Datasets:

---

Francesco-A

/

github-issues_huggingface-datasets

like 0

[]

[ "Hmmm this one broke master. I'm fixing it.\r\n\r\nMaybe because your branch was outdated ?", "Sorry @lhoestq, I forgot to update the imports... :/", "It's fine, the CI should have caught this tbh. Not sure why it did't fail" ]

[]

[ "P.S. For more context, I'm currently exploring the integration of 🤗`datasets` with JAX, so in case you need any help or want me to try something specific just let me know! (`jnp.asarray`/`jnp.array(..., copy=False)` still no zero-copy 😭)", "_The documentation is not available anymore as the PR was closed or merged._", "> Hi ! Thanks for improving this :)\r\n\r\nGlad to help, @lhoestq! Also, regarding the questions in the `## What's missing?` can I have your input? Thanks 🤗 ", "Whoops forgot to reply to these matters - sorry x)\r\n\r\nYea a JAX guide would be welcome in the documentation ! This can be done in a separate PR if you want :)\r\n\r\nPyarrow is always imported with `datasets`, so it doesn't really matter if it's under TYPE_CHECKING or not.\r\n\r\nRegarding the license : yes indeed it should be in every file, thanks for reporting.\r\n\r\nNo big preference between jnp.array and jnp.asarray, unless one offers better performance", "> Whoops forgot to reply to these matters - sorry x)\r\n> \r\n> Yea a JAX guide would be welcome in the documentation ! This can be done in a separate PR if you want :)\r\n> \r\n> Pyarrow is always imported with `datasets`, so it doesn't really matter if it's under TYPE_CHECKING or not.\r\n> \r\n> Regarding the license : yes indeed it should be in every file, thanks for reporting.\r\n> \r\n> No big preference between jnp.array and jnp.asarray, unless one offers better performance\r\n\r\nCool @lhoestq thanks for the input there!\r\n\r\n1. I can create a separate PR for JAX-format usage\r\n2. Regarding that, makes sense, we can just not put it there, unless it's more clear that in that file `pyarrow` is just required for typing?\r\n3. Do you want me to add the License? In this PR? In a separate one?\r\n4. Ideally `jnp.asarray` is similar to `np.asarray` which in the case of `numpy` tends to be more efficient as it does zero-copy when possible, while `np.array` has `copy=True` by default, anyway as I mentioned before (and as you already know) the copy from `numpy` to `jax` is not zero-copy, while the other way around (`jax` to `numpy`) it is", "Thanks, feel free to create separate PRs for the docs and the license.\r\n\r\nI guess you can move the `pyarrow` import back to where it was for consistency with the other files and we can merge this one ;)", "> Thanks, feel free to create separate PRs for the docs and the license.\r\n> \r\n> I guess you can move the `pyarrow` import back to where it was for consistency with the other files and we can merge this one ;)\r\n\r\nCool thanks I'll do that! 👍🏻 ", "Actually I just checked and there are still tens of thousands of users with jax 0.3.25 - so we need to support older versions as well. I guess it comes from `transformers` which doesn't support jax 0.4 (and doesn't want to until the jax team stops breaking the lib all the time).\r\n\r\nCould you make sure your changes work with older versions as well ? Sorry for not spotting this earlier.\r\nIf we have `\"jax>=0.2.8,!=0.3.2,<=0.4.3\"` that'b be nice, and we can update the latest supported release from time to time.\r\n\r\nIn the CI you can add `jax==0.2.8` for the `deps-minimum` job, and use `jax~=0.4.1` for the `deps-latest`.", "> Actually I just checked and there are still tens of thousands of users with jax 0.3.25 - so we need to support older versions as well. I guess it comes from `transformers` which doesn't support jax 0.4 (and doesn't want to until the jax team stops breaking the lib all the time).\r\n> \r\n> Could you make sure your changes work with older versions as well ? Sorry for not spotting this earlier. If we have `\"jax>=0.2.8,!=0.3.2,<=0.4.3\"` that'b be nice, and we can update the latest supported release from time to time.\r\n> \r\n> In the CI you can add `jax==0.2.8` for the `deps-minimum` job, and use `jax~=0.4.1` for the `deps-latest`.\r\n\r\nOk, didn't know that @lhoestq thanks for the detailed context! Sure, I'll update it and make sure it's also compatible with older versions.", "Oops forgot to add you as co-author of the last commit @lhoestq my bad 😞 ", "So it should be fixed right now @lhoestq! The thing is that `jax` doesn't provide support for Python 3.7 due to its EOL next June (more information at https://endoflife.date/python)...\r\n\r\nAnyway, I can confirm that `jax.Array` type works with 0.3.25 and that the following code works fine:\r\n\r\n```python\r\nimport jax\r\nimport jax.numpy as jnp\r\n\r\nx = jnp.ones((1, 10), dtype=jnp.float32) # Is a `jnp.DeviceArray`\r\nassert isinstance(x, jax.Array) # Is `True`\r\n```\r\n\r\nSo we can still use 0.3.25 as the maximum supported version, as well as 0.3.6 for `jaxlib` so as to be consistent with 🤗`transformers`.\r\n\r\nThanks for your comments @lhoestq those were really useful!", "Sorry for the spam, pinning versions leads to failure runs (not related to the type-hinting); I'll check that locally instead of here to avoid spam... Not pinning the dependencies work but I'll check the minimum required versions for both `jax` and `jaxlib` in Python 3.7", "> Cool ! Thanks for trying to make the CI support it, but it's maybe not worth spending more time on this for now ^^\r\n> \r\n> merging :)\r\n\r\nDo you want me to work on the CI in a separate branch? Thanks for merging and for your help as always :)", "> Do you want me to work on the CI in a separate branch? Thanks for merging and for your help as always :)\r\n\r\nIn the end I think we can keep it as is since we didn't modify the core code for jax. Maybe later if we do further changes and need to make sure we don't break anything ;) For example when we decide to add support for more recent versions", "<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) | 0.010798 / 0.011353 (-0.000555) | 0.005690 / 0.011008 (-0.005318) | 0.116840 / 0.038508 (0.078332) | 0.041376 / 0.023109 (0.018266) | 0.345616 / 0.275898 (0.069718) | 0.413914 / 0.323480 (0.090434) | 0.009237 / 0.007986 (0.001252) | 0.004490 / 0.004328 (0.000162) | 0.085833 / 0.004250 (0.081582) | 0.050231 / 0.037052 (0.013179) | 0.367276 / 0.258489 (0.108787) | 0.393735 / 0.293841 (0.099894) | 0.043775 / 0.128546 (-0.084772) | 0.013215 / 0.075646 (-0.062432) | 0.391020 / 0.419271 (-0.028252) | 0.055102 / 0.043533 (0.011569) | 0.360333 / 0.255139 (0.105194) | 0.370531 / 0.283200 (0.087331) | 0.115484 / 0.141683 (-0.026199) | 1.694779 / 1.452155 (0.242625) | 1.756249 / 1.492716 (0.263532) |\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) | 0.230508 / 0.018006 (0.212501) | 0.478681 / 0.000490 (0.478191) | 0.010305 / 0.000200 (0.010105) | 0.000147 / 0.000054 (0.000093) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.030953 / 0.037411 (-0.006459) | 0.124320 / 0.014526 (0.109794) | 0.140417 / 0.176557 (-0.036140) | 0.189522 / 0.737135 (-0.547613) | 0.143635 / 0.296338 (-0.152704) |\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) | 0.485995 / 0.215209 (0.270786) | 4.799668 / 2.077655 (2.722014) | 2.195655 / 1.504120 (0.691535) | 1.940073 / 1.541195 (0.398879) | 2.053853 / 1.468490 (0.585363) | 0.825399 / 4.584777 (-3.759378) | 4.522180 / 3.745712 (0.776468) | 2.484626 / 5.269862 (-2.785236) | 1.727617 / 4.565676 (-2.838059) | 0.098808 / 0.424275 (-0.325467) | 0.014753 / 0.007607 (0.007146) | 0.606798 / 0.226044 (0.380754) | 5.918090 / 2.268929 (3.649162) | 2.668124 / 55.444624 (-52.776500) | 2.300447 / 6.876477 (-4.576030) | 2.411203 / 2.142072 (0.269130) | 0.999826 / 4.805227 (-3.805401) | 0.193683 / 6.500664 (-6.306981) | 0.069341 / 0.075469 (-0.006129) |\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) | 1.455816 / 1.841788 (-0.385972) | 17.176476 / 8.074308 (9.102168) | 16.359100 / 10.191392 (6.167708) | 0.199669 / 0.680424 (-0.480755) | 0.033456 / 0.534201 (-0.500745) | 0.512478 / 0.579283 (-0.066805) | 0.526350 / 0.434364 (0.091986) | 0.637669 / 0.540337 (0.097332) | 0.753821 / 1.386936 (-0.633115) |\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) | 0.008176 / 0.011353 (-0.003177) | 0.005862 / 0.011008 (-0.005147) | 0.086123 / 0.038508 (0.047615) | 0.037144 / 0.023109 (0.014035) | 0.398328 / 0.275898 (0.122430) | 0.439126 / 0.323480 (0.115647) | 0.006455 / 0.007986 (-0.001531) | 0.004575 / 0.004328 (0.000246) | 0.083396 / 0.004250 (0.079146) | 0.052827 / 0.037052 (0.015775) | 0.401039 / 0.258489 (0.142550) | 0.441374 / 0.293841 (0.147533) | 0.041671 / 0.128546 (-0.086875) | 0.014098 / 0.075646 (-0.061548) | 0.100873 / 0.419271 (-0.318398) | 0.058690 / 0.043533 (0.015157) | 0.395817 / 0.255139 (0.140678) | 0.409226 / 0.283200 (0.126026) | 0.119804 / 0.141683 (-0.021879) | 1.704583 / 1.452155 (0.252428) | 1.782527 / 1.492716 (0.289811) |\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) | 0.255166 / 0.018006 (0.237160) | 0.485091 / 0.000490 (0.484601) | 0.007458 / 0.000200 (0.007258) | 0.000116 / 0.000054 (0.000061) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.034531 / 0.037411 (-0.002880) | 0.134332 / 0.014526 (0.119806) | 0.144944 / 0.176557 (-0.031613) | 0.199352 / 0.737135 (-0.537783) | 0.152243 / 0.296338 (-0.144095) |\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) | 0.495361 / 0.215209 (0.280152) | 4.895144 / 2.077655 (2.817489) | 2.350419 / 1.504120 (0.846299) | 2.112131 / 1.541195 (0.570937) | 2.234469 / 1.468490 (0.765978) | 0.815862 / 4.584777 (-3.768915) | 4.531638 / 3.745712 (0.785926) | 2.405186 / 5.269862 (-2.864676) | 1.559020 / 4.565676 (-3.006656) | 0.100432 / 0.424275 (-0.323843) | 0.014217 / 0.007607 (0.006610) | 0.614622 / 0.226044 (0.388577) | 5.984541 / 2.268929 (3.715613) | 2.929897 / 55.444624 (-52.514727) | 2.484010 / 6.876477 (-4.392467) | 2.533538 / 2.142072 (0.391466) | 0.972119 / 4.805227 (-3.833108) | 0.193630 / 6.500664 (-6.307034) | 0.073694 / 0.075469 (-0.001775) |\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) | 1.503725 / 1.841788 (-0.338063) | 17.421529 / 8.074308 (9.347221) | 15.686433 / 10.191392 (5.495041) | 0.216688 / 0.680424 (-0.463736) | 0.020929 / 0.534201 (-0.513272) | 0.512523 / 0.579283 (-0.066760) | 0.499878 / 0.434364 (0.065514) | 0.639238 / 0.540337 (0.098900) | 0.769598 / 1.386936 (-0.617338) |\n\n</details>\n</details>\n\n![](https://cml.dev/watermark.png#99200127ade6d7b7d2cfb7b88365e5844b5c9c2e \"CML watermark\")\n", "> > Do you want me to work on the CI in a separate branch? Thanks for merging and for your help as always :)\r\n> \r\n> In the end I think we can keep it as is since we didn't modify the core code for jax. Maybe later if we do further changes and need to make sure we don't break anything ;) For example when we decide to add support for more recent versions\r\n\r\nMakes sense, thank you @lhoestq!" ]

[]

[ "Cool thanks :) Feel free to merge if it's all good for you" ]

[]

[ "_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" } ]

[]

[]

[ "looks like this PR includes changes to 5000 files\r\ncould you create a new branch and a new PR ?" ]

[]

[]

[]

[ "Sorry for this. I deleted sepedi_ner_corpus as per your earlier advise. Let me check. " ]

[]

[]

[]

[]

[]

[ "@lhoestq Thank you for the review. I have made all the changes you mentioned. PTAL! " ]

[ { "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" } ]

[ "Thanks for your contribution, @jaketae. Are you still interested in adding this dataset?\r\n\r\nWe are removing the dataset scripts from this GitHub repo and moving them to the Hugging Face Hub: https://huggingface.co/datasets\r\n\r\nWe would suggest you create this dataset there. Please, feel free to tell us if you need some help." ]

[]

[ "Wrongly branching from existing branch of wisesight_sentiment. Closing and opening another one specifically for prachathai67k" ]

[]

[ "Unfortunately, I can't reproduce the error. What does the following code return for you?\r\n```python\r\nimport requests\r\nfrom huggingface_hub import hf_hub_url\r\nr = requests.get(hf_hub_url(\"codeparrot/codeparrot-clean-train\", \"dataset_infos.json\", repo_type=\"dataset\"))\r\n```\r\n\r\nAlso, can you provide more info about your network (region, proxies, 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\nthe fix was merged and will be available in the next release of `datasets`.\r\nIn the meantime, you can use it by installing `datasets` directly from master as follows:\r\n```\r\npip install git+https://github.com/huggingface/datasets.git\r\n```" ]

[]

