r\n\r\nSorry about that, it's fixed now.\r\n" ]

[]

[]

[]

[]

[]

[]

[]

[ "How to resolve these failed checks?", "Thanks for adding this one :) \r\n\r\nTo fix the check_code_quality, please run `make style` with the latest version of black, isort, flake8\r\nTo fix the test_no_encoding_on_file_open, make sure to specify the encoding each time you call `open()` on a text file.\r\nFor example : `encoding=\"utf-8\"`\r\nTo fix the test_load_dataset_pec , you must add the dummy_data.zip file. It is used to test the dataset script and make sure it runs fine. To add it, please refer to the steps in https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-add-a-dataset\r\n\r\n", "Could you also add a dataset card ? you can find a template here : https://github.com/huggingface/datasets/blob/master/templates/README.md\r\n\r\nThat would be awesome", "> Thanks for adding this one :)\r\n> \r\n> To fix the check_code_quality, please run `make style` with the latest version of black, isort, flake8\r\n> To fix the test_no_encoding_on_file_open, make sure to specify the encoding each time you call `open()` on a text file.\r\n> For example : `encoding=\"utf-8\"`\r\n> To fix the test_load_dataset_pec , you must add the dummy_data.zip file. It is used to test the dataset script and make sure it runs fine. To add it, please refer to the steps in https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-add-a-dataset\r\n\r\nThank you for the detailed suggestion.\r\n\r\nI have added dummy_data but it still failed the DistributedDatasetTest check. My dataset has a central file (containing a python dict) that needs to be accessed by each data example. Is it because the central file cannot be distributed (which would lead to a partial dictionary)?\r\n\r\nSpecifically, the central file contains a dictionary of speakers with their attributes. Each data example is also associated with a speaker. As of now, I keep the central file and data files separately. If I remove the central file by appending the speaker attributes to each data example, then there would be lots of redundancy because there are lots of duplicate speakers in the data files.", "The `DistributedDatasetTest` fail and the changes of this PR are not related, there was just a bug in the CI. You can ignore it", "> Really cool thanks !\r\n> \r\n> Could you make the dummy files smaller ? For example by reducing the size of persona.txt ?\r\n> I also left a comment about the files concatenation. It would be cool to replace that with simple iterations through the different files.\r\n> \r\n> Then once this is done, you can add a dataset card using the template guide here : https://github.com/huggingface/datasets/blob/master/templates/README_guide.md\r\n> If some fields can't be filled, just leave `[N/A]`\r\n\r\nSmall change: if you don't have the information for a field, please leave `[More Information Needed]` rather than `[N/A]`\r\n\r\nThe full information can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md#manually-tag-the-dataset-and-write-the-dataset-card)" ]

[ { "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" } ]

[ "Hi ! It's simply not added yet :)\r\n\r\nIf someone wants to contribute to add the `stratify` parameter I'd be happy to give some pointers.\r\n\r\nIn the meantime, I guess you can use `sklearn` or other tools to do a stratified train/test split over the **indices** of your dataset and then do\r\n```\r\ntrain_dataset = dataset.select(train_indices)\r\ntest_dataset = dataset.select(test_indices)\r\n```", "Hi @lhoestq I would like to add `stratify` parameter, can you give me some pointers for adding the same ?", "Hi ! Sure :)\r\n\r\nThe `train_test_split` method is defined here: \r\n\r\nhttps://github.com/huggingface/datasets/blob/dc62232fa1b3bcfe2fbddcb721f2d141f8908943/src/datasets/arrow_dataset.py#L3253-L3253\r\n\r\nand inside `train_test_split ` we need to create the right `train_indices` and `test_indices` that are passed here to `.select()`:\r\n\r\nhttps://github.com/huggingface/datasets/blob/dc62232fa1b3bcfe2fbddcb721f2d141f8908943/src/datasets/arrow_dataset.py#L3450-L3464\r\n\r\nFor example if your dataset is like\r\n| | label |\r\n|---:|--------:|\r\n| 0 | 1 |\r\n| 1 | 1 |\r\n| 2 | 0 |\r\n| 3 | 0 |\r\n\r\nand the user passes `stratify=dataset[\"label\"]`, then you should get indices that look like this\r\n```\r\ntrain_indices = [0, 2]\r\ntest_indices = [1, 3]\r\n```\r\n\r\nthese indices will be passed to `.select` to return the stratified train and test splits :)\r\n\r\nFeel free to îng me if you have any question !", "@lhoestq \r\nI just added the implementation for `stratify` option here #4322 " ]

