Use yaml instead of get data patterns when possible #6154

lhoestq · 2023-08-17T09:17:05Z

This would make the data files resolution faster: no need to list all the data files to infer the dataset builder to use.

fix #6140

thomwolf · 2023-08-17T09:18:57Z

src/datasets/load.py

+        if self.data_files is not None:
+            patterns = sanitize_patterns(self.data_files)
+        if metadata_configs and "data_files" in next(iter(metadata_configs.values())):
+            patterns = sanitize_patterns(next(iter(metadata_configs.values()))["data_files"])


what if various configs in the metadata_configs use various type of files? (edge case maybe)

This is not supported yet, and would require a refactor.

In a subsequent PR we can raise an error if it happens. Currently it raises an error if it fails to load data files if it uses the wrong dataset builder but this is confusing

HuggingFaceDocBuilderDev · 2023-08-17T09:25:25Z

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

github-actions · 2023-08-17T09:25:38Z

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)	0.006829 / 0.011353 (-0.004524)	0.004535 / 0.011008 (-0.006473)	0.085255 / 0.038508 (0.046747)	0.080861 / 0.023109 (0.057752)	0.366023 / 0.275898 (0.090125)	0.403095 / 0.323480 (0.079615)	0.005615 / 0.007986 (-0.002370)	0.003830 / 0.004328 (-0.000498)	0.064502 / 0.004250 (0.060251)	0.053916 / 0.037052 (0.016863)	0.366010 / 0.258489 (0.107521)	0.414565 / 0.293841 (0.120724)	0.031500 / 0.128546 (-0.097046)	0.009252 / 0.075646 (-0.066394)	0.289584 / 0.419271 (-0.129688)	0.052984 / 0.043533 (0.009451)	0.352626 / 0.255139 (0.097487)	0.390964 / 0.283200 (0.107764)	0.025118 / 0.141683 (-0.116565)	1.462316 / 1.452155 (0.010161)	1.565682 / 1.492716 (0.072966)

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)	0.294432 / 0.018006 (0.276426)	0.618366 / 0.000490 (0.617876)	0.003270 / 0.000200 (0.003071)	0.000081 / 0.000054 (0.000027)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.031194 / 0.037411 (-0.006217)	0.088892 / 0.014526 (0.074366)	0.102580 / 0.176557 (-0.073977)	0.159449 / 0.737135 (-0.577686)	0.104434 / 0.296338 (-0.191905)

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)	0.385690 / 0.215209 (0.170481)	3.832782 / 2.077655 (1.755128)	1.862521 / 1.504120 (0.358401)	1.685674 / 1.541195 (0.144479)	1.724984 / 1.468490 (0.256494)	0.483700 / 4.584777 (-4.101077)	3.664154 / 3.745712 (-0.081558)	3.323023 / 5.269862 (-1.946839)	2.055958 / 4.565676 (-2.509718)	0.056990 / 0.424275 (-0.367285)	0.007674 / 0.007607 (0.000067)	0.460642 / 0.226044 (0.234598)	4.609964 / 2.268929 (2.341036)	2.434868 / 55.444624 (-53.009756)	2.003347 / 6.876477 (-4.873130)	2.209520 / 2.142072 (0.067448)	0.629363 / 4.805227 (-4.175864)	0.135434 / 6.500664 (-6.365230)	0.060498 / 0.075469 (-0.014971)

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)	1.253917 / 1.841788 (-0.587870)	19.988953 / 8.074308 (11.914645)	14.353739 / 10.191392 (4.162347)	0.165987 / 0.680424 (-0.514437)	0.018299 / 0.534201 (-0.515902)	0.395532 / 0.579283 (-0.183751)	0.418708 / 0.434364 (-0.015656)	0.460865 / 0.540337 (-0.079472)	0.633925 / 1.386936 (-0.753011)

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)	0.006631 / 0.011353 (-0.004722)	0.004109 / 0.011008 (-0.006899)	0.065003 / 0.038508 (0.026495)	0.080407 / 0.023109 (0.057297)	0.362966 / 0.275898 (0.087068)	0.389727 / 0.323480 (0.066247)	0.005588 / 0.007986 (-0.002397)	0.003517 / 0.004328 (-0.000812)	0.065821 / 0.004250 (0.061570)	0.057614 / 0.037052 (0.020561)	0.367422 / 0.258489 (0.108932)	0.400706 / 0.293841 (0.106865)	0.031560 / 0.128546 (-0.096986)	0.008659 / 0.075646 (-0.066987)	0.070756 / 0.419271 (-0.348516)	0.049821 / 0.043533 (0.006288)	0.360836 / 0.255139 (0.105697)	0.383981 / 0.283200 (0.100781)	0.023719 / 0.141683 (-0.117963)	1.485197 / 1.452155 (0.033043)	1.544899 / 1.492716 (0.052182)

