Fix NonMatchingSplitsSizesError/ExpectedMoreSplits when passing data_dir/data_files in no-code Hub datasets

[albertvillanova]

Fix NonMatchingSplitsSizesError or ExpectedMoreSplits error for no-code Hub datasets if the user passes:

• data_dir
• data_files

The proposed solution is to avoid using exported dataset info (from Parquet exports) in these cases.
Additionally, also if the user passes revision other than "main" (so that no network requests are made).

This PR fixes a bug introduced by:

• Expand no-code dataset info with datasets-server info #6714

Fix #6918, fix #6939.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[albertvillanova]

Do you think this is worth making a patch release for?
CC: @huggingface/datasets

[albertvillanova]

I will add some regression tests before merging.

And I will make a patch release afterwards.

[github-actions]

Show benchmarks

PyArrow==8.0.0

Show updated benchmarks!

Benchmark: benchmark_array_xd.json

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
new / old (diff)	0.004959 / 0.011353 (-0.006394)	0.003654 / 0.011008 (-0.007354)	0.064087 / 0.038508 (0.025579)	0.031942 / 0.023109 (0.008833)	0.236830 / 0.275898 (-0.039068)	0.265359 / 0.323480 (-0.058121)	0.003108 / 0.007986 (-0.004878)	0.002824 / 0.004328 (-0.001504)	0.049102 / 0.004250 (0.044852)	0.046070 / 0.037052 (0.009017)	0.248830 / 0.258489 (-0.009659)	0.283900 / 0.293841 (-0.009941)	0.027799 / 0.128546 (-0.100747)	0.010572 / 0.075646 (-0.065074)	0.223595 / 0.419271 (-0.195677)	0.036951 / 0.043533 (-0.006582)	0.238813 / 0.255139 (-0.016326)	0.253841 / 0.283200 (-0.029359)	0.018471 / 0.141683 (-0.123212)	1.131969 / 1.452155 (-0.320186)	1.173763 / 1.492716 (-0.318954)

Benchmark: benchmark_getitem_100B.json

metric	get_batch_of_1024_random_rows	get_batch_of_1024_rows	get_first_row	get_last_row
new / old (diff)	0.095504 / 0.018006 (0.077498)	0.301469 / 0.000490 (0.300979)	0.000212 / 0.000200 (0.000012)	0.000052 / 0.000054 (-0.000003)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.019194 / 0.037411 (-0.018217)	0.062313 / 0.014526 (0.047787)	0.075852 / 0.176557 (-0.100704)	0.121996 / 0.737135 (-0.615140)	0.076416 / 0.296338 (-0.219923)

Benchmark: benchmark_iterating.json

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
new / old (diff)	0.292465 / 0.215209 (0.077256)	2.910234 / 2.077655 (0.832579)	1.479672 / 1.504120 (-0.024448)	1.332281 / 1.541195 (-0.208913)	1.354095 / 1.468490 (-0.114395)	0.573438 / 4.584777 (-4.011339)	2.382406 / 3.745712 (-1.363307)	2.708289 / 5.269862 (-2.561572)	1.739665 / 4.565676 (-2.826011)	0.063514 / 0.424275 (-0.360761)	0.005008 / 0.007607 (-0.002599)	0.350070 / 0.226044 (0.124025)	3.475837 / 2.268929 (1.206909)	1.804639 / 55.444624 (-53.639985)	1.520472 / 6.876477 (-5.356005)	1.658061 / 2.142072 (-0.484011)	0.648495 / 4.805227 (-4.156732)	0.118394 / 6.500664 (-6.382270)	0.042557 / 0.075469 (-0.032912)

Benchmark: benchmark_map_filter.json

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
new / old (diff)	0.960772 / 1.841788 (-0.881016)	11.451629 / 8.074308 (3.377321)	9.613331 / 10.191392 (-0.578061)	0.130259 / 0.680424 (-0.550164)	0.015828 / 0.534201 (-0.518373)	0.287581 / 0.579283 (-0.291702)	0.266517 / 0.434364 (-0.167847)	0.327334 / 0.540337 (-0.213003)	0.427881 / 1.386936 (-0.959055)

PyArrow==latest

Show updated benchmarks!

Benchmark: benchmark_array_xd.json

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
new / old (diff)	0.005364 / 0.011353 (-0.005989)	0.003723 / 0.011008 (-0.007285)	0.049990 / 0.038508 (0.011482)	0.032023 / 0.023109 (0.008913)	0.258609 / 0.275898 (-0.017289)	0.281250 / 0.323480 (-0.042230)	0.004222 / 0.007986 (-0.003764)	0.002799 / 0.004328 (-0.001529)	0.049546 / 0.004250 (0.045296)	0.040298 / 0.037052 (0.003246)	0.273552 / 0.258489 (0.015063)	0.304042 / 0.293841 (0.010201)	0.030116 / 0.128546 (-0.098430)	0.010792 / 0.075646 (-0.064855)	0.058427 / 0.419271 (-0.360845)	0.033415 / 0.043533 (-0.010118)	0.258794 / 0.255139 (0.003655)	0.275304 / 0.283200 (-0.007896)	0.017944 / 0.141683 (-0.123739)	1.109291 / 1.452155 (-0.342864)	1.156627 / 1.492716 (-0.336090)