[ { "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 ! Interesting :) What kind of new types would you like to use ?\r\n\r\nNote that you can already implement your own decoding by using `set_transform` that can decode data on-the-fly when rows are accessed", "An interesting proposal indeed. \r\n\r\nPandas and Polars have the \"extension API\", so doing something similar on our side could be useful, too. However, this requires defining a common interface for the existing feature types before discussing the API/workflow for defining/sharing custom feature types, and this could take some time.\r\n\r\nIt would also be nice if the datasets viewer could render these custom types.", "Thank you for your replies! @lhoestq I have a use case involving whole-slide images in digital pathology. These are very large images (potentially gigapixel scale), so standard image tools are not suitable. Essentially, encoding/decoding can be done from/to [`OpenSlide`](https://openslide.org/api/python/) objects. Though there may be interest in this use case from the digital pathology community, it may not be sufficiently useful to suggest adding the feature type, but there will likely be many other use cases for a generic custom feature type.\r\n\r\nThank you for pointing out `set_transform`! I will make sure to keep this in mind in the future.\r\n\r\n@mariosasko An \"extension API\" sounds like a good idea, though I understand that this needs to be properly defined, and that you will need to discuss it internally. Support from the viewer would be awesome, too, though the generalization to arbitrary types sounds challenging.\r\n\r\nFor now, happy to know that you're considering the feature. Feel free to let me know if I can do anything to support the process." ]

[]

[]

[ { "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" } ]

[ "very much related: https://github.com/huggingface/datasets/pull/2333", "Hi @VictorSanh, thank you for reporting this issue with duplicated keys.\r\n\r\n- The issue with \"adversarial_qa\" was fixed 23 days ago: #2433. Current version of `datasets` (1.8.0) includes the patch.\r\n- I am investigating the issue with `drop`. I'll ping you to keep you informed.", "Hi @VictorSanh, the issue is already fixed and merged into master branch and will be included in our next release version 1.9.0.", "thank you!" ]

[]

[ "Hi! You can get this behavior by specifying `stratify_by_column=\"label\"` in `train_test_split`.\r\n\r\nThis is the full example:\r\n```python\r\nimport numpy as np\r\nfrom datasets import Dataset, ClassLabel\r\n\r\ndata = [\r\n {'label': 0, 'text': \"example1\"},\r\n {'label': 1, 'text': \"example2\"},\r\n {'label': 1, 'text': \"example3\"},\r\n {'label': 1, 'text': \"example4\"},\r\n {'label': 0, 'text': \"example5\"},\r\n {'label': 1, 'text': \"example6\"},\r\n {'label': 2, 'text': \"example7\"},\r\n {'label': 2, 'text': \"example8\"}\r\n]\r\n\r\nfor _ in range(10):\r\n data_set = Dataset.from_list(data)\r\n data_set = data_set.cast_column(\"label\", ClassLabel(num_classes=3))\r\n data_set = data_set.train_test_split(test_size=0.5, stratify_by_column=\"label\")\r\n unique_labels_train = np.unique(data_set[\"train\"][:][\"label\"])\r\n unique_labels_test = np.unique(data_set[\"test\"][:][\"label\"])\r\n assert len(unique_labels_train) >= len(unique_labels_test) \r\n```\r\n" ]

[]

[ "Thanks!" ]

[ { "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": "bfdadc", "default": false, "description": "Vision datasets", "id": 3608941089, "name": "vision", "node_id": "LA_kwDODunzps7XHBIh", "url": "https://api.github.com/repos/huggingface/datasets/labels/vision" } ]

