Remove tasks

[albertvillanova]

Remove tasks, as part of the 3.0 release.

[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.

[lhoestq]

Maybe we can leave this (or ignore it in a post_init) since otherwise it will break ILSVRC/imagenet-1k and ylecun/mnist and many other datasets.

We could also keep the task classes to be imported and instantiated without errors but still remove all their methods like align_feature since they are unused

[albertvillanova]

So we keep these deprecated classes and parameters for datasets 3.0...

[lhoestq]

Alternatively we could just hardcode a patch in dataset-viewer to keep supporting imagenet and mnist until they are converted to no-code datasets ?

[lhoestq]

Actually breaking this will be in the right direction of making people stop using code datasets. I'm just concerned that the mnist repo will stop working but if ylecun doesn't merge the PR to convert the dataset to parquet we can still hardcode something for the viewer

(and imagenet we can merge by ourselves)

[albertvillanova]

I understand your concerns, but at the same time I would push to not keep the deprecated tasks (since version 2.13.0, more than a year ago) in the new major version.

So I would propose, before making the next release:

• Identify all the datasets using tasks
• Open PRs to convert them to Parquet
• Wait 1/2 weeks for the owners to merge before forcing the merge ourselves for "maintenance" reasons
• Only then, make the release

I can work on that with a script.

[lhoestq]

Sounds good, though no need to convert all the datasets - just the ones that are important

[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)	0.005330 / 0.011353 (-0.006023)	0.003946 / 0.011008 (-0.007062)	0.063530 / 0.038508 (0.025022)	0.030529 / 0.023109 (0.007419)	0.239364 / 0.275898 (-0.036534)	0.261683 / 0.323480 (-0.061797)	0.003197 / 0.007986 (-0.004789)	0.003485 / 0.004328 (-0.000844)	0.049575 / 0.004250 (0.045325)	0.046164 / 0.037052 (0.009112)	0.246129 / 0.258489 (-0.012360)	0.281365 / 0.293841 (-0.012476)	0.029480 / 0.128546 (-0.099066)	0.012450 / 0.075646 (-0.063196)	0.203696 / 0.419271 (-0.215575)	0.036539 / 0.043533 (-0.006994)	0.241664 / 0.255139 (-0.013475)	0.260930 / 0.283200 (-0.022270)	0.019931 / 0.141683 (-0.121752)	1.221075 / 1.452155 (-0.231080)	1.246315 / 1.492716 (-0.246402)

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)	0.095061 / 0.018006 (0.077055)	0.304773 / 0.000490 (0.304283)	0.000208 / 0.000200 (0.000008)	0.000050 / 0.000054 (-0.000004)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.019032 / 0.037411 (-0.018380)	0.062521 / 0.014526 (0.047995)	0.075668 / 0.176557 (-0.100889)	0.121634 / 0.737135 (-0.615501)	0.075456 / 0.296338 (-0.220882)

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)	0.291721 / 0.215209 (0.076512)	2.845445 / 2.077655 (0.767790)	1.450971 / 1.504120 (-0.053149)	1.334586 / 1.541195 (-0.206609)	1.358095 / 1.468490 (-0.110396)	0.729624 / 4.584777 (-3.855153)	2.411504 / 3.745712 (-1.334208)	2.858871 / 5.269862 (-2.410991)	1.893074 / 4.565676 (-2.672603)	0.079068 / 0.424275 (-0.345207)	0.005476 / 0.007607 (-0.002131)	0.329816 / 0.226044 (0.103771)	3.305361 / 2.268929 (1.036432)	1.799924 / 55.444624 (-53.644700)	1.512130 / 6.876477 (-5.364347)	1.635195 / 2.142072 (-0.506877)	0.801486 / 4.805227 (-4.003741)	0.134677 / 6.500664 (-6.365987)	0.042266 / 0.075469 (-0.033203)

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)	0.969835 / 1.841788 (-0.871952)	11.421833 / 8.074308 (3.347524)	9.799120 / 10.191392 (-0.392272)	0.144888 / 0.680424 (-0.535536)	0.014191 / 0.534201 (-0.520010)	0.301037 / 0.579283 (-0.278246)	0.263329 / 0.434364 (-0.171034)	0.403013 / 0.540337 (-0.137324)	0.463805 / 1.386936 (-0.923131)

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)	0.005913 / 0.011353 (-0.005440)	0.003890 / 0.011008 (-0.007118)	0.049995 / 0.038508 (0.011487)	0.032497 / 0.023109 (0.009387)	0.269926 / 0.275898 (-0.005972)	0.295567 / 0.323480 (-0.027913)	0.004365 / 0.007986 (-0.003620)	0.002818 / 0.004328 (-0.001510)	0.049055 / 0.004250 (0.044805)	0.040683 / 0.037052 (0.003630)	0.283043 / 0.258489 (0.024554)	0.321072 / 0.293841 (0.027232)	0.032760 / 0.128546 (-0.095787)	0.012370 / 0.075646 (-0.063277)	0.061574 / 0.419271 (-0.357698)	0.033714 / 0.043533 (-0.009819)	0.276287 / 0.255139 (0.021148)	0.290078 / 0.283200 (0.006878)	0.017250 / 0.141683 (-0.124432)	1.165291 / 1.452155 (-0.286863)	1.213687 / 1.492716 (-0.279029)

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)	0.096122 / 0.018006 (0.078115)	0.311954 / 0.000490 (0.311464)	0.000213 / 0.000200 (0.000013)	0.000052 / 0.000054 (-0.000002)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.022142 / 0.037411 (-0.015270)	0.076470 / 0.014526 (0.061945)	0.088340 / 0.176557 (-0.088216)	0.128594 / 0.737135 (-0.608542)	0.089780 / 0.296338 (-0.206558)

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)	0.298129 / 0.215209 (0.082920)	2.943735 / 2.077655 (0.866080)	1.574351 / 1.504120 (0.070231)	1.446688 / 1.541195 (-0.094506)	1.477714 / 1.468490 (0.009223)	0.722195 / 4.584777 (-3.862582)	0.967675 / 3.745712 (-2.778037)	2.803346 / 5.269862 (-2.466515)	1.895882 / 4.565676 (-2.669794)	0.079193 / 0.424275 (-0.345082)	0.005250 / 0.007607 (-0.002357)	0.350193 / 0.226044 (0.124149)	3.514562 / 2.268929 (1.245634)	1.962743 / 55.444624 (-53.481881)	1.677308 / 6.876477 (-5.199169)	1.811473 / 2.142072 (-0.330600)	0.796234 / 4.805227 (-4.008993)	0.131810 / 6.500664 (-6.368854)	0.041301 / 0.075469 (-0.034168)

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)	1.030700 / 1.841788 (-0.811088)	12.108809 / 8.074308 (4.034501)	10.426112 / 10.191392 (0.234720)	0.139829 / 0.680424 (-0.540595)	0.015133 / 0.534201 (-0.519068)	0.307782 / 0.579283 (-0.271501)	0.130554 / 0.434364 (-0.303810)	0.342728 / 0.540337 (-0.197610)	0.435426 / 1.386936 (-0.951510)