Datasets:

---

sphyuung

/

datasets-github-issues

like 0

[]

[]

[]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7234). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.", "merging this one for now, let me know if you'd like to see additional changes for error handling (I'll take care of them before doing a release)" ]

[]

[]

[]

[]

[]

[ "You are welcome!" ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7231). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7230). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[ "not sure it's a good idea, we always need a config name so better have the correct default and not support None (which could lead to think it doesn't have a config name, while it does)" ]

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

[]

[]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7227). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.", "I've updated the most straightforward failing test cases - lmk if you agree with those.\r\n\r\nMight need some help / pointers on the remaining new failing tests, which seem a little bit more subtle.", "@lhoestq I've had a go at fixing a few more test cases but getting quite uncertain about the remaining ones (as well as about some of the array writing ones that I tried to fix in my last commit). There are still 27 failures vs 21 on main. I'm not completely sure in some cases what intended behaviour is and my understanding of the flow for typed writing is a bit vague." ]

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

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[ "I am encountering this same issue. It seems that the library manages to recognise an optional column (but not **exclusively** null) if there is at least one non-null instance within the same file. For example, given a `test_0.jsonl` file:\r\n```json\r\n{\"a\": \"a1\", \"b\": \"b1\", \"c\": null, \"d\": null}\r\n{\"a\": \"a2\", \"b\": null, \"c\": \"c2\", \"d\": null}\r\n```\r\nthe data is correctly loaded, recognising that columns `b` & `c` are optional, while `d` is null.\r\n```python\r\n{'a': ['a1', 'a2'], 'b': ['b1', None], 'c': [None, 'c2'], 'd': [None, None]}\r\n```\r\n\r\nBut if the `config` has another file, say `test_1.jsonl` where `d` now has some non-null values:\r\n```json\r\n{\"a\": null, \"b\": \"b3\", \"c\": \"c3\", \"d\": \"d3\"}\r\n{\"a\": \"a4\", \"b\": \"b4\", \"c\": null, \"d\": null}\r\n```\r\nthen, an error is raised:\r\n```\r\nTypeError Traceback (most recent call last)\r\n\r\n[/usr/local/lib/python3.10/dist-packages/datasets/builder.py](https://localhost:8080/#) in _prepare_split_single(self, gen_kwargs, fpath, file_format, max_shard_size, job_id)\r\n 1869 try:\r\n-> 1870 writer.write_table(table)\r\n 1871 except CastError as cast_error:\r\n\r\n14 frames\r\n\r\nTypeError: Couldn't cast array of type string to null\r\n\r\n\r\nThe above exception was the direct cause of the following exception:\r\n\r\nDatasetGenerationError Traceback (most recent call last)\r\n\r\n[/usr/local/lib/python3.10/dist-packages/datasets/builder.py](https://localhost:8080/#) in _prepare_split_single(self, gen_kwargs, fpath, file_format, max_shard_size, job_id)\r\n 1895 if isinstance(e, DatasetGenerationError):\r\n 1896 raise\r\n-> 1897 raise DatasetGenerationError(\"An error occurred while generating the dataset\") from e\r\n 1898 \r\n 1899 yield job_id, True, (total_num_examples, total_num_bytes, writer._features, num_shards, shard_lengths)\r\n\r\nDatasetGenerationError: An error occurred while generating the dataset\r\n```\r\n\r\n---\r\n\r\nI have created a [sample repository](https://huggingface.co/datasets/KurtMica/optional_columns_mutiple_files) if that helps. Interestingly, the dataset viewer correctly shows the data across files, although it still indicates the above error." ]

[]

[]

[ "@lhoestq would you be open to supporting this kind of extensibility?", "I suggested a fix in https://github.com/huggingface/datasets/issues/7220 that would not necessarily require a parent class for custom features, lmk what you think" ]

[]

[]