[ "Awesome ! Let us know if you have questions or if we can help ;) I'm assigning you\r\n\r\nPS: if possible, please try to not use Google Drive links in your dataset script, since Google Drive has download quotas and is not always reliable.", "HI, I was waiting to come out in the second version to do the implementation.\r\n\r\n- Paper: https://dx.doi.org/10.2139/ssrn.4126881\r\n- Data: [Data mendelay](http://doi.org/10.17632/wg4bpm33hj.2)", "Nice ! 🚀 ", "The link of CCAgT dataset is: https://huggingface.co/datasets/lapix/CCAgT" ]

[ { "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 @jacobbieker, thanks for your request and study of possible alternatives.\r\n\r\nWe are very interested in finding a way to make `datasets` useful to you.\r\n\r\nLooking at the Zarr docs, I saw that among its storage alternatives, there is the ZIP file format: https://zarr.readthedocs.io/en/stable/api/storage.html#zarr.storage.ZipStore\r\n\r\nThis might be convenient for many reasons:\r\n- On the one hand, we avoid the Git issue with huge number of small files: chunks files are compressed into a single ZIP file\r\n- On the other hand, the ZIP file format is specially suited for streaming data because it allows random access to its component files (i.e. it supports random access to its chunks)\r\n\r\nAnyway, I think that a Python loading script will be necessary: you need to implement additional logic to select certain chunks (based on date or other criteria).\r\n\r\nPlease, let me know if this makes sense to you.", "Ah okay, I missed the option of zip files for zarr, I'll try that with our repos and see if it works! Thanks a lot!", "Hi @jacobbieker, does the Zarr ZipStore work for your use case?", "Hi,\r\n\r\nYes, it seems to! I got it working for https://huggingface.co/datasets/openclimatefix/mrms thanks for the help! ", "On behalf of the Zarr developers, let me say THANK YOU for working to support Zarr on HF! 🙏 Zarr is a 100% open-source and community driven project (fiscally sponsored by NumFocus). We see it as an ideal format for ML training datasets, particularly in scientific domains.\r\n\r\nI think the solution of zipping the Zarr store is a reasonable way to balance the constraints of Git LFS with the structure of Zarr.\r\n\r\nIt would be amazing to get something on the [Hugging Face Datasets Docs](https://huggingface.co/docs/datasets/index) about how to best work with Zarr. Let me know if there's a way I could help with that effort.", "Also just noting here that I was able to lazily open @jacobbieker's dataset over the internet from HF hub 🚀 !\r\n\r\n```python\r\nimport xarray as xr\r\nurl = \"https://huggingface.co/datasets/openclimatefix/mrms/resolve/main/data/2016_001.zarr.zip\"\r\nzip_url = 'zip:///::' + url\r\nds = xr.open_dataset(zip_url, engine='zarr', chunks={})\r\n```\r\n\r\n<img width=\"740\" alt=\"image\" src=\"https://user-images.githubusercontent.com/1197350/164508663-bc75cdc0-734d-44f4-9562-2877ecfdf433.png\">\r\n", "However, I wasn't able to get streaming working using the Datasets api:\r\n\r\n```python\r\nfrom datasets import load_dataset\r\nds = load_dataset(\"openclimatefix/mrms\", streaming=True, split='train')\r\nitem = next(iter(ds))\r\n```\r\n\r\n<details>\r\n<summary>FileNotFoundError