Datasets:

---

Francesco-A

/

github-issues_huggingface-datasets

like 0

[ { "color": "a2eeef", "default": true, "description": "New feature or request", "id": 1935892871, "name": "enhancement", "node_id": "MDU6TGFiZWwxOTM1ODkyODcx", "url": "https://api.github.com/repos/huggingface/datasets/labels/enhancement" }, { "color": "c5def5", "default": false, "description": "Generic discussion on the library", "id": 2067400324, "name": "generic discussion", "node_id": "MDU6TGFiZWwyMDY3NDAwMzI0", "url": "https://api.github.com/repos/huggingface/datasets/labels/generic%20discussion" } ]

[]

[]

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

[]

[ "@vblagoje nice work, please add the README.md file and it would be ready", "@lhoestq @tanmoyio @yjernite please have a look at the dataset card. Don't forget that the dataset is still hosted on my private gs bucket and should eventually be moved to the HF bucket", "I will move the files soon and ping you when it's done and with the new URLs :) ", "Hi !\r\n\r\nI just moved the file to a HF bucket. It's available at https://storage.googleapis.com/huggingface-nlp/datasets/cc_news/cc_news.tar.gz\r\n\r\nSorry for the delay ^^'", "@lhoestq no worries, updated PR with the new URL and rebased to master\r\n" ]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_5239). All of your documentation changes will be reflected on that endpoint.", "I ended up moving `num_proc` to `AbstractDatasetReader.__init__` :)\r\n\r\nLet me know if it sounds good to you now" ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "The CI is failing because of issues in the dataset cards that are unrelated to this PR - merging" ]

[ { "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 @lewtun, thanks for reporting.\r\n\r\nApparently, `pyarrow.json` tries to cast timestamp-like fields in your JSON file to pyarrow timestamp type, and it fails with `ArrowNotImplementedError`.\r\n\r\nI will investigate if there is a way to tell pyarrow not to try that timestamp casting.", "I think the issue is more complex than that...\r\n\r\nI just took one of your JSON lines and pyarrow.json read it without problem.", "> I just took one of your JSON lines an pyarrow.json read it without problem.\r\n\r\nyes, and for some peculiar reason the error is non-deterministic (i was eventually able to load the whole dataset by just re-running the `load_dataset` cell multiple times 🤔)\r\n\r\nthanks for looking into this 🙏 !", "I think the error is generated by the `pyarrow.json.read()` option: `read_options=paj.ReadOptions(block_size=block_size)`...\r\ncc: @lhoestq ", "The code works fine on my side.\r\nNot sure what's going on here :/\r\n\r\nI remember we did a few changes in the JSON loader in #2638 , did you do an update `datasets` when debugging this ?\r\n", "OK after upgrading `datasets` to v1.12.1 the issue seems to have gone away. Closing this now :)", "Oops, I spoke too soon 😓 \r\n\r\nAfter deleting the cache and trying the above code snippet again I am hitting the same error. You can also reproduce it in the Colab notebook I linked to in the issue description. ", "@albertvillanova @lhoestq I noticed the same issue using datasets v1.12.1. Is there an update on when this could be fixed?", "Apparently it's possible to make it work by increasing the `block_size`, let me open a PR", "I just opened a PR with a fix, feel free to install `datasets` from source from source and let me know if it helps", "@zijwang did PR #3000 solve the problem for you? It did for me, so it all is good on your end we can close this issue. Thanks again to @lhoestq for the pyarrow magic 🤯 " ]

[]

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

[]

[ "We could add a \"threads\" parallel backend to `datasets.parallel.parallel_backend` to support downloading with threads but note that `download_and_extract` also decompresses archives, and this is a CPU-intensive task, which is not ideal for (Python) threads (good for IO-intensive tasks).", "> We could add a \"threads\" parallel backend to `datasets.parallel.parallel_backend` to support downloading with threads but note that `download_and_extract` also decompresses archives, and this is a CPU-intensive task, which is not ideal for (Python) threads (good for IO-intensive tasks).\r\n\r\nGreat! Download takes more time than extract, multiple threads can download in parallel, which can speed up a lot." ]

[ { "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, @abhishekkrthakur. 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." ]

[]

[]

[ { "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": "2edb81", "default": false, "description": "A bug in a dataset script provided in the library", "id": 2067388877, "name": "dataset bug", "node_id": "MDU6TGFiZWwyMDY3Mzg4ODc3", "url": "https://api.github.com/repos/huggingface/datasets/labels/dataset%20bug" } ]

[]

[ { "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 @julien-c,\r\n\r\nOut of the box, we only support JSON lines (NDJSON) data files, but your data file is a regular JSON file. The reason is we use `pyarrow.json.read_json` and this only supports line-delimited JSON. " ]

[]

[ "Awesome, looking forward to using it :)", "Few additional comments:\r\n* the current API doesn't meet the requirements mentioned in #3145 (e.g. image mime-type). However, this will be doable soon as we also plan to store image bytes alongside paths in arrow files (see https://github.com/huggingface/datasets/pull/3129#discussion_r738426187). Then, PIL can return the correct mime-type: \r\n ```python\r\n from PIL import Image\r\n import io\r\n\r\n mimetype = Image.open(io.BytesIO(image_bytes)).get_format_mimetype()\r\n ``` \r\n I plan to add this change in a separate PR.\r\n* currently, I'm returning an `np.ndarray` object after decoding for consistency with the Audio feature. However, the vision models from Transformers prefer an `Image` object to avoid the `Image.fromarray` call in the corresponding feature extractors (see [this warning](https://huggingface.co/transformers/master/model_doc/vit.html#transformers.ViTFeatureExtractor.__call__) in the Transformers docs) cc @NielsRogge \r\n\r\nSo I'm not entirely sure whether to return only a NumPy array, only a PIL Image, or both when decoding. The last point worries me because we shouldn't provide an API that leads to a warning in Transformers (in the docs, not in code :)). At the same time, it makes sense to preserve consistency with the Audio feature and return a NumPy array. \r\n\r\nThat's why I would appreciate your opinions on this.", "That is a good question. Also pinging @nateraw .\r\n\r\nCurrently we only support returning numpy arrays because of numpy/tf/torch/jax formatting features that we have, and to keep things simple. See the [set_format docs](https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.set_format) for more info", "I don't think centering the discussion on what ViT expects is good, as the vision Transformers model are still in an experimental stage and we can adapt those depending on what you do here :-).\r\n\r\nIMO, the discussion should revolve on what a user will want to do with a vision dataset, and they will want to:\r\n- lazily decode their images\r\n- maybe apply data augmentation (for the training set)\r\n- resize to a fixed shape for batching\r\n\r\nThe libraries that provide step 2 and 3 either use PIL (thinking torchvision) or cv2 (thinking albumentations). NumPy does not have any function to resize an image or do basic data augmentation (like a rotate) so I think it shouldn't be the default format for an image dataset, PIL or cv2 (in an ideal world with the ability to switch between the two depending on what the users prefer) would be better.\r\n\r\nSide note: I will work on the vision integration in Transformers with Niels next month so please keep me in the loop for those awesome new vision features!", "@sgugger I completely agree with you, especially after trying to convert the `run_image_classification` script from Transformers to use this feature. The current API doesn't seem intuitive there due to the torchvision transforms, which, as you say, prefer PIL over NumPy arrays. \r\n\r\nSo the default format would return `Image` (PIL) / `np.ndarray` (cv2) and `set_format(numpy/tf/pt)` would return image tensors if I understand you correctly. IMO this makes a lot more sense (and flexibility) than the current API.", "Also, one additional library worth mentioning here is AugLy which supports image file paths and `PIL.Image.Image` objects.", "That's so nice !\r\n\r\nAlso I couldn't help myself so I've played with it already ^^\r\nI was agreeably surprised that with minor additions I managed to even allow this, which I find very satisfactory:\r\n```python\r\nimport PIL.Image\r\nfrom datasets import Dataset\r\n\r\npath = \"docs/source/imgs/datasets_logo_name.jpg\"\r\n\r\ndataset = Dataset.from_dict({\"img\": [PIL.Image.open(path)]})\r\nprint(dataset.features)\r\n# {'img': Image(id=None)}\r\nprint(dataset[0][\"img\"])\r\n# <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=1200x300 at 0x129DE4AC8>\r\n```\r\n\r\nLet me know if that's a behavior you'd also like to see \r\n\r\nEDIT: just pushed my changes on a branch, you can see the diff [here](https://github.com/mariosasko/datasets-1/compare/add-image-feature...huggingface:image-type-inference) if you want", "Thanks, @lhoestq! I like your change. Very elegant indeed.\r\n\r\nP.S. I have to write a big comment that explains all the changes/things left to consider. Will do that in the next few days!", "I'm marking this PR as ready for review.\r\n\r\nThanks to @sgugger's comment, the API is much more flexible now as it decodes images (lazily) as `PIL.Image.Image` objects and supports transforms directly on them.\r\n\r\nAlso, we no longer return paths explicitly (previously, we would return `{\"path\": image_path, \"image\": pil_image}`) for the following reasons:\r\n* what to return when reading an image from an URL or a NumPy array. We could set `path` to `None` in these situations, but IMO we should avoid redundant information.\r\n* returning a dict doesn't match nicely with the requirement of supporting image modifications - what to do if the user modifies both the image path and the image\r\n\r\n(Btw, for the images stored locally, you can access their paths with `dset[idx][\"image\"].filename`, or by avoiding decoding with `paths = [ex[\"path\"] for ex in dset]`. @lhoestq @albertvillanova WDYT about having an option to skip decoding for complex features, e. g. `Audio(decode=False)`? This way, the user can easily access the underlying data.)\r\n\r\nExamples of what you can do:\r\n```python\r\n# load local images\r\ndset = Dataset.from_dict(\"image\": [local_image_path], features=Features({\"images\": Image()}))\r\n# load remote images (we got this for free by adding support for streaming)\r\ndset = Dataset.from_dict(\"image\": [image_url], features=Features({\"images\": Image()}))\r\n# from np.ndarray\r\ndset = Dataset.from_dict({\"image\": [np.array(...)]}, features=Features({\"images\": Image()}))\r\n# cast column\r\ndset = Dataset.from_dict({\"image\": [local_image_path]})\r\ndset.cast_column(\"image\", Image())\r\n\r\n# automatic type inference\r\ndset = Dataset.from_dict({\"image\": [PIL.Image.open(local_image_path)]})\r\n\r\n# transforms\r\ndef img_transform(example):\r\n ...\r\n example[\"image\"] = transformed_pil_image_or_np_ndarray\r\n return example\r\ndset.map(img_trnasform)\r\n\r\n# transform that adds a new column with images (automatic inference of the feature type)\r\ndset.map(lambda ex: {\"image_resized\": ex[\"image\"].resize((100, 100))})\r\nprint(dset.features[\"image_resized\"]) # will print Image()\r\n```\r\n\r\nSome more cool features:\r\n* We store the image filename (`pil_image.filename`) whenever possible to avoid costly conversion to bytes\r\n* if possible, we use native compression when encoding images. Otherwise, we fall back to the lossless PNG format (e.g. after image ops or when storing NumPy arrays)\r\n\r\nHints to make reviewing easier:\r\n* feel free to ignore the extension type part because it's related to PyArrow internals.\r\n* also, let me know if we are too strict/ too flexible in terms of types the Image feature can encode/decode. Hints:\r\n * `encode_example` handles encoding during dataset generation (you can think of it as `yield key, features.encode_example(example)`)\r\n * `objects_to_list_of_image_dicts` handles encoding of returned examples in `map`\r\n\r\nP.S. I'll fork the PR branch and start adding the Image feature to the existing image datasets (will also update the `ImageClassification` template while doing that).", "> WDYT about having an option to skip decoding for complex features, e. g. Audio(decode=False)?\r\n\r\nYes definitely, also I think it could be useful for the dataset viewer to not decode the data but instead return either the bytes or the (possibly chained) URL. cc @severo ", "We want to merge this today/tomorrow, so I'd really appreciate your reviews @sgugger @nateraw.\r\n\r\nAlso, you can test this feature on the existing image datasets (MNIST, beans, food101, ...) by installing `datasets` from the PR branch:\r\n```\r\npip install git+https://github.com/huggingface/datasets.git@adapt-image-datasets\r\n```\r\n", "Thanks for the review @nateraw!\r\n\r\n1. This is a copy of your notebook with the fixed map call: https://colab.research.google.com/gist/mariosasko/e351a717682a9392ca03908e65a2600e/image-feature-demo.ipynb\r\n (Sorry for misleading you with the map call in my un-updated notebook)\r\n Also, we can avoid this cast by trying to infer the type of the column (`\"pixel_values\"`) returned by the image feature extractor (we are already doing something similar for the columns with names: `\"attention_mask\"`, `\"input_ids\"`, ...). I plan to add this QOL improvement soon. \r\n2. It should work OK even without updating Pillow and PyArrow (these two libraries are pre-installed in Colab, so updating them requires a restart of the runtime). \r\n > I noticed an error that I'm guessing you ran into when I tried using the older version\r\n\r\n Do you recall which type of error it was because everything works fine on my side if I run the notebooks with the lowest supported version of Pillow (`6.2.1`)?", "Thanks for playing with it @nateraw and for sharing your notebook, this is useful :)\r\n\r\nI think this is ready now, congrats @mariosasko !", "Love this feature and hope to release soon!" ]