[ "I think you can fix this simply by replacing the line with hardcoded features with `hastattr(schema, \"encode_example\")` actually", "#7284 " ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7219). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[ "Hi ! `map()` reads all the columns and writes the resulting dataset with all the columns as well, while df.column_name.apply only reads and writes one column and does it in memory. So this is speed difference is actually expected.\r\n\r\nMoreover using multiprocessing on a dataset that lives in memory (from_pandas uses the same in-memory data as the pandas DataFrame while load_dataset or from_generator load from disk) requires to copy the data to each subprocess which can also be slow. Data loaded from disk don't need to be copied though since they work as a form of shared memory thanks to memory mapping.\r\n\r\nHowever you can make you map() call much faster by making it read and write only the column you want:\r\n\r\n```python\r\nhas_cover_ds = ds.map(lambda song_name: {'has_cover': has_cover(song_name)}, input_columns=[\"song_name\"], remove_columns=ds.column_names) # outputs a dataset with 1 column\r\nds = ds.concatenate_datasets([ds, has_cover_ds], axis=1)\r\n```\r\n\r\nand if your dataset is loaded from disk you can pass num_proc=10 and get a nice speed up as well (no need to copy the data to subprocesses)" ]

[]

[]

[]

[]

[]

[ "possibly due to this logic:\r\n\r\n```python\r\n def _arrow_array_to_numpy(self, pa_array: pa.Array) -> np.ndarray:\r\n if isinstance(pa_array, pa.ChunkedArray):\r\n if isinstance(pa_array.type, _ArrayXDExtensionType):\r\n # don't call to_pylist() to preserve dtype of the fixed-size array\r\n zero_copy_only = _is_zero_copy_only(pa_array.type.storage_dtype, unnest=True)\r\n array: List = [\r\n row for chunk in pa_array.chunks for row in chunk.to_numpy(zero_copy_only=zero_copy_only)\r\n ]\r\n else:\r\n zero_copy_only = _is_zero_copy_only(pa_array.type) and all(\r\n not _is_array_with_nulls(chunk) for chunk in pa_array.chunks\r\n )\r\n array: List = [\r\n row for chunk in pa_array.chunks for row in chunk.to_numpy(zero_copy_only=zero_copy_only)\r\n ]\r\n else:\r\n if isinstance(pa_array.type, _ArrayXDExtensionType):\r\n # don't call to_pylist() to preserve dtype of the fixed-size array\r\n zero_copy_only = _is_zero_copy_only(pa_array.type.storage_dtype, unnest=True)\r\n array: List = pa_array.to_numpy(zero_copy_only=zero_copy_only)\r\n else:\r\n zero_copy_only = _is_zero_copy_only(pa_array.type) and not _is_array_with_nulls(pa_array)\r\n array: List = pa_array.to_numpy(zero_copy_only=zero_copy_only).tolist()\r\n```" ]

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

[]

[]

[]

[]

[]

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

[]

[]

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

[]

[]

[]

[]

[ "Yes your assumption on concatenate/interleave is ok imo.\r\n\r\nIt seems the TypedExamplesIterable can slow down things, it should take formatting into account to not convert numpy arrays to python lists\r\n\r\nright now it's slow (unrelatedly to your PR):\r\n\r\n```python\r\n>>> ds = Dataset.from_dict({\"a\": np.zeros((1000, 32, 32))}).to_iterable_dataset().with_format(\"np\")\r\n>>> filtered_ds = ds.filter(lambda x: True)\r\n>>> %time sum(1 for _ in ds)\r\nCPU times: user 175 ms, sys: 8.1 ms, total: 183 ms\r\nWall time: 184 ms\r\n1000\r\n>>> %time sum(1 for _ in filtered_ds)\r\nCPU times: user 4.1 s, sys: 8.41 ms, total: 4.1 s\r\nWall time: 4.12 s\r\n1000\r\n```", "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7209). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.", "> It seems the TypedExamplesIterable can slow down things, it should take formatting into account to not convert numpy arrays to python lists\r\n\r\nShould be fixed by updated #7207 I hope!" ]

[]

[]

[ "closed by https://github.com/huggingface/datasets/pull/7209, thanks @alex-hh !" ]

[]

[]