[ "looks like this PR also includes changes about another dataset `covid_qa_deepset`\r\n\r\nCould you create another branch and another PR that only includes the changes for the wikitext-tl-39 dataset ?" ]

[]

[]

[ { "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/flickr30k-captions)." ]

[ { "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, @david1542." ]

[]

[ "looks like this PR contains changes in many other files than the ones for PerSenT\r\ncan you create another branch and another PR ?", "closing since #1142 was merged" ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "Need some more time to fix the tests, especially with pickle", "> And I actually don't quite understand the idea - what's the motivation behind making only IterableDataset compatible with torch DataLoader without setting the format explicitly?\r\n\r\nSetting the format to pytorch = set the output types of the dataset to be pytorch tensors. However sometimes your dataset is not made of tensors but you still want to be able to use a pytorch DataLoader", "A bit more context. \r\n\r\nThe arrow-backed `Dataset` supports `DataLoader(ds)` (even if the format is not \"torch\"), and we want to be able to do the same with `IterableDataset` for consistency. However, this is when the PyTorch internals come into play - an iterable dataset needs to be an instance of `torch.utils.data.IterableDataset` due to [this](https://github.com/pytorch/pytorch/blob/abc54f93145830b502400faa92bec86e05422fbd/torch/utils/data/dataloader.py#L276) check (notice there is no check for the map-style version). Hence the explicit subclassing in this PR.", "Exactly :) Btw I just took your comments into account @polinaeterna , so feel free to review again", "@lhoestq just checking, does this change still preserve the fix to the \"data duplicate when setting num_works > 1 with streaming data\" issue from before?\r\n\r\nhttps://github.com/huggingface/datasets/issues/3423", "Yes :)" ]

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

[ "Hi! I agree it would be nice to have a method like that. Instead of `keep_columns`, we can name it `select_columns` to be more aligned with PyArrow's naming convention (`pa.Table.select`).", "Hi, I am a newbie to open source and would like to contribute. @mariosasko can I take up this issue ?", "Hey, I also want to work on this issue I am a newbie to open source. ", "This sounds related to https://github.com/huggingface/datasets/issues/5474\r\n\r\nI'm fine with `select_columns`, or we could also override `select` to also accept a list of columns maybe ?", "@lhoestq, I am planning to add a member function to the dataset class to perform the selection operation. Do you think its the right way to proceed? or there is a better option ?", "Unless @mariosasko thinks otherwise, I think it can go in `Dataset.select()` :)\r\nThough some parameters like keep_in_memory, indices_cache_file_name or writer_batch_size wouldn't when selecting columns, so we would need to update the docstring as well", "If someone wants to give it a shot, feel free to comment `#self-assign` and it will assign the issue to you.\r\n\r\nFeel free to ping us here if you have questions or if we can help :)", "I would rather have this functionality as a separate method. IMO it's always better to be explicit than to have an API where a single method can do different/uncorrelated things (somewhat reminds me of Pandas, and there is probably a good reason why PyArrow is more rigid in this aspect).", "In the end I also think it would be nice to have it as a separate method, this way we can also have it for `IterableDataset` (which can't have `select` for indices)" ]

[]

[]

[]

[ "Closing since this is a duplicate of #5711", "> Closing since this is a duplicate of #5711\r\n\r\nSorry @mariosasko , my internet went down went submitting the issue, and somehow it ended up creating a duplicate" ]

[]

[ "@lhoestq Hi~, does this support that `datasets.map(func, batched=True, batch_size, num_proc>1, new_fingerprint=\"func_v1\")` even if `func` can't pickle. I also notice that you said \"Unfortunately you need picklable mapping functions to make multiprocessing work :confused: Also feel free to open an issue or send me a dm if you are in a situation where the caching fails. I can help you with that :slight_smile:\" in [here](https://discuss.huggingface.co/t/how-to-deal-with-unpickable-objects-in-map/1547/8). So, I want to ask that is there a way for users to use multiprocessing in `datasets.map` when the `func` can't pickle? \r\nThanks in advance!", "Yea you need your function to be picklable for multiprocessing, otherwise the main process is not able to pass your function to the child processes." ]

[]

[]

[]

[ "Hi ! Maybe we can rename it `narrativeqa_manual` to make it explicit that this one requires manual download contrary to `narrativeqa` ?\r\nIt's important to have this one as well, in case the `narrativeqa` one suffers from download issues (checksums or dead links for example).\r\n\r\nYou can also copy the dataset card from `narrativeqa` and add the dummy data as well", "Thanks @lhoestq will do all this and submit a request in the coming days. 😊 ", "Closing this as another pull request is already done. " ]

[]

[ "Updated with Translation feature type. Working on dataset tags and README", "merging since the CI is fixed on master" ]

[]

[]

[]

[]

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

[ "Hi @bhavitvyamalik, thanks for reporting.\r\n\r\nYes, currently we do not support 7zip archive compression: I think we should.\r\n\r\nAs a workaround, you could uncompress it explicitly, like done in e.g. `samsum` dataset: \r\n\r\nhttps://github.com/huggingface/datasets/blob/fedf891a08bfc77041d575fad6c26091bc0fce52/datasets/samsum/samsum.py#L106-L110\r\n", "Related to this issue: https://github.com/huggingface/datasets/issues/3541", "Sure, let me look into and check what can be done. Will keep you guys updated here!", "Initially, I thought of solving this without any external dependency. Almost everywhere I saw `lzma` can be used for this but there is a caveat that lzma doesn’t work with 7z archives but only single files. In my case the 7z archive has multiple files so it didn't work. Is it fine to use external library here?", "Hi @bhavitvyamalik, thanks for your investigation.\r\n\r\nOn Monday, I started a PR that will eventually close this issue as well: I'm linking it to this.\r\n- #4672\r\n\r\nLet me know what you think. " ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "<details>\n<summary>Show benchmarks</summary>\n\nPyArrow==8.0.0\n\n<details>\n<summary>Show updated benchmarks!</summary>\n\n### Benchmark: benchmark_array_xd.json\n\n| metric | read_batch_formatted_as_numpy after write_array2d | read_batch_formatted_as_numpy after write_flattened_sequence | read_batch_formatted_as_numpy after write_nested_sequence | read_batch_unformated after write_array2d | read_batch_unformated after write_flattened_sequence | read_batch_unformated after write_nested_sequence | read_col_formatted_as_numpy after write_array2d | read_col_formatted_as_numpy after write_flattened_sequence | read_col_formatted_as_numpy after write_nested_sequence | read_col_unformated after write_array2d | read_col_unformated after write_flattened_sequence | read_col_unformated after write_nested_sequence | read_formatted_as_numpy after write_array2d | read_formatted_as_numpy after write_flattened_sequence | read_formatted_as_numpy after write_nested_sequence | read_unformated after write_array2d | read_unformated after write_flattened_sequence | read_unformated after write_nested_sequence | write_array2d | write_flattened_sequence | write_nested_sequence |\n|--------|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 0.007448 / 0.011353 (-0.003905) | 0.005182 / 0.011008 (-0.005826) | 0.098718 / 0.038508 (0.060210) | 0.034594 / 0.023109 (0.011485) | 0.317301 / 0.275898 (0.041403) | 0.357800 / 0.323480 (0.034320) | 0.005860 / 0.007986 (-0.002126) | 0.004267 / 0.004328 (-0.000061) | 0.074876 / 0.004250 (0.070626) | 0.048002 / 0.037052 (0.010950) | 0.333360 / 0.258489 (0.074871) | 0.362080 / 0.293841 (0.068239) | 0.035957 / 0.128546 (-0.092589) | 0.012245 / 0.075646 (-0.063401) | 0.332970 / 0.419271 (-0.086301) | 0.050825 / 0.043533 (0.007293) | 0.313936 / 0.255139 (0.058797) | 0.340684 / 0.283200 (0.057485) | 0.106630 / 0.141683 (-0.035053) | 1.427898 / 1.452155 (-0.024257) | 1.547518 / 1.492716 (0.054801) |\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) | 0.296952 / 0.018006 (0.278945) | 0.515708 / 0.000490 (0.515218) | 0.004225 / 0.000200 (0.004025) | 0.000082 / 0.000054 (0.000027) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.029365 / 0.037411 (-0.008046) | 0.111142 / 0.014526 (0.096616) | 0.124414 / 0.176557 (-0.052142) | 0.185227 / 0.737135 (-0.551908) | 0.129545 / 0.296338 (-0.166793) |\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) | 0.403303 / 0.215209 (0.188094) | 4.044138 / 2.077655 (1.966483) | 1.803622 / 1.504120 (0.299502) | 1.615436 / 1.541195 (0.074242) | 1.703576 / 1.468490 (0.235086) | 0.706398 / 4.584777 (-3.878379) | 3.912995 / 3.745712 (0.167283) | 4.004575 / 5.269862 (-1.265287) | 2.101592 / 4.565676 (-2.464085) | 0.087280 / 0.424275 (-0.336995) | 0.012564 / 0.007607 (0.004957) | 0.508484 / 0.226044 (0.282440) | 5.089351 / 2.268929 (2.820422) | 2.269022 / 55.444624 (-53.175602) | 1.933375 / 6.876477 (-4.943102) | 2.136783 / 2.142072 (-0.005289) | 0.862624 / 4.805227 (-3.942603) | 0.172107 / 6.500664 (-6.328557) | 0.066694 / 0.075469 (-0.008775) |\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) | 1.172513 / 1.841788 (-0.669275) | 15.877519 / 8.074308 (7.803211) | 14.687476 / 10.191392 (4.496084) | 0.189392 / 0.680424 (-0.491032) | 0.017334 / 0.534201 (-0.516866) | 0.420201 / 0.579283 (-0.159082) | 0.418502 / 0.434364 (-0.015862) | 0.489130 / 0.540337 (-0.051207) | 0.580678 / 1.386936 (-0.806258) |\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) | 0.007942 / 0.011353 (-0.003411) | 0.005312 / 0.011008 (-0.005696) | 0.074684 / 0.038508 (0.036176) | 0.035952 / 0.023109 (0.012843) | 0.349672 / 0.275898 (0.073774) | 0.377157 / 0.323480 (0.053678) | 0.006399 / 0.007986 (-0.001586) | 0.005769 / 0.004328 (0.001441) | 0.074283 / 0.004250 (0.070032) | 0.053217 / 0.037052 (0.016165) | 0.342545 / 0.258489 (0.084056) | 0.383663 / 0.293841 (0.089822) | 0.037234 / 0.128546 (-0.091312) | 0.012349 / 0.075646 (-0.063298) | 0.086522 / 0.419271 (-0.332749) | 0.049888 / 0.043533 (0.006355) | 0.337686 / 0.255139 (0.082547) | 0.361564 / 0.283200 (0.078365) | 0.104902 / 0.141683 (-0.036781) | 1.478259 / 1.452155 (0.026104) | 1.576376 / 1.492716 (0.083660) |\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) | 0.339760 / 0.018006 (0.321753) | 0.530946 / 0.000490 (0.530456) | 0.000474 / 0.000200 (0.000274) | 0.000063 / 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) | 0.029685 / 0.037411 (-0.007726) | 0.109409 / 0.014526 (0.094883) | 0.125579 / 0.176557 (-0.050978) | 0.175378 / 0.737135 (-0.561757) | 0.130672 / 0.296338 (-0.165667) |\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) | 0.428456 / 0.215209 (0.213247) | 4.238731 / 2.077655 (2.161077) | 2.046703 / 1.504120 (0.542583) | 1.850701 / 1.541195 (0.309506) | 1.909290 / 1.468490 (0.440800) | 0.714314 / 4.584777 (-3.870463) | 3.816056 / 3.745712 (0.070344) | 2.118567 / 5.269862 (-3.151295) | 1.348017 / 4.565676 (-3.217659) | 0.087140 / 0.424275 (-0.337135) | 0.012546 / 0.007607 (0.004938) | 0.538041 / 0.226044 (0.311997) | 5.381822 / 2.268929 (3.112893) | 2.525685 / 55.444624 (-52.918939) | 2.178659 / 6.876477 (-4.697817) | 2.381054 / 2.142072 (0.238981) | 0.844404 / 4.805227 (-3.960823) | 0.171802 / 6.500664 (-6.328862) | 0.065630 / 0.075469 (-0.009839) |\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) | 1.262187 / 1.841788 (-0.579600) | 16.197668 / 8.074308 (8.123360) | 15.148636 / 10.191392 (4.957244) | 0.152601 / 0.680424 (-0.527823) | 0.020238 / 0.534201 (-0.513963) | 0.420141 / 0.579283 (-0.159142) | 0.416295 / 0.434364 (-0.018068) | 0.487051 / 0.540337 (-0.053286) | 0.581942 / 1.386936 (-0.804994) |\n\n</details>\n</details>\n\n![](https://cml.dev/watermark.png#9615e5af75b190c4e7b66792f9ba444f352765a0 \"CML watermark\")\n" ]

[]