[ { "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" } ]

[ "@lhoestq I tested the script without accelerator, and I confirm this is due to datasets part as this gets similar results without accelerator.", "Hi ! You said it works on 1 GPU but doesn't wortk without accelerator - what's the difference between running on 1 GPU and running without accelerator in your case ?", "Hi @lhoestq \r\nthanks for coming back to me. Sorry for the confusion I made. I meant this works fine on 1 GPU, but on multi-gpu it is freezing. \"accelerator\" is not an issue as if you adapt the code without accelerator this still gets the same issue.\r\nIn order to test it. Please run \"accelerate config\", then use the setup for multi-gpu in one node.\r\nAfter that run \"accelerate launch code.py\" and then you would see the freezing occurs.", "Hi @lhoestq \r\ncould you have the chance to reproduce the error by running the minimal example shared?\r\nthanks", "I think you need to do `train_dataset = train_dataset.with_format(\"torch\")` to work with the DataLoader in a multiprocessing setup :)\r\n\r\nThe hang is probably caused by our streamign lib `fsspec` which doesn't work in multiprocessing out of the box - but we made it work with the PyTorch DataLoader when the dataset format is set to \"torch\"", "Hi @lhoestq \r\nthanks for the response. I added the line suggested right before calling `with accelerator.main_process_first():` in the code above and I confirm this also freezes. to reproduce it please run \"accelerate launch code.py\". I was wondering if you could have more suggestions for me? I do not have an idea how to fix this or debug this freezing. many thanks.", "Maybe the `fsspec` stuff need to be clearer even before - can you try to run this function at the very beginning of your script ?\r\n```python\r\nimport fsspec\r\n\r\ndef _set_fsspec_for_multiprocess() -> None:\r\n \"\"\"\r\n Clear reference to the loop and thread.\r\n This is necessary otherwise HTTPFileSystem hangs in the ML training loop.\r\n Only required for fsspec >= 0.9.0\r\n See https://github.com/fsspec/gcsfs/issues/379\r\n \"\"\"\r\n fsspec.asyn.iothread[0] = None\r\n fsspec.asyn.loop[0] = None\r\n\r\n_set_fsspec_for_multiprocess()\r\n```", "Hi @lhoestq \r\nthank you. I tried it, I am getting `AttributeError: module 'fsspec' has no attribute 'asyn'`. which version of fsspect do you use?\r\nI am using \r\n```fsspec 2022.8.2 pypi_0 pypi```\r\nthank you.", "Hi @lhoestq \r\nI solved `fsspec` error with this hack for now https://discuss.huggingface.co/t/attributeerror-module-fsspec-has-no-attribute-asyn/19255 but this is still freezing, I greatly appreciate if you could run this script on your side. Many thanks.\r\n\r\n```\r\nimport fsspec\r\n\r\ndef _set_fsspec_for_multiprocess() -> None:\r\n \"\"\"\r\n Clear reference to the loop and thread.\r\n This is necessary otherwise HTTPFileSystem hangs in the ML training loop.\r\n Only required for fsspec >= 0.9.0\r\n See https://github.com/fsspec/gcsfs/issues/379\r\n \"\"\"\r\n fsspec.asyn.iothread[0] = None\r\n fsspec.asyn.loop[0] = None\r\n\r\n\r\n_set_fsspec_for_multiprocess()\r\n\r\nfrom accelerate import Accelerator\r\nfrom accelerate.logging import get_logger\r\nfrom datasets import load_dataset\r\nfrom torch.utils.data.dataloader import DataLoader\r\nimport torch\r\nfrom datasets import load_dataset\r\nfrom transformers import AutoTokenizer\r\nimport torch\r\nfrom accelerate.logging import get_logger\r\nfrom torch.utils.data import IterableDataset\r\nfrom torch.utils.data.datapipes.iter.combinatorics import ShufflerIterDataPipe\r\n\r\n\r\nlogger = get_logger(__name__)\r\n\r\n\r\nclass ConstantLengthDataset(IterableDataset):\r\n \"\"\"\r\n Iterable dataset that returns constant length chunks of tokens from stream of text files.\r\n Args:\r\n tokenizer (Tokenizer): The processor used for proccessing the data.\r\n dataset (dataset.Dataset): Dataset with text files.\r\n infinite (bool): If True the iterator is reset after dataset reaches end else stops.\r\n max_seq_length (int): Length of token sequences to return.\r\n num_of_sequences (int): Number of token sequences to keep in buffer.\r\n chars_per_token (int): Number of characters per token used to estimate number of tokens in text buffer.\r\n \"\"\"\r\n\r\n def __init__(\r\n self,\r\n tokenizer,\r\n dataset,\r\n infinite=False,\r\n max_seq_length=1024,\r\n num_of_sequences=1024,\r\n chars_per_token=3.6,\r\n ):\r\n self.tokenizer = tokenizer\r\n # self.concat_token_id = tokenizer.bos_token_id\r\n self.dataset = dataset\r\n self.max_seq_length = max_seq_length\r\n self.epoch = 0\r\n self.infinite = infinite\r\n self.current_size = 0\r\n self.max_buffer_size = max_seq_length * chars_per_token * num_of_sequences\r\n self.content_field = \"text\"\r\n\r\n def __iter__(self):\r\n iterator = iter(self.dataset)\r\n more_examples = True\r\n while more_examples:\r\n buffer, buffer_len = [], 0\r\n while True:\r\n if buffer_len >= self.max_buffer_size:\r\n break\r\n try:\r\n buffer.append(next(iterator)[self.content_field])\r\n buffer_len += len(buffer[-1])\r\n except StopIteration:\r\n if self.infinite:\r\n iterator = iter(self.dataset)\r\n self.epoch += 1\r\n logger.info(f\"Dataset epoch: {self.epoch}\")\r\n else:\r\n more_examples = False\r\n break\r\n tokenized_inputs = self.tokenizer(buffer, truncation=False)[\"input_ids\"]\r\n all_token_ids = []\r\n for tokenized_input in tokenized_inputs:\r\n all_token_ids.extend(tokenized_input)\r\n for i in range(0, len(all_token_ids), self.max_seq_length):\r\n input_ids = all_token_ids[i : i + self.max_seq_length]\r\n if len(input_ids) == self.max_seq_length:\r\n self.current_size += 1\r\n yield torch.tensor(input_ids)\r\n\r\n def shuffle(self, buffer_size=1000):\r\n return ShufflerIterDataPipe(self, buffer_size=buffer_size)\r\n\r\n\r\ndef create_dataloaders(tokenizer, accelerator):\r\n ds_kwargs = {\"streaming\": True}\r\n # In distributed training, the load_dataset function gaurantees that only one process\r\n # can concurrently download the dataset.\r\n datasets = load_dataset(\r\n \"c4\",\r\n \"en\",\r\n cache_dir=\"cache_dir\",\r\n **ds_kwargs,\r\n )\r\n train_data, valid_data = datasets[\"train\"], datasets[\"validation\"]\r\n with accelerator.main_process_first():\r\n train_data = train_data.shuffle(buffer_size=10000, seed=None)\r\n train_dataset = ConstantLengthDataset(\r\n tokenizer,\r\n train_data,\r\n infinite=True,\r\n max_seq_length=256,\r\n )\r\n valid_dataset = ConstantLengthDataset(\r\n tokenizer,\r\n valid_data,\r\n infinite=False,\r\n max_seq_length=256,\r\n )\r\n train_dataset = train_dataset.shuffle(buffer_size=10000)\r\n train_dataloader = DataLoader(train_dataset, batch_size=160, shuffle=True)\r\n eval_dataloader = DataLoader(valid_dataset, batch_size=160)\r\n return train_dataloader, eval_dataloader\r\n\r\n\r\ndef main():\r\n # Accelerator.\r\n logging_dir = \"data_save_dir/log\"\r\n accelerator = Accelerator(\r\n gradient_accumulation_steps=1,\r\n mixed_precision=\"bf16\",\r\n log_with=\"tensorboard\",\r\n logging_dir=logging_dir,\r\n )\r\n # We need to initialize the trackers we use, and also store our configuration.\r\n # The trackers initializes automatically on the main process.\r\n if accelerator.is_main_process:\r\n accelerator.init_trackers(\"test\")\r\n tokenizer = AutoTokenizer.from_pretrained(\"bert-base-uncased\")\r\n\r\n # Load datasets and create dataloaders.\r\n train_dataloader, _ = create_dataloaders(tokenizer, accelerator)\r\n\r\n train_dataloader = accelerator.prepare(train_dataloader)\r\n for step, batch in enumerate(train_dataloader, start=1):\r\n print(step)\r\n accelerator.end_training()\r\n\r\n\r\nif __name__ == \"__main__\":\r\n main()\r\n```", "Are you using `Pytorch 1.11`? Otherwise the script freezes because of the shuffling in this line: \r\n```\r\n return ShufflerIterDataPipe(self, buffer_size=buffer_size)\r\n```\r\n`ShufflerIterDataPipe` behavior must have changed for newer Pytorch versions. But this doesn't change whether you're using streaming or not in `datasets`, so probably not the same issue, but something to try.", "> Are you using `Pytorch 1.11`? Otherwise the script freezes because of the shuffling in this line:\r\n> \r\n> ```\r\n> return ShufflerIterDataPipe(self, buffer_size=buffer_size)\r\n> ```\r\n> \r\n> `ShufflerIterDataPipe` behavior must have changed for newer Pytorch versions. But this doesn't change whether you're using streaming or not in `datasets`, so probably not the same issue, but something to try.\r\n\r\nI met the same issue for pytorch 1.12 and 1.13, is there a way to work around for this function for newer pytorch versions?" ]

