Finish deprecating the fs argument

[dconathan]

See #5385 for some discussion on this

The fs= arg was depcrecated from Dataset.save_to_disk and Dataset.load_from_disk in 2.8.0 (to be removed in 3.0.0). There are a few other places where the fs= arg was still used (functions/methods in datasets.info and datasets.load). This PR adds a similar behavior, warnings and the storage_options= arg to these functions and methods.

One question: should the "deprecated" / "added" versions be 2.8.1 for the docs/warnings on these? Right now I'm going with "fs was deprecated in 2.8.0" but "storage_options= was added in 2.8.1" where appropriate.

@mariosasko

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[albertvillanova]

Thanks for the deprecation. Some minor suggested fixes below...

Also note that the corresponding tests should be updated as well.

[dconathan]

Thanks for the deprecation. Some minor suggested fixes below...

Also note that the corresponding tests should be updated as well.

Thanks for the suggestions/typo fixes. I updated the failing test - passing locally now

[lhoestq]

Nice thanks !

I believe you also need to update _load_info and _save_info in builder.py - they're still passing fs=self._fs instead of storage_options=self._fs.storage_options

This should remove the remaining warnings in the CI such as

tests/test_builder.py::test_builder_with_filesystem_download_and_prepare_reload
tests/test_load.py::test_load_dataset_local[False]
tests/test_load.py::test_load_dataset_local[True]
tests/test_load.py::test_load_dataset_zip_csv[csv_path-False]
tests/test_load.py::test_load_dataset_then_move_then_reload
  /opt/hostedtoolcache/Python/3.7.15/x64/lib/python3.7/site-packages/datasets/info.py:344: FutureWarning: 'fs' was deprecated in favor of 'storage_options' in version 2.9.0 and will be removed in 3.0.0.
  You can remove this warning by passing 'storage_options=fs.storage_options' instead.

[albertvillanova]

Thanks again for all the work, @dconathan.

I agree with @lhoestq that we should better address all remaining fs deprecation warnings with this PR.

For example, there are still some deprecation warnings when calling Dataset.load_from_disk with fs. See:

datasets/src/datasets/load.py

Line 1819 in 232a439

return Dataset.load_from_disk(dataset_path, fs, keep_in_memory=keep_in_memory)

or DatasetDict.load_from_disk with fs. See:

datasets/src/datasets/load.py

Line 1821 in 232a439

return DatasetDict.load_from_disk(dataset_path, fs, keep_in_memory=keep_in_memory)

These docstrings should also be updated:

datasets/src/datasets/filesystems/s3filesystem.py

Line 98 in 232a439

>>> dataset = load_from_disk('s3://my-private-datasets/imdb/train', fs=s3) # doctest: +SKIP

datasets/src/datasets/filesystems/s3filesystem.py

Line 111 in 232a439

>>> dataset.save_to_disk('s3://my-private-datasets/imdb/train', fs=s3) # doctest: +SKIP

[dconathan]

re: docstring, I assume passing in storage_options=s3.storage_options is correct/necessary to pass the secrets?

[dconathan]

what about

datasets/src/datasets/filesystems/__init__.py

Lines 43 to 54 in 5b793dd

	def is_remote_filesystem(fs: fsspec.AbstractFileSystem) -> bool:
	"""
	Validates if filesystem has remote protocol.
	Args:
	fs (`fsspec.spec.AbstractFileSystem`):
	An abstract super-class for pythonic file-systems, e.g. `fsspec.filesystem(\'file\')` or [`datasets.filesystems.S3FileSystem`].
	"""
	if fs is not None and fs.protocol != "file":
	return True
	else:
	return False

leave as is? Is this function no longer necessary?

[albertvillanova]

Thanks again for all your work on this PR, @dconathan.

I think the function is_remote_filesystem should be kept as it is.

We are going to re-run the CI. Once all green, we can merge.

[github-actions]

Show benchmarks

PyArrow==6.0.0

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)	0.008877 / 0.011353 (-0.002475)	0.004725 / 0.011008 (-0.006283)	0.100738 / 0.038508 (0.062230)	0.030251 / 0.023109 (0.007141)	0.301483 / 0.275898 (0.025585)	0.374161 / 0.323480 (0.050681)	0.007225 / 0.007986 (-0.000761)	0.003654 / 0.004328 (-0.000674)	0.078400 / 0.004250 (0.074149)	0.035786 / 0.037052 (-0.001267)	0.309744 / 0.258489 (0.051255)	0.355834 / 0.293841 (0.061994)	0.034344 / 0.128546 (-0.094202)	0.011584 / 0.075646 (-0.064062)	0.321462 / 0.419271 (-0.097810)	0.041201 / 0.043533 (-0.002332)	0.298808 / 0.255139 (0.043669)	0.332626 / 0.283200 (0.049426)	0.089131 / 0.141683 (-0.052552)	1.477888 / 1.452155 (0.025734)	1.530365 / 1.492716 (0.037649)