[ "I think the problem is that the underlying ex_iterable will not use iter_arrow unless the formatting type is arrow, which leads to conversion from arrow -> python -> numpy in this case rather than arrow -> numpy.\r\n\r\nIdea of updated fix is to use the ex_iterable's iter_arrow in any case where it's available and any formatting is specified. The formatter then works directly on arrow tables; the outputs of the formatter get passed to the function to be mapped.\r\n\r\nWith updated version:\r\n\r\n```python\r\nimport numpy as np\r\nimport time\r\nfrom datasets import Dataset, Features, Array3D\r\n\r\nfeatures=Features(**{\"array0\": Array3D((None, 10, 10), dtype=\"float32\"), \"array1\": Array3D((None,10,10), dtype=\"float32\")})\r\ndataset = Dataset.from_dict({f\"array{i}\": [np.zeros((x,10,10), dtype=np.float32) for x in [2000,1000]*25] for i in range(2)}, features=features)\r\n```\r\n\r\n```python\r\nds = dataset.to_iterable_dataset()\r\nds = ds.with_format(\"numpy\").map(lambda x: x, batched=True, batch_size=10)\r\nt0 = time.time()\r\nfor ex in ds:\r\n pass\r\nt1 = time.time()\r\n```\r\nTotal time: < 0.01s (~30s on main)\r\n\r\n```python\r\nds = dataset.to_iterable_dataset()\r\nds = ds.with_format(\"numpy\").map(lambda x: x, batched=False)\r\nt0 = time.time()\r\nfor ex in ds:\r\n pass\r\nt1 = time.time()\r\n```\r\nTime: ~0.02 s (~30s on main)\r\n\r\n```python\r\nds = dataset.to_iterable_dataset()\r\nds = ds.with_format(\"numpy\")\r\nt0 = time.time()\r\nfor ex in ds:\r\n pass\r\nt1 = time.time()\r\n```\r\nTime: ~0.02s", "also now working for filter with similar performance improvements:\r\n\r\n```python\r\nfiltered_examples = []\r\nds = dataset.to_iterable_dataset()\r\nds = ds.with_format(\"numpy\").filter(lambda x: [arr.shape[0]==2000 for arr in x[\"array0\"]], batch_size=10, batched=True)\r\nt0 = time.time()\r\nfor ex in ds:\r\n filtered_examples.append(ex)\r\nt1 = time.time()\r\nassert len(filtered_examples) == 25\r\n```\r\n0.01s vs 50s on main\r\n\r\n\r\n```python\r\nfiltered_examples = []\r\nds = dataset.to_iterable_dataset()\r\nds = ds.with_format(\"numpy\").filter(lambda x: x[\"array0\"].shape[0]==2000, batched=False)\r\nt0 = time.time()\r\nfor ex in ds:\r\n filtered_examples.append(ex)\r\nt1 = time.time()\r\nassert len(filtered_examples) == 25\r\n```\r\n0.04s vs 50s on main\r\n", "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7207). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.", "(the distributed tests failing in the CI are unrelated)", "There also appears to be a separate? issue with chaining filter and map bc filter iter_arrow only returns _iter_arrow if arrow formatting is applied (and vv presumably)\r\n\r\nI don't have a good minimal example atm", "> issue with chaining filter and map bc filter iter_arrow only returns _iter_arrow if arrow formatting is applied (and vv presumably)\r\n\r\nMaybe related to this issue ?\r\n\r\n```python\r\nds = Dataset.from_dict({\"a\": range(10)}).to_iterable_dataset()\r\nds = ds.with_format(\"arrow\").map(lambda x: x, features=Features({\"a\": Value(\"string\")})).with_format(None)\r\nprint(list(ds)) # yields integers instead of strings\r\n```", "I feel like we could get rid of TypedExampleIterable altogether and apply formatting with feature conversion with `formatted_python_examples_iterator ` and `formatted_arrow_examples_iterator`\r\n\r\nbtw you can pass `features=` in `get_formatter()` to get a formatter that does the feature conversion at the same time as formatting\r\n\r\n(edit:\r\n\r\nexcept maybe the arrow formatter doesn't use `features` yet, we can fix it like this if it's really needed\r\n```diff\r\nclass ArrowFormatter(Formatter[pa.Table, pa.Array, pa.Table]):\r\n def format_row(self, pa_table: pa.Table) -> pa.Table:\r\n- return self.simple_arrow_extractor().extract_row(pa_table)\r\n+ pa_table = self.simple_arrow_extractor().extract_row(pa_table)\r\n+. return cast_table_to_features(pa_table, self.features) if self.features else pa_table\r\n \r\n```\r\n\r\n\r\n)", "> I feel like we could get rid of TypedExampleIterable altogether and apply formatting with feature conversion with formatted_python_examples_iterator and formatted_arrow_examples_iterator\r\n\r\nOh nice didn't know about the feature support in get_formatter. Haven't thought through whether this works but would a FormattedExampleIterable (with feature conversion) be able to solve this and fit the API better?", "> Oh nice didn't know about the feature support in get_formatter. Haven't thought through whether this works but would a FormattedExampleIterable (with feature conversion) be able to solve this and fit the API better?\r\n\r\nYes this is surely the way to go actually !", "ok i've fixed the chaining issue with my last two commits.\r\n\r\nWill see if I can refactor into a FormattedExampleIterable\r\n\r\nThe other issue you posted seems to be unrelated (maybe something to do with feature decoding?)", "updated with FormattedExamplesIterable.\r\n\r\nthere might be a few unnecessary format calls once the data is already formatted - doesn't seem like a big performance bottleneck but could maybe be fixed with e.g. an is_formatted property\r\n\r\nIt also might be possible to do a wider refactor and use FormattedExamplesIterable elsewhere. But I'd personally prefer not to try that rn.", "Thinking about this in the context of #7210 - am wondering if it would make sense for Features to define their own extraction arrow->object logic? e.g. Arrays should *always* be extracted with NumpyArrowExtractor, not only in case with_format is set to numpy (which a user can easily forget or not know to do)\r\n", "> Thinking about this in the context of https://github.com/huggingface/datasets/issues/7210 - am wondering if it would make sense for Features to define their own extraction arrow->object logic? e.g. Arrays should always be extracted with NumpyArrowExtractor, not only in case with_format is set to numpy (which a user can easily forget or not know to do)\r\n\r\nFor `ArrayND` they already implement `to_pylist` to decode arrow data and it can be updated to return a numpy array (see the `ArrayExtensionArray` class for more details)", "@lhoestq im no longer sure my specific concern about with_format(None) was well-founded - I didn't appreciate that the python formatter tries to do nothing to python objects including numpy arrays, so the existing with_format(None) should I *think* do what I want. Do you think with_format(None) is ok as is after all? If so think this is hopefully ready for final review!", "@lhoestq I've updated to make compatible with latest changes on main, and think the current with_format None behaviour is probably fine - please let me know if there's anything else I can do!", "Hi Alex, I will be less available from today and for a week. I'll review your PR and play with it once I come back if you don't mind !" ]