[]

[ "@lhoestq I fixed issues except for the dummy_data zip file. But I think I know why is it happening. So when unzipping dummy_data.zip it gets save in /tmp directory where glob doesn't pick it up. For regular downloads, the archive gets unzipped in ~/.cache/huggingface. Could that be a reason?", "Nice thanks :)\r\n\r\nWhen testing with the dummy data, the `download_manager.download_and_extract()` call returns the path to the unzipped dummy_data.zip archive. Therefore glob should be able to find your dummy .epub.txt file", "@lhoestq I understand but for some reason, it is not happening. I added logs to see where dummy_data.zip gets unzipped in /tmp but I suppose when the test process finishes that tmp is gone. I also tried to glob anything in _generate_examples from that directory using /* instead of **/*.epub.txt and nothing is being returned. Always an empty array. ", "Ok weird ! I can take a look tomorrow if you want", "Please do, I will take a fresh look as well. ", "In _generate_examples_ I wrote the following:\r\n```\r\nglob_target = os.path.join(directory, \"**/*.epub.txt\")\r\nprint(f\"Glob target {glob_target }\")\r\n```\r\n\r\nAnd here is the test failure:\r\n\r\n\r\n========================================================================================== FAILURES ===========================================================================================\r\n________________________________________________________________ LocalDatasetTest.test_load_dataset_all_configs_bookcorpusopen ________________________________________________________________\r\n\r\nself = <tests.test_dataset_common.LocalDatasetTest testMethod=test_load_dataset_all_configs_bookcorpusopen>, dataset_name = 'bookcorpusopen'\r\n\r\n @slow\r\n def test_load_dataset_all_configs(self, dataset_name):\r\n configs = self.dataset_tester.load_all_configs(dataset_name, is_local=True)\r\n> self.dataset_tester.check_load_dataset(dataset_name, configs, is_local=True)\r\n\r\ntests/test_dataset_common.py:232: \r\n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _\r\ntests/test_dataset_common.py:193: in check_load_dataset\r\n self.parent.assertTrue(len(dataset[split]) > 0)\r\nE AssertionError: False is not true\r\n------------------------------------------------------------------------------------ Captured stdout call -------------------------------------------------------------------------------------\r\nDownloading and preparing dataset book_corpus_open/plain_text (download: 1.00 MiB, generated: 1.00 MiB, post-processed: Unknown size, total: 2.00 MiB) to /var/folders/y_/6k6zhblx0k9dsdz5nd_z9x5c0000gp/T/tmpmuu0_ln2/book_corpus_open/plain_text/1.0.0...\r\nGlob target /var/folders/y_/6k6zhblx0k9dsdz5nd_z9x5c0000gp/T/tmpm6tpvb3f/extracted/d953b414cceb4fe3985eeaf68aec2f4435f166b2edf66863d805e3825b7d336b/dummy_data/**/*.epub.txt\r\nDataset book_corpus_open downloaded and prepared to /var/folders/y_/6k6zhblx0k9dsdz5nd_z9x5c0000gp/T/tmpmuu0_ln2/book_corpus_open/plain_text/1.0.0. Subsequent calls will reuse this data.\r\n------------------------------------------------------------------------------------ Captured stderr call -------------------------------------------------------------------------------------\r\n \r\n", "And when I do os.listdir on the given directory I get:\r\n\r\n glob_target = os.path.join(directory, \"**/*.epub.txt\")\r\n print(f\"Glob target {glob_target }\")\r\n> print(os.listdir(path=directory))\r\nE FileNotFoundError: [Errno 2] No such file or directory: '/var/folders/y_/6k6zhblx0k9dsdz5nd_z9x5c0000gp/T/tmpbu_aom5q/extracted/d953b414cceb4fe3985eeaf68aec2f4435f166b2edf66863d805e3825b7d336b/dummy_data'\r\n", "Thanks for the info, I'm looking at it right now", "Ok found the issue !\r\n\r\nThe dummy_data.zip file must be an archive of a folder named dummy_data. Currently the dummy_data.zip is an archive of a folder named book1. In order to have a valid dummy_data.zip file you must first take the dummy book1 folder, place it inside a folder named dummy_data and then compress the dummy_data folder to get dummy_data.zip", "Excellent, I am on it @lhoestq ", "> Awesome thank you so much for adding it :)\r\n\r\nYou're welcome, ok all tests are green now! I needed it asap as well. Thanks for your help @lhoestq .", "I just wanted to say thank you to everyone involved in making this happen! I was certain that I would have to add bookcorpusnew myself, but then @vblagoje came along and did it, and @lhoestq gave some great support in a timely fashion.\r\n\r\nBy the way @vblagoje, are you on Twitter? I'm https://twitter.com/theshawwn if you'd like to DM and say hello. Once again, thanks for doing this!\r\n\r\nI'll mention over at https://github.com/soskek/bookcorpus/issues/27 that this was merged.", "Thank you Shawn. You did all the heavy lifting ;-)", "@vblagoje Would you be interested in adding books3 as well? https://twitter.com/theshawwn/status/1320282149329784833\r\n\r\nHuggingface is interested and asked me to add it, but I had a bit of trouble during setup (https://github.com/huggingface/datasets/issues/790) and never got around to it. At this point you have much more experience than I do with the datasets lib.\r\n\r\nIt *seems* like it might simply be a matter of copy-pasting this PR, changing books1 to books3, and possibly trimming off the leading paths -- each book is at e.g. the-eye/Books/Bibliotok/J/Jurassic Park.epub.txt, which is rather lengthy compared to just the filename -- but the full path is probably fine, so feel free to do the least amount of work that gets the job done. Otherwise I suppose I'll get around to it eventually; thanks again!", "@shawwn I'll take a look as soon as I clear my work queue. TBH, I would likely work on making sure HF datasets has all the datasets used to train https://github.com/alexa/bort/ and these are: Wikipedia, Wiktionary, OpenWebText (Gokaslan and Cohen, 2019), UrbanDictionary, Onel Billion Words (Chelba et al., 2014), the news subset of Common Crawl (Nagel, 2016)10, and Bookcorpus. cc @lhoestq ", "@shawwn is your books3 corpus as a part of any dataset now?", "@snarb Books3 has been used in LLaMA (https://twitter.com/theshawwn/status/1643987377516580870) and in BloombergGPT (https://twitter.com/theshawwn/status/1641938293209047041). I don't know whether it's in a HuggingFace dataset yet, but you can access it via the original announcement tweet here: https://twitter.com/theshawwn/status/1320282149329784833\r\n\r\nIf you'd like to make it a huggingface dataset, I'd be grateful! I'm not sure what the process is.\r\n\r\nLLaMA also noted that they deduplicated the books in books3, so it might be worth running some sort of dedup pass on it.", "It's available here already :) https://huggingface.co/datasets/the_pile_books3", "@shawwn how the pictures and tables are handled in such datasets? For example in IQ test or geometry it is hard to imagine to understand topic without images. I want to create a dataset with limited vocabulary. To make possible the training of llmm without big money possible. But still to get model that is able to reason and formulate thoughts well. Trying to use books for children, school educational resources, simplified wiki. Maybe you can sujjest some good data sources with your experience?" ]