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)	0.336480 / 0.018006 (0.318474)	0.532839 / 0.000490 (0.532349)	0.003767 / 0.000200 (0.003567)	0.000087 / 0.000054 (0.000032)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.034132 / 0.037411 (-0.003280)	0.090131 / 0.014526 (0.075605)	0.104086 / 0.176557 (-0.072471)	0.158385 / 0.737135 (-0.578751)	0.106417 / 0.296338 (-0.189922)

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)	0.416462 / 0.215209 (0.201253)	4.160409 / 2.077655 (2.082755)	2.195355 / 1.504120 (0.691235)	2.051234 / 1.541195 (0.510040)	2.012116 / 1.468490 (0.543626)	0.477414 / 4.584777 (-4.107363)	3.590326 / 3.745712 (-0.155386)	3.318490 / 5.269862 (-1.951371)	2.064124 / 4.565676 (-2.501553)	0.057040 / 0.424275 (-0.367235)	0.007283 / 0.007607 (-0.000324)	0.480490 / 0.226044 (0.254445)	4.804013 / 2.268929 (2.535084)	2.625940 / 55.444624 (-52.818685)	2.231537 / 6.876477 (-4.644939)	2.441649 / 2.142072 (0.299576)	0.573207 / 4.805227 (-4.232020)	0.131685 / 6.500664 (-6.368979)	0.060112 / 0.075469 (-0.015357)

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)	1.358587 / 1.841788 (-0.483200)	20.457562 / 8.074308 (12.383254)	14.236304 / 10.191392 (4.044912)	0.152860 / 0.680424 (-0.527563)	0.018466 / 0.534201 (-0.515735)	0.401391 / 0.579283 (-0.177893)	0.410252 / 0.434364 (-0.024111)	0.484335 / 0.540337 (-0.056002)	0.663818 / 1.386936 (-0.723118)

github-actions · 2023-08-17T09:29:39Z

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)	0.007725 / 0.011353 (-0.003628)	0.004448 / 0.011008 (-0.006560)	0.098689 / 0.038508 (0.060180)	0.082919 / 0.023109 (0.059809)	0.380707 / 0.275898 (0.104809)	0.452977 / 0.323480 (0.129497)	0.004430 / 0.007986 (-0.003555)	0.003712 / 0.004328 (-0.000616)	0.076675 / 0.004250 (0.072425)	0.062281 / 0.037052 (0.025228)	0.403370 / 0.258489 (0.144881)	0.464557 / 0.293841 (0.170716)	0.035646 / 0.128546 (-0.092900)	0.009776 / 0.075646 (-0.065870)	0.341955 / 0.419271 (-0.077316)	0.059515 / 0.043533 (0.015983)	0.388421 / 0.255139 (0.133282)	0.439496 / 0.283200 (0.156296)	0.029090 / 0.141683 (-0.112593)	1.727473 / 1.452155 (0.275319)	1.810448 / 1.492716 (0.317732)

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)	0.221215 / 0.018006 (0.203208)	0.486660 / 0.000490 (0.486171)	0.005467 / 0.000200 (0.005267)	0.000110 / 0.000054 (0.000056)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.032491 / 0.037411 (-0.004920)	0.094446 / 0.014526 (0.079920)	0.110339 / 0.176557 (-0.066217)	0.175004 / 0.737135 (-0.562131)	0.109209 / 0.296338 (-0.187129)

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)	0.453966 / 0.215209 (0.238757)	4.515842 / 2.077655 (2.438187)	2.240512 / 1.504120 (0.736392)	2.059911 / 1.541195 (0.518717)	2.150635 / 1.468490 (0.682145)	0.564509 / 4.584777 (-4.020268)	4.055208 / 3.745712 (0.309496)	3.614084 / 5.269862 (-1.655778)	2.295760 / 4.565676 (-2.269917)	0.066507 / 0.424275 (-0.357768)	0.008909 / 0.007607 (0.001302)	0.542604 / 0.226044 (0.316560)	5.412162 / 2.268929 (3.143233)	2.758757 / 55.444624 (-52.685867)	2.430693 / 6.876477 (-4.445784)	2.669866 / 2.142072 (0.527793)	0.681756 / 4.805227 (-4.123471)	0.156524 / 6.500664 (-6.344140)	0.069499 / 0.075469 (-0.005970)