[]

[]

[ "The below easily eats up 32G of RAM. Leaving it for a while bricked the laptop with 16GB.\r\n\r\n```\r\ndataset = load_dataset(\"Voxel51/OxfordFlowers102\", data_dir=\"data\").with_format(\"numpy\")\r\nprocessed_dataset = dataset.map(lambda x: x)\r\n```\r\n\r\n\r\n\r\nSimilar problems occur if using a real transform function in `.map()`." ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7205). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7204). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7203). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "Hi ! yes now `datasets` uses `huggingface_hub` to download and cache files from the HF Hub so you need to use `HF_HOME` (or manually `HF_HUB_CACHE` and `HF_DATASETS_CACHE` if you want to separate HF Hub cached files and cached datasets Arrow files)\r\n\r\nSo in your change I guess it needs to be `HF_HOME` instead of `HF_CACHE` ?", "Thank you for your comment. You are right. I am sorry for my mistake, I fixed it.", "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7200). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.", "I just had this issue, and needed to move the setting the env code in the python file to top, before the import of the lib \r\nie. \r\n```python\r\nimport os\r\nLOCAL_DISK_MOUNT = '/mnt/data'\r\n\r\nos.environ['HF_HOME'] = f'{LOCAL_DISK_MOUNT}/hf_cache/'\r\nos.environ['HF_DATASETS_CACHE'] = f'{LOCAL_DISK_MOUNT}/datasets/'\r\n\r\nfrom datasets import load_dataset\r\nfrom datasets import load_dataset_builder\r\nfrom psutil._common import bytes2human\r\n\r\n\r\n```" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "Also cant download \"allenai/c4\", but with different error reported:\r\n```\r\nTraceback (most recent call last): \r\n File \"/***/lib/python3.10/site-packages/datasets/load.py\", line 2074, in load_dataset \r\n builder_instance = load_dataset_builder( \r\n File \"/***/lib/python3.10/site-packages/datasets/load.py\", line 1795, in load_dataset_builder \r\n dataset_module = dataset_module_factory( \r\n File \"/***/lib/python3.10/site-packages/datasets/load.py\", line 1659, in dataset_module_factory \r\n raise e1 from None \r\n File \"/***/lib/python3.10/site-packages/datasets/load.py\", line 1647, in dataset_module_factory \r\n ).get_module() \r\n File \"/***/lib/python3.10/site-packages/datasets/load.py\", line 1069, in get_module \r\n module_name, default_builder_kwargs = infer_module_for_data_files( \r\n File \"/***/lib/python3.10/site-packages/datasets/load.py\", line 594, in infer_module_for_data_files \r\n raise DataFilesNotFoundError(\"No (supported) data files found\" + (f\" in {path}\" if path else \"\")) \r\ndatasets.exceptions.DataFilesNotFoundError: No (supported) data files found in allenai/c4 \r\n```\r\n\r\n## Code to reproduce\r\n```\r\ndataset = load_dataset(\"allenai/c4\", \"en\", split=\"train\", streaming=True,trust_remote_code=True,\r\n cache_dir=\"dataset/en\",\r\n download_mode=\"force_redownload\")\r\n```\r\n\r\n## Environment\r\ndatasets 3.0.1 \r\nhuggingface_hub 0.25.1" ]