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

[ "Thanks, looking at it", "Not sure what happened 😬, but it's fixed" ]

[]

[]

[]

[ "Do you think this is trying to bring the dataset into memory and if I can avoid it to save on memory so it only brings a batch into memory? @lhoestq thank you", "It's not trying to bring the dataset into memory.\r\n\r\nActually, it's trying to memory map the dataset file, which is different. It allows to load large dataset files without filling up memory.\r\n\r\nWhat dataset did you use to get this error ?\r\nOn what OS are you running ? What's your python and pyarrow version ?", "Dear @lhoestq \r\nthank you so much for coming back to me. Please find info below:\r\n1) Dataset name: I used wikipedia with config 20200501.en\r\n2) I got these pyarrow in my environment:\r\npyarrow 2.0.0 <pip>\r\npyarrow 3.0.0 <pip>\r\n\r\n3) python version 3.7.10\r\n4) OS version \r\n\r\nlsb_release -a\r\nNo LSB modules are available.\r\nDistributor ID:\tDebian\r\nDescription:\tDebian GNU/Linux 10 (buster)\r\nRelease:\t10\r\nCodename:\tbuster\r\n\r\n\r\nIs there a way I could solve the memory issue and if I could run this model, I am using GeForce GTX 108, \r\nthanks \r\n", "I noticed that the error happens when loading the validation dataset.\r\nWhat value of `data_args.validation_split_percentage` did you use ?", "Dear @lhoestq \r\n\r\nthank you very much for the very sharp observation, indeed, this happens there, I use the default value of 5, I basically plan to subsample a part of the large dataset and choose it as validation set. Do you think this is bringing the data into memory during subsampling? Is there a way I could avoid this?\r\n\r\nThank you very much for the great help.\r\n\r\n\r\nOn Mon, Mar 8, 2021 at 11:28 AM Quentin Lhoest ***@***.***>\r\nwrote:\r\n\r\n> I noticed that the error happens when loading the validation dataset.\r\n> What value of data_args.validation_split_percentage did you use ?\r\n>\r\n> —\r\n> You are receiving this because you authored the thread.\r\n> Reply to this email directly, view it on GitHub\r\n> <https://github.com/huggingface/datasets/issues/1990#issuecomment-792655644>,\r\n> or unsubscribe\r\n> <https://github.com/notifications/unsubscribe-auth/AS37NMS337ZUJ7HGGVVCCR3TCSREFANCNFSM4YTYAQ2A>\r\n> .\r\n>\r\n", "Methods like `dataset.shard`, `dataset.train_test_split`, `dataset.select` etc. don't bring the dataset in memory. \r\nThe only time when samples are brought to memory is when you access elements via `dataset[0]`, `dataset[:10]`, `dataset[\"my_column_names\"]`.\r\n\r\nBut it's possible that trying to use those methods to build your validation set doesn't fix the issue since, if I understand correctly, the error happens when when the dataset arrow file is opened (just before the 5% percentage is applied).\r\n\r\nDid you try to reproduce this issue in a google colab ? This would be super helpful to investigate why this happened.\r\n\r\nAlso maybe you can try clearing your cache at `~/.cache/huggingface/datasets` and try again. If the arrow file was corrupted somehow, removing it and rebuilding may fix the issue." ]

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

[ "Hi ! This is because we are actually using the arrow streaming format. We plan to switch to the arrow IPC format.\r\nMore info at #1933 ", "Not sure if this was resolved, but I am getting a similar error when trying to load a dataset.arrow file directly: `ArrowInvalid: Not an Arrow file`", "Since we're using the streaming format, you need to use `open_stream`:\r\n\r\n```python\r\nimport pyarrow as pa\r\n\r\ndef in_memory_arrow_table_from_file(filename: str) -> pa.Table:\r\n in_memory_stream = pa.input_stream(filename)\r\n opened_stream = pa.ipc.open_stream(in_memory_stream)\r\n pa_table = opened_stream.read_all()\r\n return pa_table\r\n\r\ndef memory_mapped_arrow_table_from_file(filename: str) -> pa.Table:\r\n memory_mapped_stream = pa.memory_map(filename)\r\n opened_stream = pa.ipc.open_stream(memory_mapped_stream)\r\n pa_table = opened_stream.read_all()\r\n return pa_table\r\n```" ]

[]

[ "This scenario currently requires explicitly passing the target features (to avoid the error): \r\n```python\r\nimport datasets\r\n\r\n...\r\n\r\nfeatures = dataset.features\r\nfeatures[\"output\"] = = [{\"test\": datasets.Value(\"int64\")}]\r\ntest2 = dataset.map(lambda row, idx: test_func(row, idx), with_indices=True, num_proc=32, features=features)\r\n```" ]

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

[ "Hi ! You're completely right, this must be mentioned in the dataset card.\r\nIf you're interesting in contributing, feel free to open a pull request to mention this in the `trivia_qa` dataset card in the \"Licensing Information\" section at https://github.com/huggingface/datasets/blob/master/datasets/trivia_qa/README.md" ]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_3819). All of your documentation changes will be reflected on that endpoint." ]

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

[ "I refreshed the dataset viewer manually, it's fixed now. Sorry for the inconvenience.\r\n<img width=\"1557\" alt=\"Capture d’écran 2022-07-28 à 09 17 39\" src=\"https://user-images.githubusercontent.com/1676121/181514666-92d7f8e1-ddc1-4769-84f3-f1edfdb902e8.png\">\r\n\r\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" } ]

[ "@albertvillanova I would like to take this up. It would be great if you could point me as to how the dataset size is calculated in HF. Thanks!" ]

[]

[ "I think you can just `rm` this directory and it should be good :)", "@lhoestq - this seems to happen quite often (already the 2nd issue). Can we maybe delete this automatically?", "Yes I have an idea of what's going on. I'm sure I can fix that", "Hi, I rebase my local copy to `fix-empty-cache-dir`, and try to run again `python nlp-cli test datasets/bookcorpus --save_infos --all_configs`.\r\n\r\nI got this, \r\n```\r\nTraceback (most recent call last):\r\n File \"nlp-cli\", line 10, in <module>\r\n from nlp.commands.run_beam import RunBeamCommand\r\n File \"/home/yisiang/nlp/src/nlp/commands/run_beam.py\", line 6, in <module>\r\n import apache_beam as beam\r\nModuleNotFoundError: No module named 'apache_beam'\r\n```\r\nAnd after I installed it. I got this\r\n```\r\nFile \"/home/yisiang/nlp/src/nlp/datasets/bookcorpus/aea0bd5142d26df645a8fce23d6110bb95ecb81772bb2a1f29012e329191962c/bookcorpus.py\", line 88, in _split_generators\r\n downloaded_path_or_paths = dl_manager.download_custom(_GDRIVE_FILE_ID, download_file_from_google_drive)\r\n File \"/home/yisiang/nlp/src/nlp/utils/download_manager.py\", line 128, in download_custom\r\n downloaded_path_or_paths = map_nested(url_to_downloaded_path, url_or_urls)\r\n File \"/home/yisiang/nlp/src/nlp/utils/py_utils.py\", line 172, in map_nested\r\n return function(data_struct)\r\n File \"/home/yisiang/nlp/src/nlp/utils/download_manager.py\", line 126, in url_to_downloaded_path\r\n return os.path.join(self._download_config.cache_dir, hash_url_to_filename(url))\r\n File \"/home/yisiang/miniconda3/envs/nlppr/lib/python3.7/posixpath.py\", line 80, in join\r\n a = os.fspath(a)\r\n```\r\nThe problem is when I print `self._download_config.cache_dir` using pdb, it is `None`.\r\n\r\nDid I miss something ? Or can you provide a workaround first so I can keep testing my script ?", "I'll close this issue because I brings more reports in another issue #249 ." ]