[ "This PR is amazing!!!\r\n\r\nI only looked at `arrow_dataset.py` and `formatting/formatting.py` but those look good to me.\r\n\r\nMy only (tiny) concern is the name of the function: I don't think it's self-evident that `set_format` applies a generic transformation, and some people might not look too far into the doc.\r\n\r\nMaybe we could have an `apply_transform` or `process_columns` method which is called by `set_format` (to keep backward compatibility)?", "What about something like `.set_format` and `.set_transform` ?\r\n- set_format would be the same as right now, i.e. defined by a format type.\r\n- set_transform would define the transformation that is applied on output batches on-the-fly.\r\n\r\nI was also thinking about `._with_format` and `.with_transform`. It could be their equivalent but would create a **new** dataset with the corresponding format or transform ? I know @sgugger was interested in something like that.", "Yup, I think that would make all of these options very clear!", "I like all those options as well (as long as the `_` in `_with_format` is a typo ;-) )", "Yes it's a typo indeed ;)\r\n\r\nAlright I'll do the changes !", "I took all your suggestions into account, thanks :)\r\nLet me know if you have more comments", "Hi @lhoestq , thanks for offering the set_transform() function. It is very handy to process large datasets on the fly. But I ran into a problem when using it (error message shown below). Since we are working with a large collection, there's no way to filter all invalid data points beforehand. Those invalid data points will be problematic with the set_transform and I don't find a good work-around to ignore them. I wonder if you can offer some advice on dealing with invalid data points in this case. Thank you!\r\n\r\n```\r\nTraceback (most recent call last):\r\n File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/torch/utils/data/_utils/worker.py\", line 198, in _worker_loop\r\n data = fetcher.fetch(index)\r\n File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py\", line 44, in fetch\r\n data = [self.dataset[idx] for idx in possibly_batched_index]\r\n File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py\", line 44, in <listcomp>\r\n data = [self.dataset[idx] for idx in possibly_batched_index]\r\n File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/datasets/arrow_dataset.py\", line 1763, in __getitem__\r\n return self._getitem(\r\n File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/datasets/arrow_dataset.py\", line 1748, in _getitem\r\n formatted_output = format_table(\r\n File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/datasets/formatting/formatting.py\", line 532, in format_table\r\n return formatter(pa_table, query_type=query_type)\r\n File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/datasets/formatting/formatting.py\", line 281, in __call__\r\n return self.format_row(pa_table)\r\n File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/datasets/formatting/formatting.py\", line 391, in format_row\r\n raise TypeError(\r\nTypeError: Custom formatting function must return a dict to be able to pick a row, but got None\r\n\r\n```\r\n", "> Hi @lhoestq , thanks for offering the set_transform() function. It is very handy to process large datasets on the fly. But I ran into a problem when using it (error message shown below). Since we are working with a large collection, there's no way to filter all invalid data points beforehand. Those invalid data points will be problematic with the set_transform and I don't find a good work-around to ignore them. I wonder if you can offer some advice on dealing with invalid data points in this case. Thank you!\r\n> \r\n> ```\r\n> Traceback (most recent call last):\r\n> File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/torch/utils/data/_utils/worker.py\", line 198, in _worker_loop\r\n> data = fetcher.fetch(index)\r\n> File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py\", line 44, in fetch\r\n> data = [self.dataset[idx] for idx in possibly_batched_index]\r\n> File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py\", line 44, in <listcomp>\r\n> data = [self.dataset[idx] for idx in possibly_batched_index]\r\n> File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/datasets/arrow_dataset.py\", line 1763, in __getitem__\r\n> return self._getitem(\r\n> File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/datasets/arrow_dataset.py\", line 1748, in _getitem\r\n> formatted_output = format_table(\r\n> File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/datasets/formatting/formatting.py\", line 532, in format_table\r\n> return formatter(pa_table, query_type=query_type)\r\n> File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/datasets/formatting/formatting.py\", line 281, in __call__\r\n> return self.format_row(pa_table)\r\n> File \"/export/share/ruimeng/env/anaconda/envs/ir/lib/python3.8/site-packages/datasets/formatting/formatting.py\", line 391, in format_row\r\n> raise TypeError(\r\n> TypeError: Custom formatting function must return a dict to be able to pick a row, but got None\r\n> ```\r\n\r\nI found this trick can be helpful: return an empty dict in exception:\r\n```\r\ndef transform_fn(example):\r\n try:\r\n process_your_data(example)\r\n except Exception as e:\r\n print(e)\r\n return {'input_ids': [[]], 'token_type_ids': [[]], 'attention_mask': [[]]}\r\ntrain_dataset = datasets.load_dataset(...)\r\ntrain_dataset = train_dataset.with_transform(parse_fn)\r\n```" ]

[]

[]

[]

[ "I've added the changes / fixes - ready for a second pass :)" ]

[]

[]

[]

[]

[]

[ "Hi @puzzler10, \r\nThose examples where `gold_label` field was empty, -1 label was alloted to it. In order to remove it you can filter the samples from train/val/test splits. Here's how you can drop those rows from the dataset:\r\n`dataset = load_dataset(\"snli\")`\r\n`dataset_test_filter = dataset['test'].filter(lambda example: example['label'] != -1)`\r\n\r\nI agree it should have been mentioned in the documentation. I'll raise a PR regarding the same. Thanks for pointing out!" ]

[ { "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" } ]

[ "wooo that was fast! thank you @lhoestq !\r\nit is able to process now, though it's ignoring all files and ending up with 0 examples now haha :/\r\n\r\nFor subset \"unfiltered\":\r\n```python\r\n>>> load_dataset(\"trivia_qa\", \"unfiltered\")\r\nDownloading and preparing dataset trivia_qa/unfiltered (download: 3.07 GiB, generated: 27.23 GiB, post-processed: Unknown size, total: 30.30 GiB) to /gpfsscratch/rech/six/commun/datasets/trivia_qa/unfiltered/1.1.0/910043a609bb2bdf62b4874f68e0c24fb648cf81e40a358f4bd54c919d72c9ab...\r\n100%|███████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 1354.53it/s]\r\n100%|█████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 40.60it/s]\r\nTraceback (most recent call last):\r\n File \"<stdin>\", line 1, in <module>\r\n File \"/gpfswork/rech/six/commun/conda/victor/lib/python3.7/site-packages/datasets/load.py\", line 1198, in load_dataset\r\n use_auth_token=use_auth_token,\r\n File \"/gpfswork/rech/six/commun/conda/victor/lib/python3.7/site-packages/datasets/builder.py\", line 647, in download_and_prepare\r\n dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs\r\n File \"/gpfswork/rech/six/commun/conda/victor/lib/python3.7/site-packages/datasets/builder.py\", line 748, in _download_and_prepare\r\n verify_splits(self.info.splits, split_dict)\r\n File \"/gpfswork/rech/six/commun/conda/victor/lib/python3.7/site-packages/datasets/utils/info_utils.py\", line 74, in verify_splits\r\n raise NonMatchingSplitsSizesError(str(bad_splits))\r\ndatasets.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=2906575347, num_examples=10832, dataset_name='trivia_qa'), 'recorded': SplitInfo(name='test', num_bytes=0, num_examples=0, dataset_name='trivia_qa')}, {'expected': SplitInfo(name='validation', num_bytes=3038966234, num_examples=11313, dataset_name='trivia_qa'), 'recorded': SplitInfo(name='validation', num_bytes=0, num_examples=0, dataset_name='trivia_qa')}]\r\n```\r\nFor subset \"rc\":\r\n```python\r\n>>> load_dataset(\"trivia_qa\", \"rc\")\r\nDownloading and preparing dataset trivia_qa/rc (download: 2.48 GiB, generated: 14.92 GiB, post-processed: Unknown size, total: 17.40 GiB) to /gpfsscratch/rech/six/commun/datasets/trivia_qa/rc/1.1.0/910043a609bb2bdf62b4874f68e0c24fb648cf81e40a358f4bd54c919d72c9ab...\r\n100%|███████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 3806.08it/s]\r\n100%|█████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 51.57it/s]\r\nTraceback (most recent call last):\r\n File \"<stdin>\", line 1, in <module>\r\n File \"/gpfswork/rech/six/commun/conda/victor/lib/python3.7/site-packages/datasets/load.py\", line 1198, in load_dataset\r\n use_auth_token=use_auth_token,\r\n File \"/gpfswork/rech/six/commun/conda/victor/lib/python3.7/site-packages/datasets/builder.py\", line 647, in download_and_prepare\r\n dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs\r\n File \"/gpfswork/rech/six/commun/conda/victor/lib/python3.7/site-packages/datasets/builder.py\", line 748, in _download_and_prepare\r\n verify_splits(self.info.splits, split_dict)\r\n File \"/gpfswork/rech/six/commun/conda/victor/lib/python3.7/site-packages/datasets/utils/info_utils.py\", line 74, in verify_splits\r\n raise NonMatchingSplitsSizesError(str(bad_splits))\r\ndatasets.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=1577814583, num_examples=17210, dataset_name='trivia_qa'), 'recorded': SplitInfo(name='test', num_bytes=0, num_examples=0, dataset_name='trivia_qa')}, {'expected': SplitInfo(name='train', num_bytes=12750976012, num_examples=138384, dataset_name='trivia_qa'), 'recorded': SplitInfo(name='train', num_bytes=0, num_examples=0, dataset_name='trivia_qa')}, {'expected': SplitInfo(name='validation', num_bytes=1688535379, num_examples=18669, dataset_name='trivia_qa'), 'recorded': SplitInfo(name='validation', num_bytes=0, num_examples=0, dataset_name='trivia_qa')}]\r\n```\r\n\r\nCould you look into that when you get a chance?\r\nI wonder if it's not something they changed on the file to download? i couldn't find any information", "@VictorSanh have you tried passing `download_mode=\"force_redownload\"`?\r\n```python\r\nds = load_dataset(\"trivia_qa\", \"unfiltered\", download_mode=\"force_redownload\")\r\n```", "I aggressively rmed caches, especially rming the `datasets/downloads/extracted/c3d265fa20d99a147a76e4f5e...` solved the issue.\r\nthank you both!\r\n" ]

[]