traceback</summary>\r\n\r\n```\r\nNo config specified, defaulting to: mrms/2021\r\nzip://::https://huggingface.co/datasets/openclimatefix/mrms/resolve/main/data/2016_001.zarr.zip\r\ndata/2016_001.zarr.zip\r\nzip://2016_001.zarr.zip::https://huggingface.co/datasets/openclimatefix/mrms/resolve/main/data/2016_001.zarr.zip\r\n---------------------------------------------------------------------------\r\nFileNotFoundError Traceback (most recent call last)\r\nInput In [1], in <cell line: 3>()\r\n 1 from datasets import load_dataset\r\n 2 ds = load_dataset(\"openclimatefix/mrms\", streaming=True, split='train')\r\n----> 3 item = next(iter(ds))\r\n\r\nFile /opt/miniconda3/envs/hugginface/lib/python3.9/site-packages/datasets/iterable_dataset.py:497, in IterableDataset.__iter__(self)\r\n 496 def __iter__(self):\r\n--> 497 for key, example in self._iter():\r\n 498 if self.features:\r\n 499 # we encode the example for ClassLabel feature types for example\r\n 500 encoded_example = self.features.encode_example(example)\r\n\r\nFile /opt/miniconda3/envs/hugginface/lib/python3.9/site-packages/datasets/iterable_dataset.py:494, in IterableDataset._iter(self)\r\n 492 else:\r\n 493 ex_iterable = self._ex_iterable\r\n--> 494 yield from ex_iterable\r\n\r\nFile /opt/miniconda3/envs/hugginface/lib/python3.9/site-packages/datasets/iterable_dataset.py:87, in ExamplesIterable.__iter__(self)\r\n 86 def __iter__(self):\r\n---> 87 yield from self.generate_examples_fn(**self.kwargs)\r\n\r\nFile ~/.cache/huggingface/modules/datasets_modules/datasets/openclimatefix--mrms/2a6f697014d7eb3caf586ca137d47ca38785ae2fe36248611b021f8248b59936/mrms.py:150, in MRMS._generate_examples(self, filepath, split)\r\n 147 filepath = \"[https://huggingface.co/datasets/openclimatefix/mrms/resolve/main/data/2016_001.zarr.zip](https://huggingface.co/datasets/openclimatefix/mrms/resolve/main/data/2016_001.zarr.zip%3C/span%3E%3Cspan) style=\"color:rgb(175,0,0)\">\"\r\n 148 # TODO: This method handles input defined in _split_generators to yield (key, example) tuples from the dataset.\r\n 149 # The `key` is for legacy reasons (tfds) and is not important in itself, but must be unique for each example.\r\n--> 150 with zarr.storage.FSStore(fsspec.open(\"zip::\" + filepath, mode='r'), mode='r') as store:\r\n 151 data = xr.open_zarr(store)\r\n 152 for key, row in enumerate(data[\"time\"].values):\r\n\r\nFile /opt/miniconda3/envs/hugginface/lib/python3.9/site-packages/zarr/storage.py:1120, in FSStore.__init__(self, url, normalize_keys, key_separator, mode, exceptions, dimension_separator, **storage_options)\r\n 1117 import fsspec\r\n 1118 self.normalize_keys = normalize_keys\r\n-> 1120 protocol, _ = fsspec.core.split_protocol(url)\r\n 1121 # set auto_mkdir to True for local file system\r\n 1122 if protocol in (None, \"file\") and not storage_options.get(\"auto_mkdir\"):\r\n\r\nFile /opt/miniconda3/envs/hugginface/lib/python3.9/site-packages/fsspec/core.py:514, in split_protocol(urlpath)\r\n 512 def split_protocol(urlpath):\r\n 513 \"\"\"Return protocol, path pair\"\"\"\r\n--> 514 urlpath = stringify_path(urlpath)\r\n 515 if \"://\" in urlpath:\r\n 516 protocol, path = urlpath.split(\"://\", 1)\r\n\r\nFile /opt/miniconda3/envs/hugginface/lib/python3.9/site-packages/fsspec/utils.py:315, in stringify_path(filepath)\r\n 313 return filepath\r\n 314 elif hasattr(filepath, \"__fspath__\"):\r\n--> 315 return filepath.