Benchmark: benchmark_getitem_100B.json

metric	get_batch_of_1024_random_rows	get_batch_of_1024_rows	get_first_row	get_last_row
new / old (diff)	0.096700 / 0.018006 (0.078693)	0.301108 / 0.000490 (0.300618)	0.000208 / 0.000200 (0.000008)	0.000054 / 0.000054 (-0.000001)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.022632 / 0.037411 (-0.014779)	0.075813 / 0.014526 (0.061287)	0.090302 / 0.176557 (-0.086254)	0.130375 / 0.737135 (-0.606760)	0.089710 / 0.296338 (-0.206629)

Benchmark: benchmark_iterating.json

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
new / old (diff)	0.297091 / 0.215209 (0.081882)	2.910379 / 2.077655 (0.832725)	1.570460 / 1.504120 (0.066340)	1.441619 / 1.541195 (-0.099576)	1.442417 / 1.468490 (-0.026073)	0.570034 / 4.584777 (-4.014743)	0.952613 / 3.745712 (-2.793099)	2.659274 / 5.269862 (-2.610588)	1.751013 / 4.565676 (-2.814663)	0.064639 / 0.424275 (-0.359636)	0.005145 / 0.007607 (-0.002462)	0.347478 / 0.226044 (0.121434)	3.443862 / 2.268929 (1.174933)	1.897246 / 55.444624 (-53.547379)	1.609267 / 6.876477 (-5.267210)	1.755116 / 2.142072 (-0.386956)	0.658982 / 4.805227 (-4.146245)	0.117000 / 6.500664 (-6.383664)	0.041453 / 0.075469 (-0.034016)

Benchmark: benchmark_map_filter.json

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
new / old (diff)	1.005843 / 1.841788 (-0.835944)	12.101306 / 8.074308 (4.026998)	10.370706 / 10.191392 (0.179314)	0.139374 / 0.680424 (-0.541050)	0.015605 / 0.534201 (-0.518596)	0.286978 / 0.579283 (-0.292305)	0.122951 / 0.434364 (-0.311413)	0.331729 / 0.540337 (-0.208609)	0.422088 / 1.386936 (-0.964848)

[meg-huggingface]

I'm hitting this error now, using Spaces. Here's what an attempt to just get the 'validation' split is doing:

code:

  | import os
  | from huggingface_hub import HfApi
  | from datasets import Dataset, load_dataset, DownloadConfig
  |  
  |  
  | GATED_IMAGENET = os.environ.get("GATED_IMAGENET")
  | api = HfApi(token=GATED_IMAGENET)
  |
  | ds = load_dataset('datacomp/imagenet-1k-random0.0', token=GATED_IMAGENET, data_files={'validation': 'data/val*'}, split='validation', trust_remote_code=True)

log:

Generating validation split:   0%|          | 0/50000 [00:00<?, ? examples/s]
Generating validation split:  12%|█▏        | 6172/50000 [00:01<00:07, 5804.90 examples/s]
Generating validation split:  25%|██▌       | 12716/50000 [00:02<00:06, 6167.77 examples/s]
Generating validation split:  38%|███▊      | 19060/50000 [00:03<00:04, 6218.99 examples/s]
Generating validation split:  51%|█████     | 25603/50000 [00:04<00:03, 6126.35 examples/s]
Generating validation split:  64%|██████▍   | 32145/50000 [00:05<00:02, 6166.95 examples/s]
Generating validation split:  77%|███████▋  | 38716/50000 [00:06<00:01, 6272.66 examples/s]
Generating validation split:  90%|█████████ | 45158/50000 [00:07<00:00, 6307.44 examples/s]
Generating validation split: 100%|██████████| 50000/50000 [00:08<00:00, 6212.19 examples/s]
Traceback (most recent call last):
  File "/home/user/app/app.py", line 12, in <module>
    ds = load_dataset('datacomp/imagenet-1k-random0.0', token=GATED_IMAGENET, data_files={'validation': 'data/val*'}, split='validation', trust_remote_code=True)
  File "/usr/local/lib/python3.10/site-packages/datasets/load.py", line 2154, in load_dataset
    builder_instance.download_and_prepare(
  File "/usr/local/lib/python3.10/site-packages/datasets/builder.py", line 924, in download_and_prepare
    self._download_and_prepare(
  File "/usr/local/lib/python3.10/site-packages/datasets/builder.py", line 1018, in _download_and_prepare
    verify_splits(self.info.splits, split_dict)
  File "/usr/local/lib/python3.10/site-packages/datasets/utils/info_utils.py", line 68, in verify_splits
    raise ExpectedMoreSplitsError(str(set(expected_splits) - set(recorded_splits)))
datasets.exceptions.ExpectedMoreSplitsError: {'train', 'test'}

[lhoestq]

Hi Meg ! Thanks for reporting, I'll see how I can fix this. In the meantime feel free to pass verification_mode="no_checks" to load_dataset