[ "@lhoestq Feel free to have a look. The implementation is slightly different from what you suggested. I have opted to overwrite `save` instead of meddling with `save_global`. `save_global` is called very late down in dill/pickle so it is hard to control for what is happening there. I might be wrong. Pickling is more complex than I thought! \r\n\r\nThe linked issue (`map` with spaCy) also works now!\r\n\r\n```python\r\nimport pickle\r\nimport spacy\r\nfrom spacy import Language\r\nfrom datasets import load_dataset\r\nfrom datasets.utils.py_utils import dumps, pklregister\r\n\r\n@pklregister(Language, allow_subclasses=True)\r\ndef hash_spacy_language(pickler, nlp: Language):\r\n pickler.save(nlp.to_bytes())\r\n\r\ndef main():\r\n fin = r\"large/file.txt\"\r\n nlp = spacy.load(\"en_core_web_sm\")\r\n\r\n def tokenize(l):\r\n return {\"tok\": [t.text for t in nlp(l[\"text\"])]}\r\n\r\n ds = load_dataset(\"text\", data_files=fin)\r\n ds = ds[\"train\"].map(tokenize)\r\n\r\n # Sanity check: load NLP from pickle created with our own `dumps`\r\n config = nlp.config\r\n lang_cls = spacy.util.get_lang_class(config[\"nlp\"][\"lang\"])\r\n nlp2 = lang_cls.from_config(config)\r\n nlp2.from_bytes(pickle.loads(dumps(nlp)))\r\n\r\n assert isinstance(nlp2, type(nlp))\r\n assert dumps(nlp) == dumps(nlp2)\r\n\r\nif __name__ == '__main__':\r\n main()\r\n```\r\n\r\nIf this all looks good to you, I'll start writing on some documentation and examples.\r\n", "One more thing. This is a reduction function for SpaCy Language that should work with the new API:\r\n```python\r\n@pklregister(Language, allow_subclasses=True)\r\ndef hash_spacy_language(pickler, obj):\r\n def create_language(config, bytes_data):\r\n lang_cls = spacy.util.get_lang_class(config[\"nlp\"][\"lang\"])\r\n nlp = lang_cls.from_config(config)\r\n return nlp.from_bytes(bytes_data)\r\n\r\n args = (obj.config, obj.to_bytes())\r\n pickler.save_reduce(create_language, args, obj=obj)\r\n```\r\nso IMO we are missing a test with `pickler.save_reduce`. ", "> One more thing. This is a reduction function for SpaCy Language that should work with the new API:\r\n> \r\n> ```python\r\n> @pklregister(Language, allow_subclasses=True)\r\n> def hash_spacy_language(pickler, obj):\r\n> def create_language(config, bytes_data):\r\n> lang_cls = spacy.util.get_lang_class(config[\"nlp\"][\"lang\"])\r\n> nlp = lang_cls.from_config(config)\r\n> return nlp.from_bytes(bytes_data)\r\n> \r\n> args = (obj.config, obj.to_bytes())\r\n> pickler.save_reduce(create_language, args, obj=obj)\r\n> ```\r\n> \r\n> so IMO we are missing a test with `pickler.save_reduce`.\r\n\r\nSure that seems a good idea, but I do not quite understand what `save_reduce` does. Could you give some more info about what reduce functions do and how they differ from regular `save` and `save_global`? I've read about it but the docs nor the built-in `pickle` code seem really helpful.", "I'm no pickle expect, but here is my understanding. I believe the pickler uses the reduce function when you do `loads` to reconstructs the original object from the parameters/arguments that were saved with `dumps`.\r\n\r\nFor example your sanity check could be simplified from\r\n```python\r\n config = nlp.config\r\n lang_cls = spacy.util.get_lang_class(config[\"nlp\"][\"lang\"])\r\n nlp2 = lang_cls.from_config(config)\r\n nlp2 = nlp2.from_bytes(pickle.loads(dumps(nlp)))\r\n```\r\nto\r\nEDIT: <s>pickle.loads(pickle.dumps(nlp))</s>\r\n```python\r\n nlp2 = loads(dumps(nlp)) # using our custom pickler\r\n```\r\n\r\nThough note that while it can be convenient for tests, we actually don't care about the reconstruction of the object since we're only using the pickler for `dumps` to compute hashes.", "> I'm no pickle expect, but here is my understanding. I believe the pickler uses the reduce function when you do `loads` to reconstructs the original object from the parameters/arguments that were saved with `dumps`.\r\n> \r\n> For example your sanity check could be simplified from\r\n> \r\n> ```python\r\n> config = nlp.config\r\n> lang_cls = spacy.util.get_lang_class(config[\"nlp\"][\"lang\"])\r\n> nlp2 = lang_cls.from_config(config)\r\n> nlp2 = nlp2.from_bytes(pickle.loads(dumps(nlp)))\r\n> ```\r\n> \r\n> to\r\n> \r\n> ```python\r\n> nlp2 = pickle.loads(pickle.dumps(nlp))\r\n> ```\r\n> \r\n> Though note that while it can be convenient for tests, we actually don't care about the reconstruction of the object since we're only using the pickler for `dumps` to compute hashes.\r\n\r\nYes, the sanity check can be simplified like that _if_ we use `pickle.dumps` - but that would not test our own `dumps` functionality and would do a naive dump instead of using `to_bytes`. It won't work if we use our own `dumps`, exactly because of the reason that we want custom pickling and being able to call `to_bytes`. To reconstruct the object from the pickled bytes from `to_bytes` we need `from_bytes`. The result of pickle/dill loads will therefore always be a `bytes` object and not a `Language` object.\r\n\r\nBut `save_reduce` is called when saving, right? Not when loading, AFAICT. I am just not sure what exactly it is saving. It is _potentially_ called [at the end of `save`](https://github.com/python/cpython/blob/24af9a40a8f85af813ea89998aa4e931fcc78cd9/Lib/pickle.py#L603) but only if we haven't returned by then. I just can't figure out what that base case is.", "I don't think we expect users to write the reduce function that isn't going to be used anyway. So maybe let's stick with `save` ?", "@BramVanroy \r\nAs I understand `save_reduce` is very similar to `copyreg.pickle`, so I'd suggest you to check the following links:\r\n* https://docs.python.org/3/library/copyreg.html#copyreg.pickle\r\n* https://docs.python.org/3/library/pickle.html#object.__reduce__\r\n\r\n\r\n@lhoestq \r\n> I don't think we expect users to write the reduce function that isn't going to be used anyway. So maybe let's stick with save ?\r\n\r\nI agree. \r\n\r\n`save_reduce` is very similar to `copyreg.pickle` and `object.__reduce__`, which are part of public API (and `save` isn't), so I expect more advanced users to know how to write their own reduction functions. But, as you say, `pklregister` should also work with `save` (even though I think `save` is a bit lower-level, and harder to understand than `save_reduce`).\r\n\r\nAll our examples in `py_utils` that use `pklregister` also use `save_reduce` in the last step, so my reduction for SpaCy is meant to be added there, and not to be written by users (because SpaCy is very popular, so the official support by us makes sense :)).\r\n\r\nAnd in the tests, let's ignore the reconstruction part of pickle/dill, because it's not important for us, and focus on the generated dumps. What do you think?", "@mariosasko What exactly do you mean with \"isn't part of the public API\"? It is [a public method](https://github.com/python/cpython/blob/24af9a40a8f85af813ea89998aa4e931fcc78cd9/Lib/pickle.py#L535) in base pickle, just like `dump` is but maybe you mean something else.", "@BramVanroy Oh sorry, it's public (not prefixed with `\"_\"`) but it's not documented in the docs. `save_reduce` is also not in the docs, but its signature/functionality is similar to `copyreg.pickle` and I see it more often being used in the projects on GH, so it's seems \"more public\" to me. ", "Unfortunately I feel that pickle in general is under-documented. 😄 \r\n\r\nFor the documentation, I can add a brief example, maybe under \"How-to Guides\"? The only thing that isn't immediately obvious to me is how I can add that doc page to the TOC?", "Yes great idea ! To add that doc page to the TOC, you just have to add it to the index.rst file in the \"How-to guides\" TOC section", "@mariosasko @lhoestq Feel free to make any edits or suggestions in the text!", "Hi @mariosasko. I wish you'd told me sooner, as I spent quite some time writing on this.\r\n\r\nI'm also not sure whether it is too advanced to have in the documentation. The spaCy use-case seems potentially frequent. Or do you wish to add that case to the defaults, and whenever new issues come up that seem like frequent/obvious cases, add those internally as well?", "Documenting the internal `pklregister` is overkill IMO (and it can be kept in docstrings), but we can document something higher level like `register_hash_func` once it's implemented.\r\n\r\nSo we keep the nice documentation you've written (thank you!), except we can rename it to \"Advanced caching\" and show an API that is similar to\r\n```python\r\n>>> @register_hash_func(Language, allow_subclasses=True)\r\n>>> def hash_spacy_language(nlp: Language):\r\n>>> return (nlp.to_bytes(),)\r\n```\r\nThis way we keep the documentation centered around the public API rather than the internals that may evolve/be too complicated to fit only one section.\r\n\r\n> Or do you wish to add that case to the defaults, and whenever new issues come up that seem like frequent/obvious cases, add those internally as well?\r\n\r\nLet's add it to the defaults since it's a frequent use-case. And also allow users to control the hashing using the API mentioned above if they face other non-trivially-hashable objects", "Sure, I can have a go at implementing spaCy as a built-in. Should it be included in the tests? (Therefore adding spaCy to the tests requirements.)\r\n\r\nNext, from your example, it seems that the return value of `register_hash_func` will be used in pickler.save automatically (calling pklregister a bit deeper). Any reason why it returns a tuple? I can work on this as well, if needed.", "> Sure, I can have a go at implementing spaCy as a built-in. Should it be included in the tests? (Therefore adding spaCy to the tests requirements.)\r\n\r\nThat would be perfect !\r\n\r\n> Next, from your example, it seems that the return value of register_hash_func will be used in pickler.save automatically (calling pklregister a bit deeper). \r\n\r\nYes I think so. For example register_hash_func can call pklregister with the user's function, but wrapped to use pickler.save.\r\n\r\n> Any reason why it returns a tuple? I can work on this as well, if needed.\r\n\r\nIt can either return an arbitrary object or a tuple. I like it a bit better if it's a tuple, so users understand more easily how to make the function take into account more than one item for the hash. It's also consistent with the streamlit caching functions, that also require a tuple. No strong opinion on this though\r\n\r\nLet me know if I can help with anything", "@lhoestq I do not have the time anymore to work on this. Can someone else pick this up?", "Hi ! Sure someone else can continue this PR (either someone from HF, or other contributors can fork the PR).\r\nI think I can work on this next week or the week after, but if anyone wants to work on this earlier feel free to comment here :)" ]

[]

[]

[]

[ "Thanks for reporting, @markovalexander.\r\n\r\nUnfortunately, I'm not able to reproduce the issue: the `tiktoken` tokenizer can be used within `Dataset.map`, both in my local machine and in a Colab notebook: https://colab.research.google.com/drive/1DhJroZgk0sNFJ2Mrz-jYgrmh9jblXaCG?usp=sharing\r\n\r\nAre you sure you are using `datasets` version 2.11.0?" ]

[ { "color": "B67A40", "default": false, "description": "Restructuring existing code without changing its external behavior", "id": 2851292821, "name": "refactoring", "node_id": "MDU6TGFiZWwyODUxMjkyODIx", "url": "https://api.github.com/repos/huggingface/datasets/labels/refactoring" } ]

[ "Hi @lhoestq,\r\n\r\nOnce that #2578 has been merged, I would like to ask you to have a look at this PR: it implements the same logic as the one in #2578 but for all the other file compression formats.\r\n\r\nThanks.", "I think all is done @lhoestq ;)" ]

[]

[ "To show the issue:\r\n```\r\npython -c \"from datasets.fingerprint import Hasher; a=[]; func = lambda : len(a); print(Hasher.hash(func))\"\r\n```\r\ndoesn't always return the same ouput since `globs` is a dictionary with \"a\" and \"len\" as keys but sometimes not in the same order" ]

[]

[ "It looks like there was an issue while building the imdb dataset.\r\nCould you provide more information about your OS and the version of python and `datasets` ?\r\n\r\nAlso could you try again with \r\n```python\r\ndataset = datasets.load_dataset(\"imdb\", split=\"train\", download_mode=\"force_redownload\")\r\n```\r\nto make sure it's not a corrupted file issue ?", "I was using version 1.1.2 and this resolved with version 1.1.3, thanks. ", "Hello,\r\nI have the same pb with 1.8.0", "Hi ! I just tried in 1.8.0 and it worked fine. Can you try again ? Maybe the dataset host had some issues that are fixed now", "Hello,\r\nIt works fine now :) !\r\nThanks !" ]

[]

[ "I think it actually comes from an issue here:\r\n\r\nhttps://github.com/huggingface/datasets/blob/810b12f763f5cf02f2e43565b8890d278b7398cd/src/datasets/utils/file_utils.py#L575-L579\r\n\r\nand \r\n\r\nhttps://github.com/huggingface/datasets/blob/810b12f763f5cf02f2e43565b8890d278b7398cd/src/datasets/utils/streaming_download_manager.py#L386-L389\r\n\r\nThis code doesn't seem to work anymore. This can probably be fixed with\r\n\r\n```python\r\nif url.startswith(\"https://drive.google.com/\"): \r\n url += \"&confirm=t\"\r\n cookies = response.cookies \r\n```\r\n\r\nbecause Google Drive doesn't return the `download_warning` cookie anymore.", "Actually it seems that is has been fixed already in https://github.com/huggingface/datasets/pull/3787 :)\r\n\r\nI think it should have fixed the gigaword dataset loading", "@lhoestq The linked PR indeed fixes the issue. This PR is still worth merging IMO to update `gigaword`'s card." ]

[]

[ "One trick if you want to add other datasets: consider running these commands each time you want to add a new dataset\r\n```\r\ngit checkout master\r\ngit fetch upstream\r\ngit rebase upstream/master\r\ngit checkout -b add-<my_dataset_name>\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" }, { "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! This is indeed a bug in our zero-copy logic.\r\n\r\nTo fix it, instead of the line:\r\nhttps://github.com/huggingface/datasets/blob/7cfac43b980ab9e4a69c2328f085770996323005/src/datasets/features/features.py#L702\r\n\r\nwe should have:\r\n```python\r\nreturn pa.types.is_primitive(pa_type) and not (pa.types.is_boolean(pa_type) or pa.types.is_temporal(pa_type))\r\n```", "@mariosasko submitted a small PR [here](https://github.com/huggingface/datasets/pull/5504)" ]