[ { "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": "E3165C", "default": false, "description": "", "id": 4190228726, "name": "transfer-to-evaluate", "node_id": "LA_kwDODunzps75wdD2", "url": "https://api.github.com/repos/huggingface/datasets/labels/transfer-to-evaluate" } ]

[ "Thanks @Aktsvigun for your fix.\r\n\r\nHowever, metrics in `datasets` are in deprecation mode:\r\n- #4739\r\n\r\nYou should transfer this PR to the `evaluate` library: https://github.com/huggingface/evaluate\r\n\r\nJust for context, here the link to the PR by @Aktsvigun on tensorflow/nmt:\r\n- https://github.com/tensorflow/nmt/pull/488" ]

[]

[]

[]

[ "Looks good to me" ]

[]

[ "Found it. https://github.com/huggingface/nlp/tree/master/datasets" ]

[ { "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" } ]

[ "If you add the line `dataset2.set_format('np')` before calling `dataset2[0]['tmp']` it should return `np.ndarray`.\r\nI believe internally it will not store it as a list, it is only returning a list when you index it.\r\n\r\n```\r\nIn [1]: import datasets, numpy as np\r\nIn [2]: dataset = datasets.load_dataset(\"glue\", \"mrpc\")[\"validation\"]\r\nIn [3]: dataset2 = dataset.map(lambda x: {\"tmp\": np.array([0.5])}, batched=False)\r\nIn [4]: dataset2[0][\"tmp\"]\r\nOut[4]: [0.5]\r\n\r\nIn [5]: dataset2.set_format('np')\r\n\r\nIn [6]: dataset2[0][\"tmp\"]\r\nOut[6]: array([0.5])\r\n```", "I see, thanks! Any idea if the default numpy → list conversion might cause precision loss?", "I'm not super familiar with our datasets works internally, but I think your `np` array will be stored in a `pyarrow` format, and then you take a view of this as a python array. In which case, I think the precision should be preserved." ]

[]

[]

[]

[ "Awesome!" ]

[]

[ "Any love?", "Hi, any update?", "@albertvillanova Sorry for bothering you again, quick follow up: is there anything else you want me to add / modify?", "Hi @rentruewang , we updated the documentation on `master`, could you merge `master` into your branch please ?", "@lhoestq I've merge `huggingface/datasets/master` into this PR. Please review. Thanks! 🤗\r\n\r\nEdit: Umm... I was experimenting with what renaming a branch would do to a pull request. Please ignore the `closed this PR` down below. 🤗" ]

[]

[]

[]

[ "It looks like you need to rebase from master to fix the CI. Could you do that please ?" ]

[ { "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": "2edb81", "default": false, "description": "A bug in a dataset script provided in the library", "id": 2067388877, "name": "dataset bug", "node_id": "MDU6TGFiZWwyMDY3Mzg4ODc3", "url": "https://api.github.com/repos/huggingface/datasets/labels/dataset%20bug" } ]

[ "Hi! Yes, the docs are outdated. Expect this to be fixed soon. \r\n\r\nIn the meantime, you can try to fix the issue yourself.\r\n\r\nThese are the configs/language pairs supported by `wmt15` from which you can choose:\r\n* `cs-en` (Czech - English)\r\n* `de-en` (German - English)\r\n* `fi-en` (Finnish- English)\r\n* `fr-en` (French - English)\r\n* `ru-en` (Russian - English)\r\n\r\nAnd the current implementation always uses all the subsets available for a language, so to define custom subsets, you'll have to clone the repo from the Hub and replace the line https://huggingface.co/datasets/wmt15/blob/main/wmt_utils.py#L688 with:\r\n`for split, ss_names in (self._subsets if self.config.subsets is None else self.config.subsets).items()`\r\n\r\nThen, you can load the dataset as follows:\r\n```python\r\nfrom datasets import load_dataset\r\ndset = load_dataset(\"path/to/local/wmt15_folder\", \"<one of 5 available configs>\", subsets=...)", "@mariosasko thanks a lot for the suggested fix! ", "Hi @mariosasko \r\n\r\nAre the docs updated? If not, I would like to get on it. I am new around here, would we helpful, if you can guide.\r\n\r\nThanks", "Hi @khushmeeet! The docs haven't been updated, so feel free to work on this issue. This is a tricky issue, so I'll give the steps you can follow to fix this:\r\n\r\nFirst, this code:\r\nhttps://github.com/huggingface/datasets/blob/7cff5b9726a223509dbd6224de3f5f452c8d924f/src/datasets/load.py#L113-L118\r\n\r\nneeds to be replaced with (makes the dataset builder search more robust and allows us to remove the ABC stuff from `wmt_utils.py`):\r\n```python\r\n for name, obj in module.__dict__.items():\r\n if inspect.isclass(obj) and issubclass(obj, main_cls_type):\r\n if inspect.isabstract(obj):\r\n continue\r\n module_main_cls = obj\r\n obj_module = inspect.getmodule(obj)\r\n if obj_module is not None and module == obj_module:\r\n break\r\n```\r\n\r\nThen, all the `wmt_utils.py` scripts need to be updated as follows (these are the diffs with the requiered changes):\r\n````diff\r\n import os\r\n import re\r\n import xml.etree.cElementTree as ElementTree\r\n-from abc import ABC, abstractmethod\r\n\r\n import datasets\r\n````\r\n\r\n````diff\r\nlogger = datasets.logging.get_logger(__name__)\r\n\r\n\r\n _DESCRIPTION = \"\"\"\\\r\n-Translate dataset based on the data from statmt.org.\r\n+Translation dataset based on the data from statmt.org.\r\n\r\n-Versions exists for the different years using a combination of multiple data\r\n-sources. The base `wmt_translate` allows you to create your own config to choose\r\n-your own data/language pair by creating a custom `datasets.translate.wmt.WmtConfig`.\r\n+Versions exist for different years using a combination of data\r\n+sources. The base `wmt` allows you to create a custom dataset by choosing\r\n+your own data/language pair. This can be done as follows:\r\n\r\n ```\r\n-config = datasets.wmt.WmtConfig(\r\n- version=\"0.0.1\",\r\n+from datasets import inspect_dataset, load_dataset_builder\r\n+\r\n+inspect_dataset(\"<insert the dataset name\", \"path/to/scripts\")\r\n+builder = load_dataset_builder(\r\n+ \"path/to/scripts/wmt_utils.py\",\r\n language_pair=(\"fr\", \"de\"),\r\n subsets={\r\n datasets.Split.TRAIN: [\"commoncrawl_frde\"],\r\n datasets.Split.VALIDATION: [\"euelections_dev2019\"],\r\n },\r\n )\r\n-builder = datasets.builder(\"wmt_translate\", config=config)\r\n-```\r\n\r\n+# Standard version\r\n+builder.download_and_prepare()\r\n+ds = builder.as_dataset()\r\n+\r\n+# Streamable version\r\n+ds = builder.as_streaming_dataset()\r\n+```\r\n \"\"\"\r\n````\r\n\r\n````diff\r\n+class Wmt(datasets.GeneratorBasedBuilder):\r\n \"\"\"WMT translation dataset.\"\"\"\r\n+\r\n+ BUILDER_CONFIG_CLASS = WmtConfig\r\n\r\n def __init__(self, *args, **kwargs):\r\n- if type(self) == Wmt and \"config\" not in kwargs: # pylint: disable=unidiomatic-typecheck\r\n- raise ValueError(\r\n- \"The raw `wmt_translate` can only be instantiated with the config \"\r\n- \"kwargs. You may want to use one of the `wmtYY_translate` \"\r\n- \"implementation instead to get the WMT dataset for a specific year.\"\r\n- )\r\n super(Wmt, self).__init__(*args, **kwargs)\r\n\r\n @property\r\n- @abstractmethod\r\n def _subsets(self):\r\n \"\"\"Subsets that make up each split of the dataset.\"\"\"\r\n````\r\n```diff\r\n \"\"\"Subsets that make up each split of the dataset for the language pair.\"\"\"\r\n source, target = self.config.language_pair\r\n filtered_subsets = {}\r\n- for split, ss_names in self._subsets.items():\r\n+ subsets = self._subsets if self.config.subsets is None else self.config.subsets\r\n+ for split, ss_names in subsets.items():\r\n filtered_subsets[split] = []\r\n for ss_name in ss_names:\r\n dataset = DATASET_MAP[ss_name]\r\n```\r\n\r\n`wmt14`, `wmt15`, `wmt16`, `wmt17`, `wmt18`, `wmt19` and `wmt_t2t` have this script, so all of them need to be updated. Also, the dataset summaries from the READMEs of these datasets need to be updated to match the new `_DESCRIPTION` string. And that's it! Let me know if you need additional help.", "Hi @mariosasko ,\r\n\r\nI have made the changes as suggested by you and have opened a PR #4537.\r\n\r\nThanks", "Resolved via #4554 " ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "thanks @albertvillanova done thank you!" ]