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)	1.571591 / 1.841788 (-0.270197)	22.543437 / 8.074308 (14.469129)	16.068426 / 10.191392 (5.877034)	0.169860 / 0.680424 (-0.510564)	0.021216 / 0.534201 (-0.512985)	0.468745 / 0.579283 (-0.110538)	0.475924 / 0.434364 (0.041560)	0.535574 / 0.540337 (-0.004763)	0.733823 / 1.386936 (-0.653113)

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)	0.008038 / 0.011353 (-0.003315)	0.004565 / 0.011008 (-0.006443)	0.076892 / 0.038508 (0.038384)	0.089559 / 0.023109 (0.066450)	0.456752 / 0.275898 (0.180854)	0.497282 / 0.323480 (0.173802)	0.005991 / 0.007986 (-0.001995)	0.003784 / 0.004328 (-0.000545)	0.076339 / 0.004250 (0.072089)	0.066050 / 0.037052 (0.028998)	0.462708 / 0.258489 (0.204219)	0.503711 / 0.293841 (0.209870)	0.037098 / 0.128546 (-0.091448)	0.009869 / 0.075646 (-0.065777)	0.083678 / 0.419271 (-0.335594)	0.058166 / 0.043533 (0.014633)	0.461839 / 0.255139 (0.206700)	0.481546 / 0.283200 (0.198347)	0.027755 / 0.141683 (-0.113928)	1.738490 / 1.452155 (0.286335)	1.832276 / 1.492716 (0.339560)

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)	0.329935 / 0.018006 (0.311929)	0.497438 / 0.000490 (0.496949)	0.034644 / 0.000200 (0.034444)	0.000199 / 0.000054 (0.000145)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.035427 / 0.037411 (-0.001984)	0.105689 / 0.014526 (0.091163)	0.117706 / 0.176557 (-0.058850)	0.177862 / 0.737135 (-0.559273)	0.116791 / 0.296338 (-0.179547)

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)	0.484851 / 0.215209 (0.269642)	4.804346 / 2.077655 (2.726691)	2.494801 / 1.504120 (0.990681)	2.320185 / 1.541195 (0.778990)	2.374090 / 1.468490 (0.905600)	0.567397 / 4.584777 (-4.017380)	4.087402 / 3.745712 (0.341690)	3.794245 / 5.269862 (-1.475616)	2.378481 / 4.565676 (-2.187195)	0.068228 / 0.424275 (-0.356047)	0.008740 / 0.007607 (0.001133)	0.574876 / 0.226044 (0.348832)	5.742644 / 2.268929 (3.473716)	3.047661 / 55.444624 (-52.396963)	2.729742 / 6.876477 (-4.146735)	2.852510 / 2.142072 (0.710438)	0.679450 / 4.805227 (-4.125777)	0.156162 / 6.500664 (-6.344502)	0.074051 / 0.075469 (-0.001418)

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)	1.576182 / 1.841788 (-0.265605)	23.298147 / 8.074308 (15.223839)	16.344621 / 10.191392 (6.153229)	0.167571 / 0.680424 (-0.512852)	0.021423 / 0.534201 (-0.512778)	0.464511 / 0.579283 (-0.114772)	0.453257 / 0.434364 (0.018893)	0.563439 / 0.540337 (0.023102)	0.764759 / 1.386936 (-0.622177)

mariosasko · 2023-08-17T18:48:19Z

This should also fix #6140, so please link it with this PR before merging.

lhoestq · 2023-08-17T20:14:51Z

Done !

mariosasko

Oops, forgot to approve 🙂!

PS: The data files resolution (and builder creation) is hard to follow when metadata_configs are present. It would be great if we could refactor this eventually to make it easier to maintain this part of the codebase.

github-actions · 2023-08-17T20:46:25Z

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)	0.006719 / 0.011353 (-0.004634)	0.004299 / 0.011008 (-0.006709)	0.085296 / 0.038508 (0.046788)	0.085144 / 0.023109 (0.062035)	0.361703 / 0.275898 (0.085805)	0.397721 / 0.323480 (0.074241)	0.005920 / 0.007986 (-0.002065)	0.003853 / 0.004328 (-0.000476)	0.065633 / 0.004250 (0.061383)	0.057000 / 0.037052 (0.019947)	0.379981 / 0.258489 (0.121492)	0.419041 / 0.293841 (0.125200)	0.031225 / 0.128546 (-0.097322)	0.008868 / 0.075646 (-0.066779)	0.288808 / 0.419271 (-0.130463)	0.052391 / 0.043533 (0.008859)	0.362349 / 0.255139 (0.107210)	0.399858 / 0.283200 (0.116658)	0.025843 / 0.141683 (-0.115840)	1.498988 / 1.452155 (0.046834)	1.547290 / 1.492716 (0.054574)