[ { "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 just opened a PR and I verified that the code you provided doesn't show any deprecation warning :)" ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "<details>\n<summary>Show benchmarks</summary>\n\nPyArrow==8.0.0\n\n<details>\n<summary>Show updated benchmarks!</summary>\n\n### Benchmark: benchmark_array_xd.json\n\n| metric | read_batch_formatted_as_numpy after write_array2d | read_batch_formatted_as_numpy after write_flattened_sequence | read_batch_formatted_as_numpy after write_nested_sequence | read_batch_unformated after write_array2d | read_batch_unformated after write_flattened_sequence | read_batch_unformated after write_nested_sequence | read_col_formatted_as_numpy after write_array2d | read_col_formatted_as_numpy after write_flattened_sequence | read_col_formatted_as_numpy after write_nested_sequence | read_col_unformated after write_array2d | read_col_unformated after write_flattened_sequence | read_col_unformated after write_nested_sequence | read_formatted_as_numpy after write_array2d | read_formatted_as_numpy after write_flattened_sequence | read_formatted_as_numpy after write_nested_sequence | read_unformated after write_array2d | read_unformated after write_flattened_sequence | read_unformated after write_nested_sequence | write_array2d | write_flattened_sequence | write_nested_sequence |\n|--------|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 0.007700 / 0.011353 (-0.003653) | 0.004680 / 0.011008 (-0.006328) | 0.098812 / 0.038508 (0.060304) | 0.085062 / 0.023109 (0.061952) | 0.371472 / 0.275898 (0.095574) | 0.412552 / 0.323480 (0.089072) | 0.004700 / 0.007986 (-0.003285) | 0.003765 / 0.004328 (-0.000564) | 0.074267 / 0.004250 (0.070017) | 0.063003 / 0.037052 (0.025951) | 0.391842 / 0.258489 (0.133353) | 0.436955 / 0.293841 (0.143114) | 0.035291 / 0.128546 (-0.093255) | 0.009309 / 0.075646 (-0.066338) | 0.313097 / 0.419271 (-0.106174) | 0.060098 / 0.043533 (0.016565) | 0.350726 / 0.255139 (0.095587) | 0.402692 / 0.283200 (0.119493) | 0.029321 / 0.141683 (-0.112361) | 1.671806 / 1.452155 (0.219651) | 1.743760 / 1.492716 (0.251044) |\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) | 0.242281 / 0.018006 (0.224275) | 0.505054 / 0.000490 (0.504564) | 0.006595 / 0.000200 (0.006395) | 0.000091 / 0.000054 (0.000037) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.032174 / 0.037411 (-0.005238) | 0.094483 / 0.014526 (0.079957) | 0.108527 / 0.176557 (-0.068030) | 0.178983 / 0.737135 (-0.558152) | 0.113766 / 0.296338 (-0.182572) |\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) | 0.419764 / 0.215209 (0.204555) | 4.282650 / 2.077655 (2.204995) | 2.075325 / 1.504120 (0.571205) | 1.897668 / 1.541195 (0.356473) | 2.027109 / 1.468490 (0.558619) | 0.519983 / 4.584777 (-4.064794) | 4.134603 / 3.745712 (0.388891) | 6.586711 / 5.269862 (1.316849) | 3.811726 / 4.565676 (-0.753951) | 0.058628 / 0.424275 (-0.365647) | 0.007586 / 0.007607 (-0.000021) | 0.502180 / 0.226044 (0.276136) | 5.101588 / 2.268929 (2.832660) | 2.534295 / 55.444624 (-52.910330) | 2.220170 / 6.876477 (-4.656307) | 2.441110 / 2.142072 (0.299038) | 0.644775 / 4.805227 (-4.160452) | 0.144716 / 6.500664 (-6.355948) | 0.067018 / 0.075469 (-0.008451) |\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) | 1.431279 / 1.841788 (-0.410508) | 21.947814 / 8.074308 (13.873506) | 15.548236 / 10.191392 (5.356844) | 0.174774 / 0.680424 (-0.505650) | 0.021182 / 0.534201 (-0.513019) | 0.441320 / 0.579283 (-0.137963) | 0.476685 / 0.434364 (0.042321) | 0.506277 / 0.540337 (-0.034060) | 0.809943 / 1.386936 (-0.576993) |\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) | 0.007172 / 0.011353 (-0.004181) | 0.004358 / 0.011008 (-0.006650) | 0.068604 / 0.038508 (0.030096) | 0.083956 / 0.023109 (0.060847) | 0.402579 / 0.275898 (0.126681) | 0.444714 / 0.323480 (0.121235) | 0.005940 / 0.007986 (-0.002046) | 0.003607 / 0.004328 (-0.000722) | 0.073134 / 0.004250 (0.068883) | 0.061722 / 0.037052 (0.024669) | 0.410957 / 0.258489 (0.152468) | 0.458819 / 0.293841 (0.164978) | 0.033710 / 0.128546 (-0.094836) | 0.010230 / 0.075646 (-0.065417) | 0.084678 / 0.419271 (-0.334593) | 0.058203 / 0.043533 (0.014670) | 0.444972 / 0.255139 (0.189833) | 0.470962 / 0.283200 (0.187763) | 0.029222 / 0.141683 (-0.112461) | 1.671460 / 1.452155 (0.219306) | 1.759471 / 1.492716 (0.266754) |\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) | 0.238894 / 0.018006 (0.220888) | 0.493605 / 0.000490 (0.493115) | 0.001979 / 0.000200 (0.001780) | 0.000084 / 0.000054 (0.000030) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.036498 / 0.037411 (-0.000913) | 0.095245 / 0.014526 (0.080719) | 0.112147 / 0.176557 (-0.064409) | 0.171128 / 0.737135 (-0.566007) | 0.115295 / 0.296338 (-0.181044) |\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) | 0.461067 / 0.215209 (0.245858) | 4.723932 / 2.077655 (2.646277) | 2.432697 / 1.504120 (0.928578) | 2.237302 / 1.541195 (0.696107) | 2.351320 / 1.468490 (0.882830) | 0.509963 / 4.584777 (-4.074813) | 4.194817 / 3.745712 (0.449105) | 6.689529 / 5.269862 (1.419667) | 3.351198 / 4.565676 (-1.214478) | 0.064563 / 0.424275 (-0.359712) | 0.008605 / 0.007607 (0.000998) | 0.575590 / 0.226044 (0.349546) | 5.644179 / 2.268929 (3.375250) | 3.021375 / 55.444624 (-52.423249) | 2.595305 / 6.876477 (-4.281172) | 2.839228 / 2.142072 (0.697156) | 0.657148 / 4.805227 (-4.148079) | 0.144831 / 6.500664 (-6.355834) | 0.067882 / 0.075469 (-0.007587) |\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) | 1.595580 / 1.841788 (-0.246208) | 22.431609 / 8.074308 (14.357301) | 15.700845 / 10.191392 (5.509453) | 0.164675 / 0.680424 (-0.515749) | 0.021322 / 0.534201 (-0.512879) | 0.455270 / 0.579283 (-0.124013) | 0.451547 / 0.434364 (0.017183) | 0.520955 / 0.540337 (-0.019383) | 0.687803 / 1.386936 (-0.699133) |\n\n</details>\n</details>\n\n![](https://cml.dev/watermark.png#7d19574e9f44bd3b59a3e47ca7c4ea66305a8e6b \"CML watermark\")\n", "<details>\n<summary>Show benchmarks</summary>\n\nPyArrow==8.0.0\n\n<details>\n<summary>Show updated benchmarks!</summary>\n\n### Benchmark: benchmark_array_xd.json\n\n| metric | read_batch_formatted_as_numpy after write_array2d | read_batch_formatted_as_numpy after write_flattened_sequence | read_batch_formatted_as_numpy after write_nested_sequence | read_batch_unformated after write_array2d | read_batch_unformated after write_flattened_sequence | read_batch_unformated after write_nested_sequence | read_col_formatted_as_numpy after write_array2d | read_col_formatted_as_numpy after write_flattened_sequence | read_col_formatted_as_numpy after write_nested_sequence | read_col_unformated after write_array2d | read_col_unformated after write_flattened_sequence | read_col_unformated after write_nested_sequence | read_formatted_as_numpy after write_array2d | read_formatted_as_numpy after write_flattened_sequence | read_formatted_as_numpy after write_nested_sequence | read_unformated after write_array2d | read_unformated after write_flattened_sequence | read_unformated after write_nested_sequence | write_array2d | write_flattened_sequence | write_nested_sequence |\n|--------|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 0.008171 / 0.011353 (-0.003182) | 0.005563 / 0.011008 (-0.005445) | 0.102265 / 0.038508 (0.063757) | 0.074755 / 0.023109 (0.051646) | 0.431317 / 0.275898 (0.155419) | 0.472179 / 0.323480 (0.148699) | 0.006153 / 0.007986 (-0.001833) | 0.003832 / 0.004328 (-0.000496) | 0.078480 / 0.004250 (0.074230) | 0.056250 / 0.037052 (0.019197) | 0.432938 / 0.258489 (0.174449) | 0.480983 / 0.293841 (0.187142) | 0.048861 / 0.128546 (-0.079685) | 0.016252 / 0.075646 (-0.059394) | 0.343508 / 0.419271 (-0.075763) | 0.065057 / 0.043533 (0.021524) | 0.468418 / 0.255139 (0.213279) | 0.463692 / 0.283200 (0.180492) | 0.032912 / 0.141683 (-0.108771) | 1.795194 / 1.452155 (0.343039) | 1.833047 / 1.492716 (0.340331) |\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) | 0.197980 / 0.018006 (0.179974) | 0.500662 / 0.000490 (0.500172) | 0.007380 / 0.000200 (0.007181) | 0.000110 / 0.000054 (0.000055) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.028323 / 0.037411 (-0.009089) | 0.089817 / 0.014526 (0.075291) | 0.102923 / 0.176557 (-0.073633) | 0.173851 / 0.737135 (-0.563284) | 0.104006 / 0.296338 (-0.192333) |\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) | 0.580277 / 0.215209 (0.365068) | 5.878739 / 2.077655 (3.801085) | 2.404673 / 1.504120 (0.900553) | 2.071765 / 1.541195 (0.530571) | 2.106024 / 1.468490 (0.637534) | 0.855217 / 4.584777 (-3.729560) | 4.918602 / 3.745712 (1.172890) | 5.354984 / 5.269862 (0.085122) | 3.141288 / 4.565676 (-1.424389) | 0.099553 / 0.424275 (-0.324723) | 0.008152 / 0.007607 (0.000545) | 0.709857 / 0.226044 (0.483813) | 7.144602 / 2.268929 (4.875673) | 3.137637 / 55.444624 (-52.306987) | 2.379851 / 6.876477 (-4.496626) | 2.346426 / 2.142072 (0.204353) | 1.033416 / 4.805227 (-3.771811) | 0.213120 / 6.500664 (-6.287544) | 0.076037 / 0.075469 (0.000568) |\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) | 1.597742 / 1.841788 (-0.244046) | 21.745366 / 8.074308 (13.671058) | 20.830698 / 10.191392 (10.639306) | 0.238727 / 0.680424 (-0.441697) | 0.027923 / 0.534201 (-0.506278) | 0.466073 / 0.579283 (-0.113210) | 0.548647 / 0.434364 (0.114283) | 0.549245 / 0.540337 (0.008908) | 0.977148 / 1.386936 (-0.409788) |\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) | 0.008252 / 0.011353 (-0.003101) | 0.004653 / 0.011008 (-0.006356) | 0.084012 / 0.038508 (0.045504) | 0.077418 / 0.023109 (0.054309) | 0.440748 / 0.275898 (0.164850) | 0.464279 / 0.323480 (0.140799) | 0.005762 / 0.007986 (-0.002224) | 0.004909 / 0.004328 (0.000581) | 0.086441 / 0.004250 (0.082190) | 0.057883 / 0.037052 (0.020831) | 0.466655 / 0.258489 (0.208166) | 0.479751 / 0.293841 (0.185910) | 0.047166 / 0.128546 (-0.081380) | 0.014480 / 0.075646 (-0.061166) | 0.092599 / 0.419271 (-0.326672) | 0.062454 / 0.043533 (0.018921) | 0.449753 / 0.255139 (0.194614) | 0.461876 / 0.283200 (0.178676) | 0.034828 / 0.141683 (-0.106855) | 1.752249 / 1.452155 (0.300095) | 1.865449 / 1.492716 (0.372732) |\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) | 0.245028 / 0.018006 (0.227022) | 0.509564 / 0.000490 (0.509074) | 0.003930 / 0.000200 (0.003730) | 0.000110 / 0.000054 (0.000056) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.034746 / 0.037411 (-0.002665) | 0.096563 / 0.014526 (0.082037) | 0.107581 / 0.176557 (-0.068975) | 0.184952 / 0.737135 (-0.552184) | 0.108747 / 0.296338 (-0.187591) |\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) | 0.613091 / 0.215209 (0.397882) | 5.994985 / 2.077655 (3.917330) | 2.711276 / 1.504120 (1.207156) | 2.415862 / 1.541195 (0.874668) | 2.391055 / 1.468490 (0.922565) | 0.868723 / 4.584777 (-3.716054) | 4.953992 / 3.745712 (1.208280) | 4.606542 / 5.269862 (-0.663319) | 2.942162 / 4.565676 (-1.623515) | 0.102737 / 0.424275 (-0.321538) | 0.008634 / 0.007607 (0.001027) | 0.722122 / 0.226044 (0.496078) | 7.245097 / 2.268929 (4.976168) | 3.428232 / 55.444624 (-52.016393) | 2.709539 / 6.876477 (-4.166938) | 2.857956 / 2.142072 (0.715884) | 1.045594 / 4.805227 (-3.759634) | 0.213344 / 6.500664 (-6.287320) | 0.073601 / 0.075469 (-0.001868) |\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) | 1.651954 / 1.841788 (-0.189834) | 22.458646 / 8.074308 (14.384338) | 19.583203 / 10.191392 (9.391811) | 0.246932 / 0.680424 (-0.433492) | 0.025730 / 0.534201 (-0.508471) | 0.473475 / 0.579283 (-0.105808) | 0.521411 / 0.434364 (0.087047) | 0.562038 / 0.540337 (0.021700) | 0.767673 / 1.386936 (-0.619263) |\n\n</details>\n</details>\n\n![](https://cml.dev/watermark.png#3869d99628329c696f6975377f65e625dd8ef3e0 \"CML watermark\")\n", "The CI error is unrelated.", "<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) | 0.006649 / 0.011353 (-0.004703) | 0.003963 / 0.011008 (-0.007045) | 0.084564 / 0.038508 (0.046056) | 0.075668 / 0.023109 (0.052559) | 0.314233 / 0.275898 (0.038335) | 0.343320 / 0.323480 (0.019841) | 0.005405 / 0.007986 (-0.002581) | 0.003356 / 0.004328 (-0.000973) | 0.065094 / 0.004250 (0.060844) | 0.058774 / 0.037052 (0.021722) | 0.320772 / 0.258489 (0.062283) | 0.353546 / 0.293841 (0.059705) | 0.030921 / 0.128546 (-0.097625) | 0.008463 / 0.075646 (-0.067184) | 0.287490 / 0.419271 (-0.131781) | 0.053188 / 0.043533 (0.009656) | 0.324023 / 0.255139 (0.068884) | 0.337828 / 0.283200 (0.054628) | 0.024764 / 0.141683 (-0.116918) | 1.458028 / 1.452155 (0.005873) | 1.521615 / 1.492716 (0.028899) |\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) | 0.209360 / 0.018006 (0.191353) | 0.461331 / 0.000490 (0.460841) | 0.000386 / 0.000200 (0.000186) | 0.000052 / 0.000054 (-0.000002) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.028405 / 0.037411 (-0.009006) | 0.081074 / 0.014526 (0.066548) | 0.094868 / 0.176557 (-0.081689) | 0.151050 / 0.737135 (-0.586085) | 0.095854 / 0.296338 (-0.200484) |\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) | 0.393957 / 0.215209 (0.178748) | 3.938649 / 2.077655 (1.860994) | 1.938190 / 1.504120 (0.434070) | 1.766458 / 1.541195 (0.225263) | 1.818028 / 1.468490 (0.349538) | 0.483926 / 4.584777 (-4.100851) | 3.641957 / 3.745712 (-0.103755) | 4.883845 / 5.269862 (-0.386016) | 2.960300 / 4.565676 (-1.605377) | 0.057227 / 0.424275 (-0.367048) | 0.007285 / 0.007607 (-0.000322) | 0.475928 / 0.226044 (0.249884) | 4.756757 / 2.268929 (2.487828) | 2.502659 / 55.444624 (-52.941966) | 2.178067 / 6.876477 (-4.698410) | 2.378298 / 2.142072 (0.236226) | 0.578639 / 4.805227 (-4.226588) | 0.132512 / 6.500664 (-6.368152) | 0.059656 / 0.075469 (-0.015813) |\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) | 1.272673 / 1.841788 (-0.569115) | 19.266884 / 8.074308 (11.192576) | 14.272930 / 10.191392 (4.081538) | 0.165897 / 0.680424 (-0.514527) | 0.018436 / 0.534201 (-0.515765) | 0.395177 / 0.579283 (-0.184107) | 0.420134 / 0.434364 (-0.014229) | 0.460781 / 0.540337 (-0.079557) | 0.645376 / 1.386936 (-0.741560) |\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) | 0.006504 / 0.011353 (-0.004849) | 0.003942 / 0.011008 (-0.007066) | 0.064936 / 0.038508 (0.026428) | 0.075015 / 0.023109 (0.051905) | 0.396871 / 0.275898 (0.120973) | 0.423448 / 0.323480 (0.099968) | 0.005239 / 0.007986 (-0.002747) | 0.003265 / 0.004328 (-0.001063) | 0.064910 / 0.004250 (0.060660) | 0.055006 / 0.037052 (0.017953) | 0.392818 / 0.258489 (0.134329) | 0.429735 / 0.293841 (0.135894) | 0.031847 / 0.128546 (-0.096699) | 0.008626 / 0.075646 (-0.067021) | 0.071591 / 0.419271 (-0.347681) | 0.049006 / 0.043533 (0.005473) | 0.384913 / 0.255139 (0.129774) | 0.408969 / 0.283200 (0.125769) | 0.023573 / 0.141683 (-0.118110) | 1.490271 / 1.452155 (0.038117) | 1.564620 / 1.492716 (0.071904) |\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) | 0.225917 / 0.018006 (0.207911) | 0.450369 / 0.000490 (0.449880) | 0.000375 / 0.000200 (0.000175) | 0.000055 / 0.000054 (0.000000) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.031196 / 0.037411 (-0.006215) | 0.090486 / 0.014526 (0.075960) | 0.102326 / 0.176557 (-0.074231) | 0.157483 / 0.737135 (-0.579653) | 0.103670 / 0.296338 (-0.192668) |\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) | 0.417577 / 0.215209 (0.202368) | 4.170798 / 2.077655 (2.093143) | 2.123689 / 1.504120 (0.619569) | 1.948231 / 1.541195 (0.407037) | 2.040277 / 1.468490 (0.571787) | 0.497919 / 4.584777 (-4.086858) | 3.633270 / 3.745712 (-0.112442) | 4.851698 / 5.269862 (-0.418164) | 2.691992 / 4.565676 (-1.873684) | 0.058641 / 0.424275 (-0.365634) | 0.007719 / 0.007607 (0.000112) | 0.500652 / 0.226044 (0.274607) | 4.988657 / 2.268929 (2.719728) | 2.604488 / 55.444624 (-52.840136) | 2.329829 / 6.876477 (-4.546648) | 2.468239 / 2.142072 (0.326167) | 0.598724 / 4.805227 (-4.206503) | 0.135959 / 6.500664 (-6.364706) | 0.061088 / 0.075469 (-0.014381) |\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) | 1.352107 / 1.841788 (-0.489681) | 19.973976 / 8.074308 (11.899668) | 14.292812 / 10.191392 (4.101420) | 0.163855 / 0.680424 (-0.516568) | 0.018402 / 0.534201 (-0.515799) | 0.393128 / 0.579283 (-0.186155) | 0.407379 / 0.434364 (-0.026985) | 0.462324 / 0.540337 (-0.078013) | 0.607501 / 1.386936 (-0.779435) |\n\n</details>\n</details>\n\n![](https://cml.dev/watermark.png#ae126ac974cad3050f90106e5909232140786811 \"CML watermark\")\n" ]

