Add fsspec support for to_json, to_csv, and to_parquet

[alvarobartt]

Hi to whoever is reading this! 🤗 (Most likely @mariosasko)

What's in this PR?

This PR replaces the open from Python with fsspec.open and adds the argument storage_options for the methods to_json, to_csv, and to_parquet, to allow users to export any 🤗Dataset into a file in a file-system as requested at #6086.

What's missing in this PR?

As per to_json, to_csv, and to_parquet docstrings for the recently included storage_options arg, I've scoped it to 2.15.0, so we should check that before merging in case we want to scope that for 2.14.2 instead.

Additionally, should we also add fsspec support for the from_csv, from_json, and from_parquet methods? If you want me to do so @mariosasko just let me know and I'll create another PR to support that too!

Fix #6086.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.

[alvarobartt]

Hi here @lhoestq @mariosasko I just realised this PR is still open, should we close it in case this is something not to include within datasets, or should we merge? Let me know whatever you decide 🤗

[lhoestq]

Thanks for the ping ! It looks good to me, I just added a few suggestions before merging:

[lhoestq]

LGTM thanks !

[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)	0.005011 / 0.011353 (-0.006342)	0.003203 / 0.011008 (-0.007806)	0.064033 / 0.038508 (0.025524)	0.029152 / 0.023109 (0.006043)	0.242884 / 0.275898 (-0.033014)	0.263517 / 0.323480 (-0.059963)	0.004088 / 0.007986 (-0.003898)	0.002570 / 0.004328 (-0.001759)	0.049061 / 0.004250 (0.044811)	0.040170 / 0.037052 (0.003117)	0.263305 / 0.258489 (0.004816)	0.286255 / 0.293841 (-0.007586)	0.028206 / 0.128546 (-0.100340)	0.010337 / 0.075646 (-0.065309)	0.206235 / 0.419271 (-0.213036)	0.038182 / 0.043533 (-0.005351)	0.246486 / 0.255139 (-0.008653)	0.263077 / 0.283200 (-0.020122)	0.017850 / 0.141683 (-0.123833)	1.173921 / 1.452155 (-0.278234)	1.255583 / 1.492716 (-0.237133)

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)	0.090278 / 0.018006 (0.072272)	0.298146 / 0.000490 (0.297657)	0.000215 / 0.000200 (0.000015)	0.000043 / 0.000054 (-0.000011)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.018021 / 0.037411 (-0.019390)	0.061434 / 0.014526 (0.046908)	0.072617 / 0.176557 (-0.103939)	0.119063 / 0.737135 (-0.618072)	0.073997 / 0.296338 (-0.222341)

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)	0.288496 / 0.215209 (0.073287)	2.794943 / 2.077655 (0.717288)	1.538299 / 1.504120 (0.034179)	1.399164 / 1.541195 (-0.142031)	1.419104 / 1.468490 (-0.049386)	0.566147 / 4.584777 (-4.018630)	2.386687 / 3.745712 (-1.359025)	2.723584 / 5.269862 (-2.546278)	1.699161 / 4.565676 (-2.866515)	0.062526 / 0.424275 (-0.361750)	0.004927 / 0.007607 (-0.002680)	0.345132 / 0.226044 (0.119087)	3.389634 / 2.268929 (1.120706)	1.898012 / 55.444624 (-53.546612)	1.599050 / 6.876477 (-5.277427)	1.614289 / 2.142072 (-0.527783)	0.656716 / 4.805227 (-4.148511)	0.118480 / 6.500664 (-6.382184)	0.041913 / 0.075469 (-0.033557)

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)	0.968676 / 1.841788 (-0.873111)	11.184668 / 8.074308 (3.110360)	9.249912 / 10.191392 (-0.941480)	0.141139 / 0.680424 (-0.539285)	0.014207 / 0.534201 (-0.519994)	0.287603 / 0.579283 (-0.291680)	0.262792 / 0.434364 (-0.171572)	0.340239 / 0.540337 (-0.200099)	0.437471 / 1.386936 (-0.949465)

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)	0.005268 / 0.011353 (-0.006085)	0.003142 / 0.011008 (-0.007866)	0.049333 / 0.038508 (0.010825)	0.029558 / 0.023109 (0.006449)	0.270716 / 0.275898 (-0.005182)	0.293834 / 0.323480 (-0.029646)	0.004285 / 0.007986 (-0.003701)	0.002703 / 0.004328 (-0.001626)	0.048857 / 0.004250 (0.044607)	0.043456 / 0.037052 (0.006404)	0.286058 / 0.258489 (0.027569)	0.313491 / 0.293841 (0.019650)	0.029336 / 0.128546 (-0.099210)	0.010287 / 0.075646 (-0.065360)	0.057753 / 0.419271 (-0.361518)	0.050867 / 0.043533 (0.007334)	0.271717 / 0.255139 (0.016578)	0.291468 / 0.283200 (0.008268)	0.018668 / 0.141683 (-0.123015)	1.137399 / 1.452155 (-0.314755)	1.186315 / 1.492716 (-0.306401)

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)	0.090289 / 0.018006 (0.072283)	0.297987 / 0.000490 (0.297497)	0.000227 / 0.000200 (0.000027)	0.000044 / 0.000054 (-0.000010)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.021012 / 0.037411 (-0.016399)	0.075046 / 0.014526 (0.060520)	0.085295 / 0.176557 (-0.091261)	0.123879 / 0.737135 (-0.613257)	0.086572 / 0.296338 (-0.209766)

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)	0.293350 / 0.215209 (0.078141)	2.875958 / 2.077655 (0.798304)	1.586460 / 1.504120 (0.082340)	1.467950 / 1.541195 (-0.073245)	1.453478 / 1.468490 (-0.015012)	0.566083 / 4.584777 (-4.018693)	2.462582 / 3.745712 (-1.283130)	2.609367 / 5.269862 (-2.660495)	1.709691 / 4.565676 (-2.855985)	0.062928 / 0.424275 (-0.361347)	0.005040 / 0.007607 (-0.002567)	0.337997 / 0.226044 (0.111952)	3.347235 / 2.268929 (1.078306)	1.923940 / 55.444624 (-53.520684)	1.657731 / 6.876477 (-5.218746)	1.747469 / 2.142072 (-0.394604)	0.657061 / 4.805227 (-4.148167)	0.116655 / 6.500664 (-6.384009)	0.040363 / 0.075469 (-0.035106)

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)	1.011171 / 1.841788 (-0.830617)	11.705905 / 8.074308 (3.631597)	10.064391 / 10.191392 (-0.127001)	0.141681 / 0.680424 (-0.538743)	0.014763 / 0.534201 (-0.519438)	0.286425 / 0.579283 (-0.292858)	0.271036 / 0.434364 (-0.163328)	0.321393 / 0.540337 (-0.218944)	0.424539 / 1.386936 (-0.962397)

[albertvillanova]

Thanks @alvarobartt.

I am linking this PR to the corresponding issue (on the right column, under "Development") and closing the issue.

For future contributions, please add to the PR description the word "fix" followed by the issue number, e.g.:

Fix #6086.

• I have edited the PR description to add this.

[alvarobartt]

Thanks @alvarobartt.

I am linking this PR to the corresponding issue (on the right column, under "Development") and closing the issue.

For future contributions, please add to the PR description the word "fix" followed by the issue number, e.g.:

Fix #6086.

• I have edited the PR description to add this.

Hi @albertvillanova, fair, I missed that, thanks for the edit and the heads up!