Add batch method to Dataset class

[lappemic]

This PR introduces a new batch method to the Dataset class, aligning its functionality with the IterableDataset.batch() method (implemented in #7054). The implementation uses as well the existing map method for efficient batching of examples.

Key changes:

• Add batch method to Dataset class in arrow_dataset.py
• Utilize map method for batching

Closes #7063

Once the approach is approved, i will create the tests and update the documentation.

[lhoestq]

Looks good to me ! :)

you might want to add the map num_proc argument as well, for people who want to make it run faster

[lappemic]

Thanks for the feedback @lhoestq! The last commits include:

• Adding the num_proc parameter to batch
• Adding tests similar to the one done for IterableDataset.batch()
• Updated the documentation -> I think they are actually misplaced in the Stream page. But could not find a better place atm. Where would you put this documentation?

WDYT?

[lhoestq]

You can put the documentation in process.mdx :)

[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.

[lappemic]

I reset the head to the commit before I added the Dataset.batch() documentation to stream.mdx and instead added the documentation to process.mdx.

[lhoestq]

LGTM thanks ! the CI failures are unrelated to your PR

[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)	0.005736 / 0.011353 (-0.005617)	0.003959 / 0.011008 (-0.007049)	0.063259 / 0.038508 (0.024751)	0.030705 / 0.023109 (0.007596)	0.245706 / 0.275898 (-0.030192)	0.278766 / 0.323480 (-0.044714)	0.003354 / 0.007986 (-0.004632)	0.004246 / 0.004328 (-0.000082)	0.049346 / 0.004250 (0.045095)	0.046439 / 0.037052 (0.009386)	0.257930 / 0.258489 (-0.000559)	0.295562 / 0.293841 (0.001722)	0.030529 / 0.128546 (-0.098017)	0.012465 / 0.075646 (-0.063182)	0.205595 / 0.419271 (-0.213677)	0.036319 / 0.043533 (-0.007214)	0.243872 / 0.255139 (-0.011267)	0.275834 / 0.283200 (-0.007366)	0.020330 / 0.141683 (-0.121353)	1.108337 / 1.452155 (-0.343817)	1.150406 / 1.492716 (-0.342310)

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)	0.113498 / 0.018006 (0.095491)	0.306654 / 0.000490 (0.306164)	0.000238 / 0.000200 (0.000038)	0.000043 / 0.000054 (-0.000012)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.019092 / 0.037411 (-0.018319)	0.063180 / 0.014526 (0.048654)	0.078244 / 0.176557 (-0.098313)	0.126106 / 0.737135 (-0.611030)	0.078651 / 0.296338 (-0.217687)

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)	0.284132 / 0.215209 (0.068923)	2.781250 / 2.077655 (0.703595)	1.471864 / 1.504120 (-0.032256)	1.354661 / 1.541195 (-0.186534)	1.362839 / 1.468490 (-0.105651)	0.719126 / 4.584777 (-3.865651)	2.396969 / 3.745712 (-1.348743)	2.987924 / 5.269862 (-2.281938)	1.910555 / 4.565676 (-2.655121)	0.078612 / 0.424275 (-0.345663)	0.005170 / 0.007607 (-0.002437)	0.333876 / 0.226044 (0.107832)	3.298340 / 2.268929 (1.029412)	1.853332 / 55.444624 (-53.591292)	1.551919 / 6.876477 (-5.324557)	1.585677 / 2.142072 (-0.556395)	0.802487 / 4.805227 (-4.002741)	0.134828 / 6.500664 (-6.365837)	0.041966 / 0.075469 (-0.033503)

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)	0.992277 / 1.841788 (-0.849511)	11.626887 / 8.074308 (3.552578)	9.715623 / 10.191392 (-0.475769)	0.140306 / 0.680424 (-0.540117)	0.014528 / 0.534201 (-0.519673)	0.306247 / 0.579283 (-0.273036)	0.263067 / 0.434364 (-0.171297)	0.342325 / 0.540337 (-0.198013)	0.432299 / 1.386936 (-0.954637)

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)	0.006004 / 0.011353 (-0.005349)	0.003890 / 0.011008 (-0.007118)	0.050408 / 0.038508 (0.011900)	0.031880 / 0.023109 (0.008771)	0.273114 / 0.275898 (-0.002784)	0.296653 / 0.323480 (-0.026826)	0.004569 / 0.007986 (-0.003416)	0.002831 / 0.004328 (-0.001497)	0.050032 / 0.004250 (0.045782)	0.040468 / 0.037052 (0.003415)	0.284718 / 0.258489 (0.026229)	0.321754 / 0.293841 (0.027913)	0.033863 / 0.128546 (-0.094684)	0.012183 / 0.075646 (-0.063463)	0.060805 / 0.419271 (-0.358466)	0.034919 / 0.043533 (-0.008614)	0.274354 / 0.255139 (0.019215)	0.293477 / 0.283200 (0.010277)	0.019418 / 0.141683 (-0.122265)	1.151571 / 1.452155 (-0.300584)	1.217174 / 1.492716 (-0.275542)

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)	0.097326 / 0.018006 (0.079320)	0.316277 / 0.000490 (0.315787)	0.000225 / 0.000200 (0.000025)	0.000045 / 0.000054 (-0.000009)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.022932 / 0.037411 (-0.014479)	0.077455 / 0.014526 (0.062929)	0.088949 / 0.176557 (-0.087608)	0.129447 / 0.737135 (-0.607688)	0.093705 / 0.296338 (-0.202634)

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)	0.303918 / 0.215209 (0.088709)	2.973866 / 2.077655 (0.896211)	1.593165 / 1.504120 (0.089045)	1.465312 / 1.541195 (-0.075883)	1.484503 / 1.468490 (0.016013)	0.731849 / 4.584777 (-3.852928)	0.953337 / 3.745712 (-2.792375)	2.887815 / 5.269862 (-2.382047)	1.923618 / 4.565676 (-2.642058)	0.080073 / 0.424275 (-0.344202)	0.005460 / 0.007607 (-0.002148)	0.359876 / 0.226044 (0.133832)	3.532251 / 2.268929 (1.263323)	1.987778 / 55.444624 (-53.456846)	1.685572 / 6.876477 (-5.190905)	1.827141 / 2.142072 (-0.314932)	0.815953 / 4.805227 (-3.989274)	0.136698 / 6.500664 (-6.363967)	0.042185 / 0.075469 (-0.033285)

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)	1.032508 / 1.841788 (-0.809280)	12.526918 / 8.074308 (4.452610)	10.202942 / 10.191392 (0.011550)	0.145920 / 0.680424 (-0.534504)	0.015643 / 0.534201 (-0.518558)	0.300465 / 0.579283 (-0.278818)	0.126786 / 0.434364 (-0.307578)	0.342885 / 0.540337 (-0.197453)	0.438139 / 1.386936 (-0.948797)