[ { "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" } ]

[ "Hi @patrickvonplaten \r\nI can have a look to this dataset later since I am trying to add the OpenSLR dataset https://github.com/huggingface/datasets/pull/2173\r\nHopefully I have enough space since the compressed file is 21GB. The release 3 is even bigger: 54GB :-0", "Resolved via https://github.com/huggingface/datasets/pull/4309" ]

[]

[]

[]

[ "Thank you! I just marked this as a draft PR. It probably would be better to create specific Array2D and Array3D classes as needed instead of a generic MultiArray for now, it should simplify the code a lot too so, I'll update it as such. Also i was meaning to reply earlier, but I wanted to thank you for the testing script you sent me earlier since it ended up being tremendously helpful. ", "Okay, I just converted the MultiArray class to Array2D, and got rid of all those \"globals()\"! \r\n\r\nThe main issues I had were that when including a \"pa.ExtensionType\" as a column, the ordinary methods to batch the data would not work and it would throw me some mysterious error, so I first cleaned up my code to order the row to match the schema (because when including extension types the row is disordered ) and then made each row a pa.Table and then concatenated all the tables. Also each n-dimensional vector class we implement will be size invariant which is some good news. ", "Okay awesome! I just added your suggestions and changed up my recursive functions. \r\n\r\nHere is the traceback for the when I use the original code in the write_on_file method:\r\n\r\n```\r\nTraceback (most recent call last):\r\n File \"<stdin>\", line 33, in <module>\r\n File \"/home/eltoto/nlp/src/nlp/arrow_writer.py\", line 214, in finalize\r\n self.write_on_file()\r\n File \"/home/eltoto/nlp/src/nlp/arrow_writer.py\", line 134, in write_on_file\r\n pa_array = pa.array(self.current_rows, type=self._type)\r\n File \"pyarrow/array.pxi\", line 269, in pyarrow.lib.array\r\n File \"pyarrow/array.pxi\", line 38, in pyarrow.lib._sequence_to_array\r\n File \"pyarrow/error.pxi\", line 106, in pyarrow.lib.check_status\r\npyarrow.lib.ArrowNotImplementedError: MakeBuilder: cannot construct builder for type extension<arrow.py_extension_type>\r\n\r\nshell returned 1\r\n```\r\n\r\nI think when trying to cast an extension array within a list of dictionaries, some method gets called that bugs out Arrow and somehow doesn't get called when adding a single row to a a table and then appending multiple tables together. I tinkered with this for a while but could not find any workaround. \r\n\r\nIn the case that this new method causes bad compression/worse performance, we can explicitly set the batch size in the pa.Table.to_batches(***batch_size***) method, which will return a list of batches. Perhaps, we can check that the batch size is not too large converting the table to batches after X many rows are appended to it by following the batch_size check below.", "> I think when trying to cast an extension array within a list of dictionaries, some method gets called that bugs out Arrow and somehow doesn't get called when adding a single row to a a table and then appending multiple tables together. I tinkered with this for a while but could not find any workaround.\r\n\r\nIndeed that's weird.\r\n\r\n> In the case that this new method causes bad compression/worse performance, we can explicitly set the batch size in the pa.Table.to_batches(batch_size) method, which will return a list of batches. Perhaps, we can check that the batch size is not too large converting the table to batches after X many rows are appended to it by following the batch_size check below.\r\n\r\nThe argument of `pa.Table.to_batches` is not `batch_size` but `max_chunksize`, which means that right now it would have no effects (each chunk is of length 1).\r\n\r\nWe can fix that just by doing `entries.combine_chunks().to_batches(batch_size)`. In that case it would write by chunk of 1000 which is what we want. I don't think it will slow down the writing by much, but we may have to do a benchmark just to make sure. If speed is ok we could even replace the original code to always write chunks this way.\r\n\r\nDo you still have errors that need to be fixed ?", "@lhoestq Nope all should be good! \r\n\r\nWould you like me to add the entries.combine_chunks().to_batch_size() code + benchmark?", "> @lhoestq Nope all should be good!\r\n\r\nAwesome :)\r\n\r\nI think it would be good to start to add some tests then.\r\nYou already have `test_multi_array.py` which is a good start, maybe you can place it in /tests and make it a `unittest.TestCase` ?\r\n\r\n> Would you like me to add the entries.combine_chunks().to_batch_size() code + benchmark?\r\n\r\nThat would be interesting. We don't want reading/writing to be the bottleneck of dataset processing for example in terms of speed. Maybe we could test the write + read speed of different datasets:\r\n- write speed + read speed a dataset with `nlp.Array2D` features\r\n- write speed + read speed a dataset with `nlp.Sequence(nlp.Sequence(nlp.Value(\"float32\")))` features\r\n- write speed + read speed a dataset with `nlp.Sequence(nlp.Value(\"float32\"))` features (same data but flatten)\r\nIt will be interesting to see the influence of `.combine_chunks()` on the `Array2D` test too.\r\n\r\nWhat do you think ?", "Well actually it looks like we're still having the `print(dataset[0])` error no ?", "I just tested your code to try to understand better.\r\n\r\n\r\n- First thing you must know is that we've switched from `dataset._data.to_pandas` to `dataset._data.to_pydict` by default when we call `dataset[0]` in #423 . Right now it raises an error but it can be fixed by adding this method to `ExtensionArray2D`:\r\n\r\n```python\r\n def to_pylist(self):\r\n return self.to_numpy().tolist()\r\n```\r\n\r\n- Second, I noticed that `ExtensionArray2D.to_numpy()` always return a (5, 5) shape in your example. I thought `ExtensionArray` was for possibly multiple examples and so I was expecting a shape like (1, 5, 5) for example. Did I miss something ?\r\nTherefore when I apply the fix I mentioned (adding to_pylist), it returns one example per row in each image (in your example of 2 images of shape 5x5, I get `len(dataset._data.to_pydict()[\"image\"]) == 10 # True`)\r\n\r\n[EDIT] I changed the reshape step in `ExtensionArray2D.to_numpy()` by\r\n```python\r\nnumpy_arr = numpy_arr.reshape(len(self), *ExtensionArray2D._construct_shape(self.storage))\r\n```\r\nand it did the job: `len(dataset._data.to_pydict()[\"image\"]) == 2 # True`\r\n\r\n- Finally, I was able to make `to_pandas` work though, by implementing custom array dtype in pandas with arrow conversion (I got inspiration from [here](https://gist.github.com/Eastsun/a59fb0438f65e8643cd61d8c98ec4c08) and [here](https://pandas.pydata.org/pandas-docs/version/1.0.0/development/extending.html#compatibility-with-apache-arrow))\r\n\r\nMaybe you could add me in your repo so I can open a PR to add these changes to your branch ?", "`combine_chunks` doesn't seem to work btw:\r\n`ArrowNotImplementedError: concatenation of extension<arrow.py_extension_type>`", "> > @lhoestq Nope all should be good!\r\n> \r\n> Awesome :)\r\n> \r\n> I think it would be good to start to add some tests then.\r\n> You already have `test_multi_array.py` which is a good start, maybe you can place it in /tests and make it a `unittest.TestCase` ?\r\n> \r\n> > Would you like me to add the entries.combine_chunks().to_batch_size() code + benchmark?\r\n> \r\n> That would be interesting. We don't want reading/writing to be the bottleneck of dataset processing for example in terms of speed. Maybe we could test the write + read speed of different datasets:\r\n> \r\n> * write speed + read speed a dataset with `nlp.Array2D` features\r\n> * write speed + read speed a dataset with `nlp.Sequence(nlp.Sequence(nlp.Value(\"float32\")))` features\r\n> * write speed + read speed a dataset with `nlp.Sequence(nlp.Value(\"float32\"))` features (same data but flatten)\r\n> It will be interesting to see the influence of `.combine_chunks()` on the `Array2D` test too.\r\n> \r\n> What do you think ?\r\n\r\nYa! that should be no problem at all, Ill use the timeit module and get back to you with the results sometime over the weekend.", "Thank you for all your help getting the pandas and row indexing for the dataset to work! For `print(dataset[0])`, I considered the workaround of doing `print(dataset[\"col_name\"][0])` a temporary solution, but ya, I was never able to figure out how to previously get it to work. I'll add you to my repo right now, let me know if you do not see the invite. Also lastly, it is strange how the to_batches method is not working, so I can check that out while I add some speed tests + add the multi dim test under the unit tests this weekend. ", "I created the PR :)\r\nI also tested `to_batches` and it works on my side", "Sorry for the bit of delay! I just added the tests, the PR into my fork, and some speed tests. It should be fairly easy to add more tests if we need. Do you think there is anything else to checkout?", "Cool thanks for adding the tests :) \r\n\r\nNext step is merge master into this branch.\r\nNot sure I understand what you did in your last commit, but it looks like you discarded all the changes from master ^^'\r\n\r\nWe've done some changes in the features logic on master, so let me know if you need help merging it.\r\n\r\nAs soon as we've merged from master, we'll have to make sure that we have extensive tests and we'll be good to do !\r\nAbout the lxmert dataset, we can probably keep it for another PR as soon as we have working 2d features. What do you think ?", "We might want to merge this after tomorrow's release though to avoid potential side effects @lhoestq ", "Yep I'm sure we can have it not for tomorrow's release but for the next one ;)", "haha, when I tried to rebase, I ran into some conflicts. In that last commit, I restored the features.py from the previous commit on the branch in my fork because upon updating to master, the pandasdtypemanger and pandas extension types disappeared. If you actually could help me with merging in what is needed, that would actually help a lot. \r\n\r\nOther than that, ya let me go ahead and move the dataloader code out of this PR. Perhaps we could discuss in the slack channelk soon about what to do with that because we can either just support the pretraining corpus for lxmert or try to implement the full COCO and visual genome datasets (+VQA +GQA) which im sure people would be pretty happy about. \r\n\r\nAlso we can talk more tests soon too when you are free. \r\n\r\nGoodluck on the release tomorrow guys!", "Not sure why github thinks there are conflicts here, as I just rebased from the current master branch.\r\nMerging into master locally works on my side without conflicts\r\n```\r\ngit checkout master\r\ngit reset --hard origin/master\r\ngit merge --no-ff eltoto1219/support_multi_dim_tensors_for_images\r\nMerge made by the 'recursive' strategy.\r\n datasets/lxmert_pretraining_beta/lxmert_pretraining_beta.py | 89 +++++++++++++++++++++++++++++++++++++\r\n datasets/lxmert_pretraining_beta/test_multi_array.py | 45 +++++++++++++++++++\r\n datasets/lxmert_pretraining_beta/to_arrow_data.py | 371 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++\r\n src/nlp/arrow_dataset.py | 24 +++++-----\r\n src/nlp/arrow_writer.py | 22 ++++++++--\r\n src/nlp/features.py | 229 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++---\r\n tests/test_array_2d.py | 210 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++\r\n 7 files changed, 969 insertions(+), 21 deletions(-)\r\n create mode 100644 datasets/lxmert_pretraining_beta/lxmert_pretraining_beta.py\r\n create mode 100644 datasets/lxmert_pretraining_beta/test_multi_array.py\r\n create mode 100644 datasets/lxmert_pretraining_beta/to_arrow_data.py\r\n create mode 100644 tests/test_array_2d.py\r\n```", "I put everything inside one commit from the master branch but the merge conflicts on github'side were still there for some reason.\r\nClosing and re-opening the PR fixed the conflict check on github's side.", "Almost done ! It still needs a pass on the docs/comments and maybe a few more tests.\r\n\r\nI had to do several changes for type inference in the ArrowWriter to make it support custom types.", "Ok this is now ready for review ! Thanks for your awesome work in this @eltoto1219 \r\n\r\nSummary of the changes:\r\n- added new feature type `Array2D`, that can be instantiated like `Array2D(\"float32\")` for example\r\n- added pyarrow extension type `Array2DExtensionType` and array `Array2DExtensionArray` that take care of converting from and to arrow. `Array2DExtensionType`'s storage is a list of list of any pyarrow array.\r\n- added pandas extension type `PandasArrayExtensionType` and array `PandasArrayExtensionArray` for conversion from and to arrow/python objects\r\n- refactor of the `ArrowWriter` write and write_batch functions to support extension types while preserving the type inference behavior.\r\n- added a utility object `TypedSequence` that is helpful to combine extension arrays and type inference inside the writer's methods.\r\n- added speed test for sequences writing (printed as warnings in pytest)\r\n- breaking: set disable_nullable to False by default as pyarrow's type inference returns nullable fields\r\n\r\nAnd there are plenty of new tests, mainly in `test_array2d.py` and `test_arrow_writer.py`.\r\n\r\nNote that there are some collisions in `arrow_dataset.py` with #513 so let's be careful when we'll merge this one.\r\n\r\nI know this is a big PR so feel free to ask questions", "I'll add Array3D, 4D.. tomorrow but it should take only a few lines. The rest won't change", "I took your comments into account and I added Array[3-5]D.\r\nI changed the storage type to fixed lengths lists. I had to update the `to_numpy` function because of that. Indeed slicing a FixedLengthListArray returns a view a of the original array, while in the previous case slicing a ListArray copies the storage.\r\n" ]