__fspath__()\r\n 316 elif isinstance(filepath, pathlib.Path):\r\n 317 return str(filepath)\r\n\r\nFile /opt/miniconda3/envs/hugginface/lib/python3.9/site-packages/fsspec/core.py:98, in OpenFile.__fspath__(self)\r\n 96 def __fspath__(self):\r\n 97 # may raise if cannot be resolved to local file\r\n---> 98 return self.open().__fspath__()\r\n\r\nFile /opt/miniconda3/envs/hugginface/lib/python3.9/site-packages/fsspec/core.py:140, in OpenFile.open(self)\r\n 132 def open(self):\r\n 133 \"\"\"Materialise this as a real open file without context\r\n 134 \r\n 135 The file should be explicitly closed to avoid enclosed file\r\n (...)\r\n 138 been deleted; but a with-context is better style.\r\n 139 \"\"\"\r\n--> 140 out = self.__enter__()\r\n 141 closer = out.close\r\n 142 fobjects = self.fobjects.copy()[:-1]\r\n\r\nFile /opt/miniconda3/envs/hugginface/lib/python3.9/site-packages/fsspec/core.py:103, in OpenFile.__enter__(self)\r\n 100 def __enter__(self):\r\n 101 mode = self.mode.replace(\"t\", \"\").replace(\"b\", \"\") + \"b\"\r\n--> 103 f = self.fs.open(self.path, mode=mode)\r\n 105 self.fobjects = [f]\r\n 107 if self.compression is not None:\r\n\r\nFile /opt/miniconda3/envs/hugginface/lib/python3.9/site-packages/fsspec/spec.py:1009, in AbstractFileSystem.open(self, path, mode, block_size, cache_options, compression, **kwargs)\r\n 1007 else:\r\n 1008 ac = kwargs.pop(\"autocommit\", not self._intrans)\r\n-> 1009 f = self._open(\r\n 1010 path,\r\n 1011 mode=mode,\r\n 1012 block_size=block_size,\r\n 1013 autocommit=ac,\r\n 1014 cache_options=cache_options,\r\n 1015 **kwargs,\r\n 1016 )\r\n 1017 if compression is not None:\r\n 1018 from fsspec.compression import compr\r\n\r\nFile /opt/miniconda3/envs/hugginface/lib/python3.9/site-packages/fsspec/implementations/zip.py:96, in ZipFileSystem._open(self, path, mode, block_size, autocommit, cache_options, **kwargs)\r\n 94 if mode != \"rb\":\r\n 95 raise NotImplementedError\r\n---> 96 info = self.info(path)\r\n 97 out = self.zip.open(path, \"r\")\r\n 98 out.size = info[\"size\"]\r\n\r\nFile /opt/miniconda3/envs/hugginface/lib/python3.9/site-packages/fsspec/archive.py:42, in AbstractArchiveFileSystem.info(self, path, **kwargs)\r\n 40 return self.dir_cache[path + \"/\"]\r\n 41 else:\r\n---> 42 raise FileNotFoundError(path)\r\n\r\nFileNotFoundError:\r\n```\r\n\r\n</details>\r\n\r\nIs this a bug? Or am I just doing it wrong...", "I'm still messing around with that dataset, so the data might have moved. I currently have each year of MRMS precipitation rate data as it's own zarr, but as they are quite large (on order of 100GB each) I'm working to split them into single days, and as such they are still being moved around, I was just trying to get a proof of concept working originally. ", "I've mostly finished rearranging the data now and uploading some more, so this works now:\r\n```python\r\nimport datasets\r\nds = datasets.load_dataset(\"openclimatefix/mrms\", streaming=True, split=\"train\")\r\nitem = next(iter(ds))\r\nprint(item.keys())\r\nprint(item[\"timestamp\"])\r\n```\r\n\r\nThe MRMS data now goes most of 2016-2022, with quite a few gaps I'm working on filling in" ]