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)	0.191647 / 0.018006 (0.173640)	0.424339 / 0.000490 (0.423849)	0.002941 / 0.000200 (0.002741)	0.000075 / 0.000054 (0.000020)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.023442 / 0.037411 (-0.013969)	0.097264 / 0.014526 (0.082738)	0.105655 / 0.176557 (-0.070901)	0.145055 / 0.737135 (-0.592081)	0.108750 / 0.296338 (-0.187588)

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)	0.422925 / 0.215209 (0.207716)	4.216022 / 2.077655 (2.138367)	1.876441 / 1.504120 (0.372322)	1.665115 / 1.541195 (0.123920)	1.711105 / 1.468490 (0.242615)	0.701820 / 4.584777 (-3.882957)	3.389319 / 3.745712 (-0.356393)	1.909868 / 5.269862 (-3.359994)	1.270482 / 4.565676 (-3.295195)	0.083680 / 0.424275 (-0.340595)	0.012347 / 0.007607 (0.004740)	0.531076 / 0.226044 (0.305031)	5.344045 / 2.268929 (3.075117)	2.310897 / 55.444624 (-53.133728)	1.971953 / 6.876477 (-4.904524)	2.113748 / 2.142072 (-0.028325)	0.823766 / 4.805227 (-3.981462)	0.150864 / 6.500664 (-6.349800)	0.066263 / 0.075469 (-0.009206)

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)	1.253190 / 1.841788 (-0.588598)	13.757887 / 8.074308 (5.683579)	13.888195 / 10.191392 (3.696803)	0.137285 / 0.680424 (-0.543139)	0.029151 / 0.534201 (-0.505050)	0.387402 / 0.579283 (-0.191881)	0.401673 / 0.434364 (-0.032691)	0.450474 / 0.540337 (-0.089863)	0.533757 / 1.386936 (-0.853179)

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)	0.006919 / 0.011353 (-0.004434)	0.004655 / 0.011008 (-0.006353)	0.096946 / 0.038508 (0.058438)	0.028697 / 0.023109 (0.005588)	0.420020 / 0.275898 (0.144122)	0.460193 / 0.323480 (0.136713)	0.005189 / 0.007986 (-0.002796)	0.003425 / 0.004328 (-0.000904)	0.074900 / 0.004250 (0.070649)	0.041844 / 0.037052 (0.004792)	0.421538 / 0.258489 (0.163049)	0.468497 / 0.293841 (0.174656)	0.032573 / 0.128546 (-0.095973)	0.011731 / 0.075646 (-0.063916)	0.320221 / 0.419271 (-0.099050)	0.042113 / 0.043533 (-0.001420)	0.422757 / 0.255139 (0.167618)	0.445372 / 0.283200 (0.162172)	0.090300 / 0.141683 (-0.051383)	1.458598 / 1.452155 (0.006443)	1.550060 / 1.492716 (0.057344)

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)	0.235489 / 0.018006 (0.217483)	0.418207 / 0.000490 (0.417718)	0.002511 / 0.000200 (0.002311)	0.000080 / 0.000054 (0.000025)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.025603 / 0.037411 (-0.011808)	0.100237 / 0.014526 (0.085711)	0.108617 / 0.176557 (-0.067939)	0.148417 / 0.737135 (-0.588719)	0.110163 / 0.296338 (-0.186176)

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)	0.474804 / 0.215209 (0.259595)	4.745370 / 2.077655 (2.667715)	2.417819 / 1.504120 (0.913699)	2.209892 / 1.541195 (0.668697)	2.263296 / 1.468490 (0.794806)	0.695537 / 4.584777 (-3.889240)	3.381028 / 3.745712 (-0.364684)	2.952271 / 5.269862 (-2.317591)	1.507041 / 4.565676 (-3.058636)	0.083334 / 0.424275 (-0.340941)	0.012554 / 0.007607 (0.004947)	0.578861 / 0.226044 (0.352817)	5.795241 / 2.268929 (3.526313)	2.858544 / 55.444624 (-52.586080)	2.516270 / 6.876477 (-4.360207)	2.557350 / 2.142072 (0.415278)	0.801799 / 4.805227 (-4.003428)	0.151579 / 6.500664 (-6.349085)	0.068765 / 0.075469 (-0.006704)

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)	1.279935 / 1.841788 (-0.561853)	14.049065 / 8.074308 (5.974757)	13.972703 / 10.191392 (3.781311)	0.140551 / 0.680424 (-0.539873)	0.016831 / 0.534201 (-0.517370)	0.383886 / 0.579283 (-0.195397)	0.385661 / 0.434364 (-0.048703)	0.444525 / 0.540337 (-0.095813)	0.532197 / 1.386936 (-0.854739)