[]

[]

[]

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

[ { "login": "severo", "id": 1676121, "node_id": "MDQ6VXNlcjE2NzYxMjE=", "avatar_url": "https://avatars.githubusercontent.com/u/1676121?v=4", "gravatar_id": "", "url": "https://api.github.com/users/severo", "html_url": "https://github.com/severo", "followers_url": "https://api.github.com/users/severo/followers", "following_url": "https://api.github.com/users/severo/following{/other_user}", "gists_url": "https://api.github.com/users/severo/gists{/gist_id}", "starred_url": "https://api.github.com/users/severo/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/severo/subscriptions", "organizations_url": "https://api.github.com/users/severo/orgs", "repos_url": "https://api.github.com/users/severo/repos", "events_url": "https://api.github.com/users/severo/events{/privacy}", "received_events_url": "https://api.github.com/users/severo/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "maybe related: https://github.com/huggingface/datasets/issues/6388", "Also look at https://github.com/huggingface/dataset-viewer/blob/f5fd117ceded990a7766e705bba1203fa907d6ad/services/worker/src/worker/job_runners/dataset/modalities.py#L241 which lists the 3D file formats that will assign the 3D modality to a dataset.", "~~we can brainstorm about the UX maybe (i don't expect we should load all models on the page at once – IMO there should be a manual action from user to load + maybe load first couple of row by default) cc @gary149 @cfahlgren1~~\r\n\r\nit's more for the viewer issue (https://github.com/huggingface/dataset-viewer/issues/1003)" ]

[]

[]

[ "Actually there is no such dataset available, that is why you are getting that error.", "Fixed with @kdutia in Slack chat. Generating a new token fixed this issue. " ]

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

[]

[ "I was curious about the same - since map is applied on the fly I was assuming that setting num_workers>1 in DataLoader would effectively do the map in parallel, have you tried that?" ]

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

[]

[ "perhaps concatenate_datasets can already be used to achieve almost the same effect? ", "`concatenate_datasets` does the job when there is a finite number of repetitions, but in case of `.repeat()` forever we need a new logic in `iterable_dataset.py`" ]

[]

[]

[ "I think the approach presented in https://github.com/huggingface/datasets/pull/6832 is the one we'll be taking.\r\n\r\nAsking user input is not a good idea since `load_dataset` is used a lot in server that don't have someone in front of them to select a split" ]

[]

[]

[ "Yes, I need to use the latest version of fsspec and datasets for my usecase. \r\nhttps://github.com/fsspec/s3fs/pull/888#issuecomment-2404204606\r\nhttps://github.com/apache/arrow/issues/34363#issuecomment-2403553473\r\n\r\nlast version where things install without conflict is: 2.14.4\r\n\r\nSo this issue starts from:\r\nhttps://github.com/huggingface/datasets/releases/tag/2.14.5" ]