[ { "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\nCould you tell us how you are running this code?\r\nI tested on my machine (M1 Mac). And it is running fine both on and off internet.\r\n\r\n<img width=\"1033\" alt=\"Screen Shot 2022-07-03 at 1 32 25 AM\" src=\"https://user-images.githubusercontent.com/8711912/177026364-4ad7cedb-e524-4513-97f7-7961bbb34c90.png\">\r\nTested on both stable and dev version. ", "Sure, I was running it on a Linux machine.\r\nI found that if I turn the Internet off, it would still try to make a HTTPS call which would slow down the cache loading. If you can't reproduce then we can close the issue.", "Hi @Muennighoff! You can set the env variable `HF_DATASETS_OFFLINE` to `1` to avoid this behavior in offline mode. More info is available [here](https://huggingface.co/docs/datasets/master/en/loading#offline)." ]

[]

[ "Yes, having `cache_dir` as an attribute looks cleaner.\r\n\r\n\r\n\r\n", "Good job! Looking forward to this new feature! 🥂", "@lhoestq Sorry for the late reply. Yes, I'll start working on tests. Thanks for the detailed explanation of the current issues with caching (like the idea of adding the `use_caching` parameter to `load_dataset`) ", "@lhoestq Sure. I'm aware this is a high-priority issue to some extent, so feel free to take over.\r\n\r\nFew suggestions I have:\r\n* there is a slight difference between setting `use_caching` to `False` in `load_dataset` and disabling caching globally with `set_caching_enabled(False)` because the former will never execute the following code (`self._cache_dir` is always `False`): \r\nhttps://github.com/huggingface/datasets/blob/c231abdb174987419bbde3360b5b9d6a4672c736/src/datasets/arrow_dataset.py#L1807-L1824\r\n, so I'm just checking whether this is intended (if yes, maybe the docs should mention this) or not?\r\n* think we should add the `use_caching` parameter to every method that has the `keep_in_memory` (and `in_memory` 😃) parameter in its signature for better consistency, but I say let's address this in a separate PR. IMO we need one more PR that will deal exclusively with consistency in the caching logic.", "Hi @mariosasko \r\nWe discussed internally and we think that this feature might not be the direction we're doing to take for these reasons:\r\n\r\n- it goes against our simple definition of caching: on-disk == uses file cache, and in-memory == nothing is written to disk. I think it adds too much complexity just for a minimal flexibility addition\r\n- there are a few edge cases which are really confusing:\r\n - map on an in memory dataset with a cache_file_name specified by the user -> should the result be in memory or from disk ?\r\n - it would require a special cache directory just for in memory datasets, since they don’t have a preferred directory for caching\r\n- it would break a lot of stuff and would require to rewrite significant parts of the core code and the tests\r\n\r\n\r\nSo in the end we're probably going to close this PR.\r\nLet me know what you think, and thank you anyway for your help on this !", "Hi,\r\n\r\nI'm fine with that. I agree this adds too much complexity. Btw, I like the idea of reverting default in-memory for small datasets that led to this PR.", "Superseded by #2460 (to close issue #2458)." ]

[]

[]

[]