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

[ "Ok, so\r\n```\r\nos.path.basename(\"/home/user/\")\r\n```\r\ngives `''` while \r\n```\r\nos.path.basename(\"/home/user\")\r\n```\r\ngives `user`. \r\nThe code should check if the last char is a slash.\r\n", "The fix is:\r\n```\r\n\"name\": os.path.basename(self.path[:-1] if self.path[-1] == \"/\" else self.path)\r\n```", "I came through the same issue , just removing the last slash in the dataset path fixed it for me, may be this repo moderators could accept this as an accepted answer atleast if this could not be integrated\r\n\r\n> The fix is:\r\n> \r\n> ```\r\n> \"name\": os.path.basename(self.path[:-1] if self.path[-1] == \"/\" else self.path)\r\n> ```\r\n\r\n@apohllo consider making a pull request on this \r\n\r\nThanks for the amazing contributions from huggingface people !!\r\n", "@apohllo Would you be interested in submitting a PR with the fix?", "@mariosasko here we go:\r\n\r\nhttps://github.com/huggingface/datasets/pull/4967\r\n\r\nTBH I haven't tested it yet, but should work, since this is a basic change." ]

[]

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

[ { "color": "2edb81", "default": false, "description": "A bug in a dataset script provided in the library", "id": 2067388877, "name": "dataset bug", "node_id": "MDU6TGFiZWwyMDY3Mzg4ODc3", "url": "https://api.github.com/repos/huggingface/datasets/labels/dataset%20bug" } ]

[ "Thanks for reporting, @DanielHesslow. We are fixing it. ", "@albertvillanova thanks for taking care of this so quickly!", "The dataset metadata is fixed. You can download it normally." ]

[ { "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 ! Is there some kind of proxy that is configured in your browser that gives you access to internet ? If it's the case it could explain why it doesn't work in the code, since the proxy wouldn't be used", "It is indeed an agency problem, thank you very, very much", "Let me know if you have other questions :)\r\n\r\nClosing this issue now" ]

[ { "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" } ]

[ "Thanks for the feedback!\r\n\r\nI think since this issue is closely related to loading, I can add a clearer explanation under [Load > local loading script](https://huggingface.co/docs/datasets/main/en/loading#local-loading-script).", "That makes sense but I think having a line about it under https://huggingface.co/docs/datasets/installation#source the \"source\" header here would be useful. My mental model of `pip install -e .` does not include the fact that the source files aren't actually being used. ", "Thanks for sharing your perspective. I think the `load_dataset` function is the only one that pulls from GitHub, and since this use-case is very specific, I don't think we need to include such a broad clarification in the Installation section.\r\n\r\nFeel free to check out the linked PR and let me know if it needs any additional explanation 😊" ]

[]

[]

[]

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

[]

[ "Hi @lhoestq,\r\nCould you please help me, I got this error message in all \"ci/circleci: run_dataset_script_tests_pyarrow*\" tests:\r\n```\r\n...\r\n \"\"\"Wrapper classes for various types of tokenization.\"\"\"\r\n \r\n from bleurt.lib import bert_tokenization\r\n import tensorflow.compat.v1 as tf\r\n> import sentencepiece as spm\r\nE ModuleNotFoundError: No module named 'sentencepiece'\r\n...\r\n```\r\nI am not sure why I do get it. Thanks.\r\n", "Hi ! This issue appeared on master since the last update of `BLEURT`.\r\nI'm working on a fix. You can ignore this issue for this PR", "> Hi ! This issue appeared on master since the last update of `BLEURT`.\r\n> I'm working on a fix. You can ignore this issue for this PR\r\n\r\nThanks for the info", "Merging since the CI is fixed on master" ]

[]

[]

[]