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

[]

[]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7188). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[]

[]

[ { "login": "albertvillanova", "id": 8515462, "node_id": "MDQ6VXNlcjg1MTU0NjI=", "avatar_url": "https://avatars.githubusercontent.com/u/8515462?v=4", "gravatar_id": "", "url": "https://api.github.com/users/albertvillanova", "html_url": "https://github.com/albertvillanova", "followers_url": "https://api.github.com/users/albertvillanova/followers", "following_url": "https://api.github.com/users/albertvillanova/following{/other_user}", "gists_url": "https://api.github.com/users/albertvillanova/gists{/gist_id}", "starred_url": "https://api.github.com/users/albertvillanova/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/albertvillanova/subscriptions", "organizations_url": "https://api.github.com/users/albertvillanova/orgs", "repos_url": "https://api.github.com/users/albertvillanova/repos", "events_url": "https://api.github.com/users/albertvillanova/events{/privacy}", "received_events_url": "https://api.github.com/users/albertvillanova/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[]

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

[]

[ "Fixed by #7205" ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7184). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7182). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.", "The CI issue is unrelated:\r\n- #7183" ]

[]

[]

[ "Linked Issue: #7037\r\nIdeas: #7039 ", "@albertvillanova / @lhoestq any early feedback?\r\n\r\nAFAIK there is no param `orient` in `load_dataset()`. So for orientations other than \"records\", the loading isn't very accurate. Any thoughts?", "`orient = \"split\"` can also be handled. I will add the changes soon", "Thanks for diving into this ! I don't think we want the JSON export to be that complex though, especially if people can do `ds.to_pandas().to_json(orient=...)`. Maybe we can just raise an error and suggest users to use pandas ? And also note that it loads the full dataset in memory so it's mainly for small scale datasets. The only acceptable option for large scale datasets is probably just JSON Lines anyway since it enables streaming.", "@lhoestq Simply doing `ds.to_pandas().to_json(orient=...)` is not going to give any batching or multiprocessing benefits right? Also, which function are you referring to - when you say that its meant for small scale datasets only?", "Yes indeed. Though I think it's fine since using something else than orient=\"lines\" is only suitable/useful for small datasets. Or you know a case where a big dataset need to be in a format that is not orient=\"lines\" ?", "@lhoestq Let me close this PR and open another one where I will add an error message, as suggested here.\r\n\r\n> Thanks for diving into this ! I don't think we want the JSON export to be that complex though, especially if people can do `ds.to_pandas().to_json(orient=...)`. Maybe we can just raise an error and suggest users to use pandas ? And also note that it loads the full dataset in memory so it's mainly for small scale datasets. The only acceptable option for large scale datasets is probably just JSON Lines anyway since it enables streaming.\r\n\r\n", "Addressed here: #7273 \r\n@lhoestq " ]

[]

[]

[ "> I've encountered a memory leak when wrapping the HuggingFace dataset into a PyTorch Dataset. The RAM usage constantly increases during iteration if items are not explicitly deleted after use.\r\n\r\nDatasets are memory mapped so they work like SWAP memory. In particular as long as you have RAM available the data will stay in RAM, and get paged out once your system needs RAM for something else (no OOM).\r\n\r\nrelated: https://github.com/huggingface/datasets/issues/4883" ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7179). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

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

[ { "login": "albertvillanova", "id": 8515462, "node_id": "MDQ6VXNlcjg1MTU0NjI=", "avatar_url": "https://avatars.githubusercontent.com/u/8515462?v=4", "gravatar_id": "", "url": "https://api.github.com/users/albertvillanova", "html_url": "https://github.com/albertvillanova", "followers_url": "https://api.github.com/users/albertvillanova/followers", "following_url": "https://api.github.com/users/albertvillanova/following{/other_user}", "gists_url": "https://api.github.com/users/albertvillanova/gists{/gist_id}", "starred_url": "https://api.github.com/users/albertvillanova/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/albertvillanova/subscriptions", "organizations_url": "https://api.github.com/users/albertvillanova/orgs", "repos_url": "https://api.github.com/users/albertvillanova/repos", "events_url": "https://api.github.com/users/albertvillanova/events{/privacy}", "received_events_url": "https://api.github.com/users/albertvillanova/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[]