[ "It's so cool seeing all these datasets fly by and see how they are still of interest. I did my internship at the research group of Liesbeth Augustinus et al. They're a very kind group of people!", "merging since the CI is fixed on master" ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "<details>\n<summary>Show benchmarks</summary>\n\nPyArrow==6.0.0\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) | 0.011875 / 0.011353 (0.000522) | 0.008188 / 0.011008 (-0.002821) | 0.131137 / 0.038508 (0.092629) | 0.038127 / 0.023109 (0.015018) | 0.383864 / 0.275898 (0.107966) | 0.458617 / 0.323480 (0.135137) | 0.010989 / 0.007986 (0.003003) | 0.004892 / 0.004328 (0.000563) | 0.101955 / 0.004250 (0.097704) | 0.045081 / 0.037052 (0.008029) | 0.409768 / 0.258489 (0.151279) | 0.446597 / 0.293841 (0.152756) | 0.058588 / 0.128546 (-0.069958) | 0.020872 / 0.075646 (-0.054774) | 0.432982 / 0.419271 (0.013711) | 0.075875 / 0.043533 (0.032342) | 0.380923 / 0.255139 (0.125784) | 0.432994 / 0.283200 (0.149795) | 0.122678 / 0.141683 (-0.019005) | 1.857865 / 1.452155 (0.405710) | 1.927801 / 1.492716 (0.435085) |\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) | 0.212941 / 0.018006 (0.194935) | 0.527977 / 0.000490 (0.527488) | 0.002996 / 0.000200 (0.002797) | 0.000105 / 0.000054 (0.000051) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.030046 / 0.037411 (-0.007366) | 0.126384 / 0.014526 (0.111858) | 0.138307 / 0.176557 (-0.038250) | 0.185338 / 0.737135 (-0.551797) | 0.144733 / 0.296338 (-0.151606) |\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) | 0.627096 / 0.215209 (0.411887) | 6.418014 / 2.077655 (4.340360) | 2.547675 / 1.504120 (1.043555) | 2.195552 / 1.541195 (0.654357) | 2.200377 / 1.468490 (0.731887) | 1.289935 / 4.584777 (-3.294842) | 5.670839 / 3.745712 (1.925127) | 5.252597 / 5.269862 (-0.017265) | 2.878470 / 4.565676 (-1.687207) | 0.143754 / 0.424275 (-0.280521) | 0.014814 / 0.007607 (0.007207) | 0.810073 / 0.226044 (0.584028) | 8.183757 / 2.268929 (5.914829) | 3.375525 / 55.444624 (-52.069099) | 2.594048 / 6.876477 (-4.282428) | 2.598095 / 2.142072 (0.456023) | 1.554493 / 4.805227 (-3.250734) | 0.263159 / 6.500664 (-6.237505) | 0.089822 / 0.075469 (0.014353) |\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) | 1.660847 / 1.841788 (-0.180941) | 18.434283 / 8.074308 (10.359975) | 21.764887 / 10.191392 (11.573495) | 0.264524 / 0.680424 (-0.415900) | 0.048519 / 0.534201 (-0.485682) | 0.587468 / 0.579283 (0.008185) | 0.634142 / 0.434364 (0.199778) | 0.675374 / 0.540337 (0.135037) | 0.777510 / 1.386936 (-0.609426) |\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) | 0.010021 / 0.011353 (-0.001332) | 0.006207 / 0.011008 (-0.004801) | 0.130490 / 0.038508 (0.091982) | 0.037957 / 0.023109 (0.014848) | 0.489381 / 0.275898 (0.213483) | 0.536522 / 0.323480 (0.213042) | 0.008611 / 0.007986 (0.000626) | 0.004894 / 0.004328 (0.000565) | 0.101617 / 0.004250 (0.097367) | 0.052629 / 0.037052 (0.015577) | 0.509211 / 0.258489 (0.250721) | 0.545023 / 0.293841 (0.251182) | 0.057468 / 0.128546 (-0.071078) | 0.023393 / 0.075646 (-0.052253) | 0.431408 / 0.419271 (0.012137) | 0.064967 / 0.043533 (0.021434) | 0.495261 / 0.255139 (0.240122) | 0.527098 / 0.283200 (0.243898) | 0.113172 / 0.141683 (-0.028511) | 1.937072 / 1.452155 (0.484918) | 2.048413 / 1.492716 (0.555697) |\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) | 0.245406 / 0.018006 (0.227399) | 0.526772 / 0.000490 (0.526283) | 0.004379 / 0.000200 (0.004179) | 0.000114 / 0.000054 (0.000060) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.031785 / 0.037411 (-0.005626) | 0.130949 / 0.014526 (0.116424) | 0.145660 / 0.176557 (-0.030896) | 0.186991 / 0.737135 (-0.550144) | 0.151000 / 0.296338 (-0.145338) |\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) | 0.708643 / 0.215209 (0.493434) | 7.179252 / 2.077655 (5.101597) | 3.143375 / 1.504120 (1.639255) | 2.714298 / 1.541195 (1.173103) | 2.773441 / 1.468490 (1.304951) | 1.312821 / 4.584777 (-3.271956) | 5.798396 / 3.745712 (2.052684) | 3.253215 / 5.269862 (-2.016646) | 2.147260 / 4.565676 (-2.418416) | 0.154673 / 0.424275 (-0.269602) | 0.014918 / 0.007607 (0.007311) | 0.860618 / 0.226044 (0.634573) | 8.774455 / 2.268929 (6.505527) | 3.925020 / 55.444624 (-51.519604) | 3.139361 / 6.876477 (-3.737115) | 3.208883 / 2.142072 (1.066810) | 1.547305 / 4.805227 (-3.257922) | 0.268814 / 6.500664 (-6.231850) | 0.084578 / 0.075469 (0.009109) |\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) | 1.694990 / 1.841788 (-0.146798) | 18.619183 / 8.074308 (10.544875) | 21.929886 / 10.191392 (11.738494) | 0.265763 / 0.680424 (-0.414661) | 0.028325 / 0.534201 (-0.505876) | 0.552910 / 0.579283 (-0.026373) | 0.616864 / 0.434364 (0.182500) | 0.637858 / 0.540337 (0.097521) | 0.744508 / 1.386936 (-0.642428) |\n\n</details>\n</details>\n\n![](https://cml.dev/watermark.png#5f819ba3d0306748aaf9fd8ea040b981dd08e5e5 \"CML watermark\")\n" ]

[]