[]

[ "<details>\n<summary>Show benchmarks</summary>\n\nPyArrow==8.0.0\n\n<details>\n<summary>Show updated benchmarks!</summary>\n\n### Benchmark: benchmark_array_xd.json\n\n| metric | read_batch_formatted_as_numpy after write_array2d | read_batch_formatted_as_numpy after write_flattened_sequence | read_batch_formatted_as_numpy after write_nested_sequence | read_batch_unformated after write_array2d | read_batch_unformated after write_flattened_sequence | read_batch_unformated after write_nested_sequence | read_col_formatted_as_numpy after write_array2d | read_col_formatted_as_numpy after write_flattened_sequence | read_col_formatted_as_numpy after write_nested_sequence | read_col_unformated after write_array2d | read_col_unformated after write_flattened_sequence | read_col_unformated after write_nested_sequence | read_formatted_as_numpy after write_array2d | read_formatted_as_numpy after write_flattened_sequence | read_formatted_as_numpy after write_nested_sequence | read_unformated after write_array2d | read_unformated after write_flattened_sequence | read_unformated after write_nested_sequence | write_array2d | write_flattened_sequence | write_nested_sequence |\n|--------|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 0.007489 / 0.011353 (-0.003864) | 0.004914 / 0.011008 (-0.006095) | 0.111626 / 0.038508 (0.073117) | 0.037920 / 0.023109 (0.014811) | 0.350571 / 0.275898 (0.074673) | 0.389667 / 0.323480 (0.066187) | 0.006309 / 0.007986 (-0.001676) | 0.005488 / 0.004328 (0.001160) | 0.083962 / 0.004250 (0.079712) | 0.050728 / 0.037052 (0.013675) | 0.360997 / 0.258489 (0.102508) | 0.392736 / 0.293841 (0.098895) | 0.031975 / 0.128546 (-0.096571) | 0.009941 / 0.075646 (-0.065705) | 0.379840 / 0.419271 (-0.039432) | 0.056522 / 0.043533 (0.012989) | 0.359379 / 0.255139 (0.104240) | 0.384487 / 0.283200 (0.101287) | 0.117523 / 0.141683 (-0.024160) | 1.683639 / 1.452155 (0.231485) | 1.791645 / 1.492716 (0.298929) |\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) | 0.236862 / 0.018006 (0.218856) | 0.481208 / 0.000490 (0.480719) | 0.007455 / 0.000200 (0.007255) | 0.000111 / 0.000054 (0.000057) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.030854 / 0.037411 (-0.006557) | 0.126892 / 0.014526 (0.112367) | 0.139207 / 0.176557 (-0.037350) | 0.206447 / 0.737135 (-0.530689) | 0.143095 / 0.296338 (-0.153244) |\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) | 0.474677 / 0.215209 (0.259468) | 4.699534 / 2.077655 (2.621879) | 2.152102 / 1.504120 (0.647983) | 1.934815 / 1.541195 (0.393620) | 1.986448 / 1.468490 (0.517958) | 0.607184 / 4.584777 (-3.977593) | 4.480385 / 3.745712 (0.734673) | 2.074729 / 5.269862 (-3.195132) | 1.182383 / 4.565676 (-3.383294) | 0.075624 / 0.424275 (-0.348651) | 0.014046 / 0.007607 (0.006439) | 0.598859 / 0.226044 (0.372814) | 5.959551 / 2.268929 (3.690622) | 2.700851 / 55.444624 (-52.743773) | 2.303775 / 6.876477 (-4.572702) | 2.456441 / 2.142072 (0.314369) | 0.747185 / 4.805227 (-4.058042) | 0.165787 / 6.500664 (-6.334878) | 0.075817 / 0.075469 (0.000348) |\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) | 1.411859 / 1.841788 (-0.429928) | 17.375495 / 8.074308 (9.301187) | 15.187098 / 10.191392 (4.995706) | 0.169953 / 0.680424 (-0.510471) | 0.020204 / 0.534201 (-0.513997) | 0.461424 / 0.579283 (-0.117859) | 0.494443 / 0.434364 (0.060080) | 0.544583 / 0.540337 (0.004246) | 0.648231 / 1.386936 (-0.738705) |\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) | 0.007785 / 0.011353 (-0.003568) | 0.005314 / 0.011008 (-0.005694) | 0.087273 / 0.038508 (0.048765) | 0.037810 / 0.023109 (0.014701) | 0.425473 / 0.275898 (0.149575) | 0.459976 / 0.323480 (0.136497) | 0.007270 / 0.007986 (-0.000716) | 0.004631 / 0.004328 (0.000303) | 0.087063 / 0.004250 (0.082812) | 0.052630 / 0.037052 (0.015578) | 0.432384 / 0.258489 (0.173895) | 0.500291 / 0.293841 (0.206450) | 0.033144 / 0.128546 (-0.095402) | 0.010101 / 0.075646 (-0.065545) | 0.096068 / 0.419271 (-0.323204) | 0.062750 / 0.043533 (0.019217) | 0.419308 / 0.255139 (0.164169) | 0.437099 / 0.283200 (0.153900) | 0.122289 / 0.141683 (-0.019394) | 1.737829 / 1.452155 (0.285674) | 1.851481 / 1.492716 (0.358765) |\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) | 0.014277 / 0.018006 (-0.003729) | 0.489835 / 0.000490 (0.489345) | 0.008423 / 0.000200 (0.008223) | 0.000188 / 0.000054 (0.000134) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.032966 / 0.037411 (-0.004445) | 0.130069 / 0.014526 (0.115544) | 0.144372 / 0.176557 (-0.032185) | 0.200400 / 0.737135 (-0.536735) | 0.149384 / 0.296338 (-0.146954) |\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) | 0.511542 / 0.215209 (0.296333) | 5.093879 / 2.077655 (3.016225) | 2.572088 / 1.504120 (1.067968) | 2.339118 / 1.541195 (0.797923) | 2.441637 / 1.468490 (0.973147) | 0.614818 / 4.584777 (-3.969959) | 4.724441 / 3.745712 (0.978729) | 5.431978 / 5.269862 (0.162116) | 2.257794 / 4.565676 (-2.307883) | 0.078109 / 0.424275 (-0.346166) | 0.013821 / 0.007607 (0.006214) | 0.639232 / 0.226044 (0.413188) | 6.424623 / 2.268929 (4.155694) | 3.163018 / 55.444624 (-52.281606) | 2.756786 / 6.876477 (-4.119690) | 2.808655 / 2.142072 (0.666583) | 0.745843 / 4.805227 (-4.059385) | 0.165562 / 6.500664 (-6.335102) | 0.076610 / 0.075469 (0.001141) |\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) | 1.738630 / 1.841788 (-0.103158) | 18.073573 / 8.074308 (9.999265) | 16.482820 / 10.191392 (6.291428) | 0.213233 / 0.680424 (-0.467191) | 0.022839 / 0.534201 (-0.511362) | 0.487043 / 0.579283 (-0.092240) | 0.512518 / 0.434364 (0.078154) | 0.549365 / 0.540337 (0.009028) | 0.656612 / 1.386936 (-0.730324) |\n\n</details>\n</details>\n\n![](https://cml.dev/watermark.png#288e92b03bd4ec91c10c8a529b32631cfaba9fb7 \"CML watermark\")\n", "Good idea!\r\n\r\nI was wondering: if we want to optimize the balance between the size of downloading a row group, and the number of rows in the group, would it make sense to compute the row group size by checking the average size of the rows?\r\n\r\neg. 32x32 images could have a larger row group size than full HD images, no? Relying on the size would even remove the need to check the column types.\r\n\r\n(in this proposal, we could use the computed row group size, eg 837, or use the nearest row group size in a list of values: 10, 100, 1000, 10000)", "Probably, but I would go for a simpler solution first :p", "Sure! I wanted to understand if the idea made sense or not, but it's not for this PR.", "I think it will be more useful for people who use the viewer and won't impact sequential io that much.", "DuckDB [paragraph](https://duckdb.org/docs/data/parquet/tips.html#selecting-a-row_group_size) that explains how to choose the `row_group_size`. Our default shard size is 500 MB in `push_to_hub`, so, ideally, we should aim for 64 MB row groups (and make this part configurable for power users 🙂).\r\n\r\nSo, before merging this PR, let's add a TODO or open an issue as a reminder that this can be improved.", "I moved the config values, improved the features check and mentioned the improvements we could do in the docstring :)", "_The documentation is not available anymore as the PR was closed or merged._", "<details>\n<summary>Show benchmarks</summary>\n\nPyArrow==8.0.0\n\n<details>\n<summary>Show updated benchmarks!</summary>\n\n### Benchmark: benchmark_array_xd.json\n\n| metric | read_batch_formatted_as_numpy after write_array2d | read_batch_formatted_as_numpy after write_flattened_sequence | read_batch_formatted_as_numpy after write_nested_sequence | read_batch_unformated after write_array2d | read_batch_unformated after write_flattened_sequence | read_batch_unformated after write_nested_sequence | read_col_formatted_as_numpy after write_array2d | read_col_formatted_as_numpy after write_flattened_sequence | read_col_formatted_as_numpy after write_nested_sequence | read_col_unformated after write_array2d | read_col_unformated after write_flattened_sequence | read_col_unformated after write_nested_sequence | read_formatted_as_numpy after write_array2d | read_formatted_as_numpy after write_flattened_sequence | read_formatted_as_numpy after write_nested_sequence | read_unformated after write_array2d | read_unformated after write_flattened_sequence | read_unformated after write_nested_sequence | write_array2d | write_flattened_sequence | write_nested_sequence |\n|--------|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 0.006211 / 0.011353 (-0.005141) | 0.004244 / 0.011008 (-0.006764) | 0.097941 / 0.038508 (0.059433) | 0.028564 / 0.023109 (0.005455) | 0.299651 / 0.275898 (0.023753) | 0.340694 / 0.323480 (0.017214) | 0.005161 / 0.007986 (-0.002824) | 0.004764 / 0.004328 (0.000435) | 0.075505 / 0.004250 (0.071255) | 0.039656 / 0.037052 (0.002603) | 0.309242 / 0.258489 (0.050753) | 0.350783 / 0.293841 (0.056942) | 0.025145 / 0.128546 (-0.103401) | 0.008498 / 0.075646 (-0.067148) | 0.317657 / 0.419271 (-0.101615) | 0.043926 / 0.043533 (0.000394) | 0.305915 / 0.255139 (0.050776) | 0.331630 / 0.283200 (0.048430) | 0.088564 / 0.141683 (-0.053119) | 1.533175 / 1.452155 (0.081021) | 1.581017 / 1.492716 (0.088301) |\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) | 0.206032 / 0.018006 (0.188025) | 0.433446 / 0.000490 (0.432956) | 0.003955 / 0.000200 (0.003755) | 0.000095 / 0.000054 (0.000040) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.023468 / 0.037411 (-0.013943) | 0.103292 / 0.014526 (0.088766) | 0.107234 / 0.176557 (-0.069322) | 0.168525 / 0.737135 (-0.568610) | 0.113218 / 0.296338 (-0.183120) |\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) | 0.431085 / 0.215209 (0.215875) | 4.302082 / 2.077655 (2.224427) | 2.068290 / 1.504120 (0.564171) | 1.850718 / 1.541195 (0.309523) | 1.964261 / 1.468490 (0.495771) | 0.547562 / 4.584777 (-4.037215) | 3.410739 / 3.745712 (-0.334974) | 1.779640 / 5.269862 (-3.490221) | 1.005466 / 4.565676 (-3.560210) | 0.066250 / 0.424275 (-0.358025) | 0.011877 / 0.007607 (0.004270) | 0.525185 / 0.226044 (0.299141) | 5.234786 / 2.268929 (2.965857) | 2.398045 / 55.444624 (-53.046580) | 2.073020 / 6.876477 (-4.803457) | 2.210753 / 2.142072 (0.068680) | 0.654897 / 4.805227 (-4.150331) | 0.134639 / 6.500664 (-6.366025) | 0.067050 / 0.075469 (-0.008419) |\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) | 1.180210 / 1.841788 (-0.661577) | 13.613091 / 8.074308 (5.538783) | 13.441837 / 10.191392 (3.250445) | 0.146048 / 0.680424 (-0.534376) | 0.016505 / 0.534201 (-0.517696) | 0.363210 / 0.579283 (-0.216073) | 0.405484 / 0.434364 (-0.028880) | 0.428712 / 0.540337 (-0.111625) | 0.522300 / 1.386936 (-0.864636) |\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) | 0.006147 / 0.011353 (-0.005206) | 0.004161 / 0.011008 (-0.006847) | 0.075861 / 0.038508 (0.037353) | 0.027948 / 0.023109 (0.004839) | 0.362466 / 0.275898 (0.086568) | 0.398227 / 0.323480 (0.074747) | 0.005014 / 0.007986 (-0.002972) | 0.004772 / 0.004328 (0.000444) | 0.075674 / 0.004250 (0.071423) | 0.039158 / 0.037052 (0.002106) | 0.363567 / 0.258489 (0.105078) | 0.410378 / 0.293841 (0.116537) | 0.025510 / 0.128546 (-0.103036) | 0.008528 / 0.075646 (-0.067118) | 0.081803 / 0.419271 (-0.337468) | 0.040954 / 0.043533 (-0.002579) | 0.358492 / 0.255139 (0.103353) | 0.381345 / 0.283200 (0.098145) | 0.092347 / 0.141683 (-0.049336) | 1.567695 / 1.452155 (0.115540) | 1.668412 / 1.492716 (0.175696) |\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) | 0.203367 / 0.018006 (0.185360) | 0.424642 / 0.000490 (0.424152) | 0.002451 / 0.000200 (0.002251) | 0.000071 / 0.000054 (0.000017) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.026129 / 0.037411 (-0.011282) | 0.102564 / 0.014526 (0.088039) | 0.110583 / 0.176557 (-0.065973) | 0.164332 / 0.737135 (-0.572804) | 0.115706 / 0.296338 (-0.180632) |\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) | 0.468925 / 0.215209 (0.253716) | 4.657266 / 2.077655 (2.579612) | 2.423280 / 1.504120 (0.919160) | 2.236284 / 1.541195 (0.695089) | 2.323019 / 1.468490 (0.854529) | 0.548120 / 4.584777 (-4.036657) | 3.455602 / 3.745712 (-0.290110) | 1.730421 / 5.269862 (-3.539441) | 1.006089 / 4.565676 (-3.559588) | 0.067478 / 0.424275 (-0.356797) | 0.011465 / 0.007607 (0.003857) | 0.574235 / 0.226044 (0.348190) | 5.744404 / 2.268929 (3.475475) | 2.882225 / 55.444624 (-52.562400) | 2.618246 / 6.876477 (-4.258231) | 2.642920 / 2.142072 (0.500847) | 0.661441 / 4.805227 (-4.143787) | 0.137358 / 6.500664 (-6.363306) | 0.070372 / 0.075469 (-0.005097) |\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) | 1.333815 / 1.841788 (-0.507973) | 14.689667 / 8.074308 (6.615359) | 14.362294 / 10.191392 (4.170902) | 0.152011 / 0.680424 (-0.528413) | 0.016869 / 0.534201 (-0.517332) | 0.370433 / 0.579283 (-0.208851) | 0.399642 / 0.434364 (-0.034722) | 0.433759 / 0.540337 (-0.106578) | 0.525443 / 1.386936 (-0.861493) |\n\n</details>\n</details>\n\n![](https://cml.dev/watermark.png#09e9f9a88edd9055b5c540e3d83b5a11d48f8ba8 \"CML watermark\")\n", "<details>\n<summary>Show benchmarks</summary>\n\nPyArrow==8.0.0\n\n<details>\n<summary>Show updated benchmarks!</summary>\n\n### Benchmark: benchmark_array_xd.json\n\n| metric | read_batch_formatted_as_numpy after write_array2d | read_batch_formatted_as_numpy after write_flattened_sequence | read_batch_formatted_as_numpy after write_nested_sequence | read_batch_unformated after write_array2d | read_batch_unformated after write_flattened_sequence | read_batch_unformated after write_nested_sequence | read_col_formatted_as_numpy after write_array2d | read_col_formatted_as_numpy after write_flattened_sequence | read_col_formatted_as_numpy after write_nested_sequence | read_col_unformated after write_array2d | read_col_unformated after write_flattened_sequence | read_col_unformated after write_nested_sequence | read_formatted_as_numpy after write_array2d | read_formatted_as_numpy after write_flattened_sequence | read_formatted_as_numpy after write_nested_sequence | read_unformated after write_array2d | read_unformated after write_flattened_sequence | read_unformated after write_nested_sequence | write_array2d | write_flattened_sequence | write_nested_sequence |\n|--------|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|\n| new / old (diff) | 0.006564 / 0.011353 (-0.004789) | 0.004350 / 0.011008 (-0.006658) | 0.096277 / 0.038508 (0.057769) | 0.032956 / 0.023109 (0.009847) | 0.303675 / 0.275898 (0.027777) | 0.336384 / 0.323480 (0.012904) | 0.005789 / 0.007986 (-0.002197) | 0.003957 / 0.004328 (-0.000371) | 0.073990 / 0.004250 (0.069740) | 0.050974 / 0.037052 (0.013922) | 0.321754 / 0.258489 (0.063265) | 0.349489 / 0.293841 (0.055648) | 0.031138 / 0.128546 (-0.097409) | 0.009000 / 0.075646 (-0.066646) | 0.325445 / 0.419271 (-0.093826) | 0.070173 / 0.043533 (0.026640) | 0.304706 / 0.255139 (0.049567) | 0.321803 / 0.283200 (0.038603) | 0.109405 / 0.141683 (-0.032278) | 1.489812 / 1.452155 (0.037657) | 1.577729 / 1.492716 (0.085013) |\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) | 0.287187 / 0.018006 (0.269181) | 0.527625 / 0.000490 (0.527135) | 0.006533 / 0.000200 (0.006333) | 0.000090 / 0.000054 (0.000036) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.026659 / 0.037411 (-0.010752) | 0.106236 / 0.014526 (0.091710) | 0.118615 / 0.176557 (-0.057941) | 0.173156 / 0.737135 (-0.563979) | 0.122883 / 0.296338 (-0.173456) |\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) | 0.407189 / 0.215209 (0.191980) | 4.055732 / 2.077655 (1.978078) | 1.865594 / 1.504120 (0.361474) | 1.664325 / 1.541195 (0.123130) | 1.668961 / 1.468490 (0.200471) | 0.521207 / 4.584777 (-4.063570) | 3.740424 / 3.745712 (-0.005288) | 3.431973 / 5.269862 (-1.837889) | 1.636669 / 4.565676 (-2.929008) | 0.065271 / 0.424275 (-0.359005) | 0.012151 / 0.007607 (0.004544) | 0.514233 / 0.226044 (0.288189) | 5.110150 / 2.268929 (2.841222) | 2.264340 / 55.444624 (-53.180284) | 1.940428 / 6.876477 (-4.936049) | 2.042286 / 2.142072 (-0.099787) | 0.639200 / 4.805227 (-4.166028) | 0.139537 / 6.500664 (-6.361127) | 0.063195 / 0.075469 (-0.012274) |\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) | 1.179501 / 1.841788 (-0.662286) | 14.600133 / 8.074308 (6.525825) | 14.902137 / 10.191392 (4.710745) | 0.144509 / 0.680424 (-0.535915) | 0.017449 / 0.534201 (-0.516752) | 0.393135 / 0.579283 (-0.186148) | 0.413103 / 0.434364 (-0.021261) | 0.459897 / 0.540337 (-0.080440) | 0.552602 / 1.386936 (-0.834334) |\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) | 0.006891 / 0.011353 (-0.004462) | 0.004633 / 0.011008 (-0.006375) | 0.073093 / 0.038508 (0.034585) | 0.032509 / 0.023109 (0.009399) | 0.348332 / 0.275898 (0.072434) | 0.381920 / 0.323480 (0.058440) | 0.005978 / 0.007986 (-0.002007) | 0.005360 / 0.004328 (0.001032) | 0.074307 / 0.004250 (0.070056) | 0.049668 / 0.037052 (0.012615) | 0.354713 / 0.258489 (0.096224) | 0.398521 / 0.293841 (0.104681) | 0.032013 / 0.128546 (-0.096534) | 0.008890 / 0.075646 (-0.066756) | 0.080013 / 0.419271 (-0.339259) | 0.051820 / 0.043533 (0.008288) | 0.349730 / 0.255139 (0.094591) | 0.369267 / 0.283200 (0.086067) | 0.103874 / 0.141683 (-0.037809) | 1.484148 / 1.452155 (0.031993) | 1.573927 / 1.492716 (0.081211) |\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) | 0.009699 / 0.018006 (-0.008307) | 0.511176 / 0.000490 (0.510686) | 0.002938 / 0.000200 (0.002738) | 0.000109 / 0.000054 (0.000054) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.027847 / 0.037411 (-0.009564) | 0.111565 / 0.014526 (0.097039) | 0.120625 / 0.176557 (-0.055932) | 0.172130 / 0.737135 (-0.565006) | 0.125949 / 0.296338 (-0.170389) |\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) | 0.430634 / 0.215209 (0.215424) | 4.315377 / 2.077655 (2.237722) | 2.070764 / 1.504120 (0.566644) | 1.881962 / 1.541195 (0.340767) | 1.904053 / 1.468490 (0.435563) | 0.524973 / 4.584777 (-4.059804) | 3.718359 / 3.745712 (-0.027353) | 3.415344 / 5.269862 (-1.854518) | 1.224568 / 4.565676 (-3.341108) | 0.065593 / 0.424275 (-0.358682) | 0.011643 / 0.007607 (0.004036) | 0.537050 / 0.226044 (0.311006) | 5.352155 / 2.268929 (3.083226) | 2.557361 / 55.444624 (-52.887263) | 2.217770 / 6.876477 (-4.658707) | 2.194975 / 2.142072 (0.052902) | 0.635142 / 4.805227 (-4.170085) | 0.140642 / 6.500664 (-6.360022) | 0.064690 / 0.075469 (-0.010779) |\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) | 1.266125 / 1.841788 (-0.575663) | 14.836413 / 8.074308 (6.762105) | 14.446870 / 10.191392 (4.255478) | 0.191545 / 0.680424 (-0.488878) | 0.017433 / 0.534201 (-0.516768) | 0.392296 / 0.579283 (-0.186987) | 0.420698 / 0.434364 (-0.013666) | 0.463225 / 0.540337 (-0.077112) | 0.556127 / 1.386936 (-0.830809) |\n\n</details>\n</details>\n\n![](https://cml.dev/watermark.png#7fcbe5b1575c8d162b65b9397b3dfda995a4e048 \"CML watermark\")\n" ]