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

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "Thanks for the changes, looks good to me!", "@stevhliu I have pushed some quality improvements both in terms of code and content. Would you be able to re-review? ", "<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) | 0.008325 / 0.011353 (-0.003028) | 0.004432 / 0.011008 (-0.006576) | 0.099794 / 0.038508 (0.061286) | 0.029469 / 0.023109 (0.006360) | 0.306554 / 0.275898 (0.030656) | 0.367373 / 0.323480 (0.043893) | 0.007532 / 0.007986 (-0.000454) | 0.003310 / 0.004328 (-0.001018) | 0.077453 / 0.004250 (0.073203) | 0.034836 / 0.037052 (-0.002216) | 0.311696 / 0.258489 (0.053207) | 0.349683 / 0.293841 (0.055842) | 0.033089 / 0.128546 (-0.095457) | 0.011339 / 0.075646 (-0.064307) | 0.321699 / 0.419271 (-0.097573) | 0.040213 / 0.043533 (-0.003320) | 0.304741 / 0.255139 (0.049602) | 0.331569 / 0.283200 (0.048369) | 0.090397 / 0.141683 (-0.051285) | 1.526001 / 1.452155 (0.073847) | 1.558863 / 1.492716 (0.066146) |\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) | 0.179446 / 0.018006 (0.161440) | 0.416308 / 0.000490 (0.415818) | 0.002390 / 0.000200 (0.002190) | 0.000075 / 0.000054 (0.000021) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.023641 / 0.037411 (-0.013770) | 0.096672 / 0.014526 (0.082147) | 0.104330 / 0.176557 (-0.072227) | 0.146338 / 0.737135 (-0.590797) | 0.108278 / 0.296338 (-0.188060) |\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) | 0.420194 / 0.215209 (0.204985) | 4.196981 / 2.077655 (2.119326) | 1.861206 / 1.504120 (0.357086) | 1.658748 / 1.541195 (0.117554) | 1.704309 / 1.468490 (0.235819) | 0.691639 / 4.584777 (-3.893138) | 3.346303 / 3.745712 (-0.399409) | 1.932962 / 5.269862 (-3.336900) | 1.299395 / 4.565676 (-3.266281) | 0.081869 / 0.424275 (-0.342406) | 0.012415 / 0.007607 (0.004808) | 0.530805 / 0.226044 (0.304761) | 5.293486 / 2.268929 (3.024558) | 2.328327 / 55.444624 (-53.116297) | 1.964956 / 6.876477 (-4.911521) | 2.002793 / 2.142072 (-0.139280) | 0.813380 / 4.805227 (-3.991847) | 0.150030 / 6.500664 (-6.350634) | 0.065194 / 0.075469 (-0.010275) |\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) | 1.259421 / 1.841788 (-0.582367) | 13.667796 / 8.074308 (5.593488) | 13.819121 / 10.191392 (3.627729) | 0.136718 / 0.680424 (-0.543706) | 0.028510 / 0.534201 (-0.505691) | 0.402246 / 0.579283 (-0.177037) | 0.405279 / 0.434364 (-0.029085) | 0.467185 / 0.540337 (-0.073153) | 0.554213 / 1.386936 (-0.832723) |\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) | 0.006738 / 0.011353 (-0.004615) | 0.004616 / 0.011008 (-0.006393) | 0.096978 / 0.038508 (0.058470) | 0.027750 / 0.023109 (0.004640) | 0.411505 / 0.275898 (0.135607) | 0.441796 / 0.323480 (0.118316) | 0.005073 / 0.007986 (-0.002913) | 0.003360 / 0.004328 (-0.000968) | 0.074445 / 0.004250 (0.070194) | 0.040654 / 0.037052 (0.003602) | 0.414277 / 0.258489 (0.155788) | 0.448665 / 0.293841 (0.154824) | 0.032346 / 0.128546 (-0.096200) | 0.011533 / 0.075646 (-0.064114) | 0.317349 / 0.419271 (-0.101923) | 0.041934 / 0.043533 (-0.001599) | 0.409102 / 0.255139 (0.153963) | 0.429977 / 0.283200 (0.146777) | 0.089459 / 0.141683 (-0.052224) | 1.518127 / 1.452155 (0.065973) | 1.569902 / 1.492716 (0.077186) |\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) | 0.232648 / 0.018006 (0.214642) | 0.413751 / 0.000490 (0.413261) | 0.000404 / 0.000200 (0.000204) | 0.000057 / 0.000054 (0.000003) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.025468 / 0.037411 (-0.011943) | 0.098195 / 0.014526 (0.083669) | 0.108882 / 0.176557 (-0.067674) | 0.150059 / 0.737135 (-0.587076) | 0.110742 / 0.296338 (-0.185597) |\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) | 0.445326 / 0.215209 (0.230117) | 4.449200 / 2.077655 (2.371545) | 2.098939 / 1.504120 (0.594819) | 1.861207 / 1.541195 (0.320012) | 1.901385 / 1.468490 (0.432894) | 0.695287 / 4.584777 (-3.889490) | 3.461775 / 3.745712 (-0.283938) | 2.998566 / 5.269862 (-2.271296) | 1.555036 / 4.565676 (-3.010641) | 0.082789 / 0.424275 (-0.341486) | 0.012772 / 0.007607 (0.005165) | 0.564855 / 0.226044 (0.338811) | 5.631049 / 2.268929 (3.362120) | 2.543771 / 55.444624 (-52.900854) | 2.194378 / 6.876477 (-4.682099) | 2.267168 / 2.142072 (0.125095) | 0.803330 / 4.805227 (-4.001898) | 0.151336 / 6.500664 (-6.349328) | 0.067015 / 0.075469 (-0.008454) |\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) | 1.298422 / 1.841788 (-0.543366) | 13.933637 / 8.074308 (5.859329) | 13.570848 / 10.191392 (3.379456) | 0.150787 / 0.680424 (-0.529637) | 0.016911 / 0.534201 (-0.517290) | 0.384771 / 0.579283 (-0.194512) | 0.397505 / 0.434364 (-0.036858) | 0.450931 / 0.540337 (-0.089406) | 0.534501 / 1.386936 (-0.852435) |\n\n</details>\n</details>\n\n![](https://cml.dev/watermark.png \"CML watermark\")\n", "@lhoestq @nateraw made some changes as per the comments. PTAL and approve as necessary. ", "<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) | 0.009037 / 0.011353 (-0.002316) | 0.004970 / 0.011008 (-0.006038) | 0.099223 / 0.038508 (0.060715) | 0.034935 / 0.023109 (0.011826) | 0.297027 / 0.275898 (0.021129) | 0.352861 / 0.323480 (0.029382) | 0.007558 / 0.007986 (-0.000427) | 0.003903 / 0.004328 (-0.000425) | 0.075663 / 0.004250 (0.071413) | 0.042577 / 0.037052 (0.005524) | 0.307182 / 0.258489 (0.048693) | 0.344237 / 0.293841 (0.050396) | 0.041438 / 0.128546 (-0.087108) | 0.012159 / 0.075646 (-0.063487) | 0.333771 / 0.419271 (-0.085501) | 0.047847 / 0.043533 (0.004314) | 0.290797 / 0.255139 (0.035658) | 0.320517 / 0.283200 (0.037318) | 0.098334 / 0.141683 (-0.043349) | 1.446187 / 1.452155 (-0.005968) | 1.495506 / 1.492716 (0.002789) |\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) | 0.203704 / 0.018006 (0.185698) | 0.441325 / 0.000490 (0.440835) | 0.001173 / 0.000200 (0.000973) | 0.000080 / 0.000054 (0.000026) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.026694 / 0.037411 (-0.010718) | 0.103819 / 0.014526 (0.089294) | 0.116377 / 0.176557 (-0.060179) | 0.158280 / 0.737135 (-0.578856) | 0.119797 / 0.296338 (-0.176541) |\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) | 0.405723 / 0.215209 (0.190514) | 4.047633 / 2.077655 (1.969979) | 1.805652 / 1.504120 (0.301532) | 1.611382 / 1.541195 (0.070187) | 1.663117 / 1.468490 (0.194627) | 0.692589 / 4.584777 (-3.892188) | 3.689970 / 3.745712 (-0.055742) | 2.089760 / 5.269862 (-3.180101) | 1.450576 / 4.565676 (-3.115101) | 0.085276 / 0.424275 (-0.338999) | 0.012042 / 0.007607 (0.004434) | 0.513159 / 0.226044 (0.287115) | 5.123235 / 2.268929 (2.854306) | 2.281864 / 55.444624 (-53.162761) | 1.926170 / 6.876477 (-4.950307) | 2.035093 / 2.142072 (-0.106979) | 0.857457 / 4.805227 (-3.947770) | 0.166088 / 6.500664 (-6.334576) | 0.062115 / 0.075469 (-0.013354) |\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) | 1.197776 / 1.841788 (-0.644012) | 14.674452 / 8.074308 (6.600144) | 14.275990 / 10.191392 (4.084598) | 0.170848 / 0.680424 (-0.509576) | 0.028613 / 0.534201 (-0.505588) | 0.438650 / 0.579283 (-0.140633) | 0.439323 / 0.434364 (0.004959) | 0.515090 / 0.540337 (-0.025247) | 0.614216 / 1.386936 (-0.772720) |\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) | 0.007159 / 0.011353 (-0.004194) | 0.005142 / 0.011008 (-0.005866) | 0.096953 / 0.038508 (0.058445) | 0.033036 / 0.023109 (0.009927) | 0.391790 / 0.275898 (0.115892) | 0.427120 / 0.323480 (0.103640) | 0.005691 / 0.007986 (-0.002294) | 0.004848 / 0.004328 (0.000519) | 0.072258 / 0.004250 (0.068008) | 0.049017 / 0.037052 (0.011965) | 0.387267 / 0.258489 (0.128778) | 0.437112 / 0.293841 (0.143272) | 0.036360 / 0.128546 (-0.092186) | 0.012249 / 0.075646 (-0.063397) | 0.336246 / 0.419271 (-0.083025) | 0.048777 / 0.043533 (0.005244) | 0.397872 / 0.255139 (0.142733) | 0.399768 / 0.283200 (0.116568) | 0.101283 / 0.141683 (-0.040400) | 1.443999 / 1.452155 (-0.008156) | 1.575496 / 1.492716 (0.082779) |\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) | 0.220952 / 0.018006 (0.202946) | 0.442220 / 0.000490 (0.441730) | 0.000406 / 0.000200 (0.000206) | 0.000058 / 0.000054 (0.000004) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.028626 / 0.037411 (-0.008786) | 0.109929 / 0.014526 (0.095403) | 0.120989 / 0.176557 (-0.055568) | 0.157377 / 0.737135 (-0.579758) | 0.125522 / 0.296338 (-0.170816) |\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) | 0.436565 / 0.215209 (0.221356) | 4.380771 / 2.077655 (2.303117) | 2.200003 / 1.504120 (0.695883) | 2.013289 / 1.541195 (0.472094) | 2.052658 / 1.468490 (0.584168) | 0.703706 / 4.584777 (-3.881071) | 3.823289 / 3.745712 (0.077577) | 2.064882 / 5.269862 (-3.204980) | 1.330834 / 4.565676 (-3.234842) | 0.085945 / 0.424275 (-0.338330) | 0.012511 / 0.007607 (0.004904) | 0.544171 / 0.226044 (0.318127) | 5.476059 / 2.268929 (3.207130) | 2.695586 / 55.444624 (-52.749039) | 2.330239 / 6.876477 (-4.546238) | 2.429290 / 2.142072 (0.287218) | 0.843154 / 4.805227 (-3.962073) | 0.169334 / 6.500664 (-6.331330) | 0.064261 / 0.075469 (-0.011209) |\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) | 1.268344 / 1.841788 (-0.573444) | 14.934342 / 8.074308 (6.860034) | 13.555389 / 10.191392 (3.363997) | 0.142725 / 0.680424 (-0.537699) | 0.017891 / 0.534201 (-0.516310) | 0.424833 / 0.579283 (-0.154450) | 0.420035 / 0.434364 (-0.014329) | 0.491009 / 0.540337 (-0.049329) | 0.586953 / 1.386936 (-0.799983) |\n\n</details>\n</details>\n\n![](https://cml.dev/watermark.png \"CML watermark\")\n", "Merging this PR with approvals from @stevhliu @lhoestq. ", "<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) | 0.008586 / 0.011353 (-0.002767) | 0.004659 / 0.011008 (-0.006350) | 0.100343 / 0.038508 (0.061835) | 0.029861 / 0.023109 (0.006751) | 0.301090 / 0.275898 (0.025192) | 0.369528 / 0.323480 (0.046048) | 0.006920 / 0.007986 (-0.001065) | 0.003513 / 0.004328 (-0.000815) | 0.078514 / 0.004250 (0.074263) | 0.035285 / 0.037052 (-0.001767) | 0.311257 / 0.258489 (0.052768) | 0.353995 / 0.293841 (0.060154) | 0.033733 / 0.128546 (-0.094813) | 0.011489 / 0.075646 (-0.064157) | 0.323095 / 0.419271 (-0.096176) | 0.040808 / 0.043533 (-0.002725) | 0.301779 / 0.255139 (0.046640) | 0.348517 / 0.283200 (0.065318) | 0.086962 / 0.141683 (-0.054721) | 1.496270 / 1.452155 (0.044115) | 1.514260 / 1.492716 (0.021544) |\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) | 0.189502 / 0.018006 (0.171496) | 0.419326 / 0.000490 (0.418837) | 0.002160 / 0.000200 (0.001960) | 0.000084 / 0.000054 (0.000029) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.023669 / 0.037411 (-0.013742) | 0.096574 / 0.014526 (0.082048) | 0.105970 / 0.176557 (-0.070587) | 0.148531 / 0.737135 (-0.588605) | 0.109948 / 0.296338 (-0.186391) |\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) | 0.424968 / 0.215209 (0.209759) | 4.246292 / 2.077655 (2.168637) | 1.911062 / 1.504120 (0.406943) | 1.700733 / 1.541195 (0.159538) | 1.760756 / 1.468490 (0.292266) | 0.696966 / 4.584777 (-3.887811) | 3.372320 / 3.745712 (-0.373392) | 2.886281 / 5.269862 (-2.383581) | 1.553082 / 4.565676 (-3.012594) | 0.082835 / 0.424275 (-0.341440) | 0.012688 / 0.007607 (0.005081) | 0.536352 / 0.226044 (0.310308) | 5.382510 / 2.268929 (3.113582) | 2.365664 / 55.444624 (-53.078960) | 1.995631 / 6.876477 (-4.880845) | 2.073865 / 2.142072 (-0.068207) | 0.819109 / 4.805227 (-3.986118) | 0.150278 / 6.500664 (-6.350386) | 0.065201 / 0.075469 (-0.010268) |\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) | 1.239835 / 1.841788 (-0.601953) | 13.911847 / 8.074308 (5.837539) | 13.500433 / 10.191392 (3.309041) | 0.137153 / 0.680424 (-0.543271) | 0.028451 / 0.534201 (-0.505750) | 0.394659 / 0.579283 (-0.184625) | 0.404915 / 0.434364 (-0.029449) | 0.458944 / 0.540337 (-0.081394) | 0.542288 / 1.386936 (-0.844648) |\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) | 0.006791 / 0.011353 (-0.004562) | 0.004590 / 0.011008 (-0.006419) | 0.098697 / 0.038508 (0.060189) | 0.027634 / 0.023109 (0.004525) | 0.344383 / 0.275898 (0.068485) | 0.385607 / 0.323480 (0.062127) | 0.005413 / 0.007986 (-0.002573) | 0.003447 / 0.004328 (-0.000881) | 0.077268 / 0.004250 (0.073018) | 0.041823 / 0.037052 (0.004770) | 0.342904 / 0.258489 (0.084414) | 0.399371 / 0.293841 (0.105530) | 0.032668 / 0.128546 (-0.095879) | 0.011598 / 0.075646 (-0.064048) | 0.319973 / 0.419271 (-0.099299) | 0.041760 / 0.043533 (-0.001773) | 0.340510 / 0.255139 (0.085371) | 0.377929 / 0.283200 (0.094730) | 0.090889 / 0.141683 (-0.050793) | 1.496068 / 1.452155 (0.043913) | 1.574884 / 1.492716 (0.082168) |\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) | 0.230489 / 0.018006 (0.212483) | 0.425234 / 0.000490 (0.424745) | 0.000406 / 0.000200 (0.000206) | 0.000059 / 0.000054 (0.000004) |\n\n### Benchmark: benchmark_indices_mapping.json\n\n| metric | select | shard | shuffle | sort | train_test_split |\n|--------|---|---|---|---|---|\n| new / old (diff) | 0.024650 / 0.037411 (-0.012761) | 0.102706 / 0.014526 (0.088180) | 0.108017 / 0.176557 (-0.068539) | 0.143645 / 0.737135 (-0.593490) | 0.110556 / 0.296338 (-0.185782) |\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) | 0.468038 / 0.215209 (0.252829) | 4.670514 / 2.077655 (2.592860) | 2.446620 / 1.504120 (0.942500) | 2.241255 / 1.541195 (0.700060) | 2.286409 / 1.468490 (0.817919) | 0.698923 / 4.584777 (-3.885854) | 3.401121 / 3.745712 (-0.344592) | 1.892399 / 5.269862 (-3.377462) | 1.163101 / 4.565676 (-3.402575) | 0.082567 / 0.424275 (-0.341708) | 0.012662 / 0.007607 (0.005055) | 0.571262 / 0.226044 (0.345218) | 5.731740 / 2.268929 (3.462812) | 2.879649 / 55.444624 (-52.564975) | 2.533846 / 6.876477 (-4.342631) | 2.654789 / 2.142072 (0.512717) | 0.811345 / 4.805227 (-3.993882) | 0.152495 / 6.500664 (-6.348169) | 0.067748 / 0.075469 (-0.007721) |\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) | 1.267852 / 1.841788 (-0.573935) | 14.114920 / 8.074308 (6.040612) | 14.355403 / 10.191392 (4.164011) | 0.150393 / 0.680424 (-0.530031) | 0.016855 / 0.534201 (-0.517346) | 0.378710 / 0.579283 (-0.200573) | 0.385380 / 0.434364 (-0.048984) | 0.439054 / 0.540337 (-0.101284) | 0.524343 / 1.386936 (-0.862593) |\n\n</details>\n</details>\n\n![](https://cml.dev/watermark.png \"CML watermark\")\n" ]

[]

[ "What do you mean 'custom' ? You may want to elaborate on it when ask a question.\r\n\r\nAnyway, there are two things you may interested\r\n`nlp.Dataset.from_file` and `load_dataset(..., cache_dir=)`", "To load a dataset you need to have a script that defines the format of the examples, the splits and the way to generate examples. As your dataset has the same format of squad, you can just copy the squad script (see the [datasets](https://github.com/huggingface/nlp/tree/master/datasets) forlder) and just replace the url to load the data to your local or remote path.\r\n\r\nThen what you can do is `load_dataset(<path/to/your/script>)`", "Also if you want to upload your script, you should be able to use the `nlp-cli`.\r\n\r\nUnfortunately the upload feature was not shipped in the latest version 0.2.0. so right now you can either clone the repo to use it or wait for the next release. We will add some docs to explain how to upload datasets.\r\n", "Since the latest release 0.2.1 you can use \r\n```bash\r\nnlp-cli upload_dataset <path/to/dataset>\r\n```\r\nwhere `<path/to/dataset>` is a path to a folder containing your script (ex: `squad.py`).\r\nThis will upload the script under your namespace on our S3.\r\n\r\nOptionally the folder can also contain `dataset_infos.json` generated using\r\n```bash\r\nnlp-cli test <path/to/dataset> --all_configs --save_infos\r\n```\r\n\r\nThen you should be able to do\r\n```python\r\nnlp.load_dataset(\"my_namespace/dataset_name\")\r\n```" ]

[]

[ "Hi! You can do `ds.get_index(\"embeddings\").faiss_index.metric_type` to get the metric type and then match the result with the FAISS metric [enum](https://github.com/facebookresearch/faiss/blob/43d86e30736ede853c384b24667fc3ab897d6ba9/faiss/MetricType.h#L22-L36) (should be L2).", "Ah! Thank you for pointing this out. FYI: the enum indicates it's using the inner product. Using `torch.inner` or `torch.dot` still produces a discrepancy compared to the built-in score. I think this is because of the compression/quantization that occurs with the FAISS index." ]

[]

[ "Thanks @TevenLeScao for finding this issue", "Thanks @lhoestq for catching this ❤️" ]

[]

[]

[ { "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" } ]

[ "Sorry, my bad... When implementing `xpathopen`, I just considered the use case in the COUNTER dataset... I'm fixing it!" ]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_4095). All of your documentation changes will be reflected on that endpoint." ]

[]

[ "_The documentation is not available anymore as the PR was closed or merged._", "I like the `kwargs` approach, thanks!" ]

[]

[]

[]

[ "I think it's ready now @stas00, did you want to add something else ?\r\nThis PR includes your changes but with the level set to warning", "LGTM, thank you, @lhoestq " ]

[]

[ "Never mind, I ran it again and it worked this time. Strange." ]

[]

[ "As you can see in the error the test checks if `indices_mappings_in_memory` is True or not, which is different from the test you do in your script. In a dataset, both the data and the indices mapping can be either on disk or in memory.\r\n\r\nThe indices mapping correspond to a mapping on top of the data table that is used to re-order/select a sample of the original data table. For example if you do `dataset.train_test_split`, then the resulting train and test datasets will have both an indices mapping to tell which examples are in train and which ones in test.\r\n\r\nBefore saving your datasets on disk, you should call `dataset.flatten_indices()` to remove the indices mapping. It should fix your issue. Under the hood it will create a new data table using the indices mapping. The new data table is going to be a subset of the old one (for example taking only the test set examples), and since the indices mapping will be gone you'll be able to concatenate your datasets.\r\n", "> As you can see in the error the test checks if `indices_mappings_in_memory` is True or not, which is different from the test you do in your script. In a dataset, both the data and the indices mapping can be either on disk or in memory.\r\n> \r\n> The indices mapping correspond to a mapping on top of the data table that is used to re-order/select a sample of the original data table. For example if you do `dataset.train_test_split`, then the resulting train and test datasets will have both an indices mapping to tell which examples are in train and which ones in test.\r\n> \r\n> Before saving your datasets on disk, you should call `dataset.flatten_indices()` to remove the indices mapping. It should fix your issue. Under the hood it will create a new data table using the indices mapping. The new data table is going to be a subset of the old one (for example taking only the test set examples), and since the indices mapping will be gone you'll be able to concatenate your datasets.\r\n\r\n`dataset.flatten_indices()` solved my problem, thanks so much!", "@lhoestq we can add a mention of `dataset.flatten_indices()` in the error message (no rush, just put it on your TODO list or I can do it when I come at it)", "Yup I agree ! And in the docs as well" ]

[]

[ "duplicate" ]

[]

[]

[]

[]

[]