[ { "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" } ]

[ "Closing this issue as the `task_templates` API has been deprecated." ]

[]

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

[]

[ "Will slim down the dummy files in the morning" ]

[]

[ "@lhoestq The test passes even without the try/except block!", "Hey, I'm a little late because I was caught up in the course work, but I double-checked this and it looks great. Thanks for fixing!" ]

[]

[ "These test errors are related to issue #4428 \r\n", "_The documentation is not available anymore as the PR was closed or merged._", "I only made a trivial modification in my commit https://github.com/huggingface/datasets/pull/4431/commits/402c893d35224d7828176717233909ac5f1e7b3e\r\n\r\nI have submitted a PR #4434 for the about issue.", "> Awesome thanks for adding this dataset :)\r\n> \r\n> I just have one comment about the licensing.\r\n> \r\n> Also it seems that you already have the dataset in https://huggingface.co/datasets/silver/personal_dialog, so it's unnecessary to add it here\r\n\r\nThank you very much for your comment.\r\n\r\nSo, should I close this PR?", "Thanks for fixing the licensing section :)\r\n\r\n> So, should I close this PR?\r\n\r\nYes you can close this PR, it's better if your dataset is under your namespace at https://huggingface.co/datasets/silver/personal_dialog :)\r\n\r\nDon't forget to update the licensing section on https://huggingface.co/datasets/silver/personal_dialog as well" ]

[]

[ "Hi, you can use the `cast_column` function to change the feature type from a `Value(int64)` to `ClassLabel`:\r\n\r\n```py\r\ndataset = dataset.cast_column(\"label\", ClassLabel(names=[\"label_1\", \"label_2\", \"label_3\"]))\r\nprint(dataset.features)\r\n{'text': Value(dtype='string', id=None),\r\n 'label': ClassLabel(names=['label_1', 'label_2', 'label_3'], id=None)}\r\n```", "thank you @stevhliu, its worked. " ]

[]

[ "@SBrandeis thanks for the review, I applied your suggested changes, but CI is failing now not sure about the error.", "Many thanks @Nilanshrajput !\r\nThe failing tests on CI are not related to your changes, merging master on your branch should fix them :)\r\nIf you're interested in what causes the CI to fail, checkout [this commit](https://github.com/huggingface/datasets/commit/9a0f1e20ca1e783cb14c1ab1cc2f54b0b5b201e8)", "@SBrandeis done!\r\n", "Hello @Nilanshrajput, your PR includes changes from other branches too now (485 files changed)\r\nWould you mind creating another branch from master with your changes and opening a new PR?", "@SBrandeis sorry i messed up the git history, #1587 I opened this new pr! ", "closing in favor of #1587 " ]

[ { "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._" ]

[]

[]

[]