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)	0.278091 / 0.018006 (0.260085)	0.621794 / 0.000490 (0.621305)	0.003770 / 0.000200 (0.003570)	0.000084 / 0.000054 (0.000029)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.029128 / 0.037411 (-0.008283)	0.082061 / 0.014526 (0.067536)	0.101758 / 0.176557 (-0.074799)	0.155724 / 0.737135 (-0.581411)	0.102173 / 0.296338 (-0.194165)

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)	0.387145 / 0.215209 (0.171935)	3.868262 / 2.077655 (1.790607)	1.886440 / 1.504120 (0.382320)	1.723305 / 1.541195 (0.182111)	1.805411 / 1.468490 (0.336921)	0.485024 / 4.584777 (-4.099753)	3.637859 / 3.745712 (-0.107853)	3.319593 / 5.269862 (-1.950269)	2.087860 / 4.565676 (-2.477817)	0.056992 / 0.424275 (-0.367283)	0.007623 / 0.007607 (0.000016)	0.468182 / 0.226044 (0.242138)	4.681112 / 2.268929 (2.412183)	2.407010 / 55.444624 (-53.037614)	2.026604 / 6.876477 (-4.849872)	2.298158 / 2.142072 (0.156086)	0.581839 / 4.805227 (-4.223388)	0.132101 / 6.500664 (-6.368563)	0.060472 / 0.075469 (-0.014997)

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)	1.236422 / 1.841788 (-0.605365)	20.505168 / 8.074308 (12.430860)	14.356081 / 10.191392 (4.164689)	0.148808 / 0.680424 (-0.531616)	0.018433 / 0.534201 (-0.515768)	0.391323 / 0.579283 (-0.187960)	0.413142 / 0.434364 (-0.021222)	0.453484 / 0.540337 (-0.086853)	0.620771 / 1.386936 (-0.766165)

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)	0.007030 / 0.011353 (-0.004323)	0.004430 / 0.011008 (-0.006578)	0.065578 / 0.038508 (0.027070)	0.090751 / 0.023109 (0.067642)	0.389121 / 0.275898 (0.113223)	0.424657 / 0.323480 (0.101177)	0.006575 / 0.007986 (-0.001410)	0.003855 / 0.004328 (-0.000473)	0.066175 / 0.004250 (0.061925)	0.063255 / 0.037052 (0.026202)	0.397161 / 0.258489 (0.138672)	0.435291 / 0.293841 (0.141450)	0.031622 / 0.128546 (-0.096925)	0.008900 / 0.075646 (-0.066747)	0.071694 / 0.419271 (-0.347577)	0.049161 / 0.043533 (0.005628)	0.386214 / 0.255139 (0.131075)	0.404571 / 0.283200 (0.121372)	0.024821 / 0.141683 (-0.116862)	1.489514 / 1.452155 (0.037359)	1.576139 / 1.492716 (0.083423)

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)	0.289884 / 0.018006 (0.271878)	0.629342 / 0.000490 (0.628852)	0.004799 / 0.000200 (0.004599)	0.000160 / 0.000054 (0.000106)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.032081 / 0.037411 (-0.005331)	0.088152 / 0.014526 (0.073626)	0.107289 / 0.176557 (-0.069267)	0.164598 / 0.737135 (-0.572537)	0.108395 / 0.296338 (-0.187944)

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)	0.426723 / 0.215209 (0.211514)	4.267719 / 2.077655 (2.190064)	2.289657 / 1.504120 (0.785537)	2.117435 / 1.541195 (0.576240)	2.187292 / 1.468490 (0.718802)	0.478387 / 4.584777 (-4.106390)	3.625096 / 3.745712 (-0.120616)	3.408036 / 5.269862 (-1.861826)	2.124117 / 4.565676 (-2.441559)	0.056537 / 0.424275 (-0.367738)	0.007489 / 0.007607 (-0.000118)	0.502434 / 0.226044 (0.276389)	5.025357 / 2.268929 (2.756428)	2.740554 / 55.444624 (-52.704070)	2.418841 / 6.876477 (-4.457635)	2.730764 / 2.142072 (0.588691)	0.600013 / 4.805227 (-4.205214)	0.133039 / 6.500664 (-6.367625)	0.061466 / 0.075469 (-0.014003)

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)	1.330211 / 1.841788 (-0.511577)	21.092100 / 8.074308 (13.017792)	14.463054 / 10.191392 (4.271662)	0.154149 / 0.680424 (-0.526274)	0.018891 / 0.534201 (-0.515310)	0.393078 / 0.579283 (-0.186205)	0.415279 / 0.434364 (-0.019085)	0.479469 / 0.540337 (-0.060868)	0.659953 / 1.386936 (-0.726983)

* use yaml data_files instead of get_data_patterns when possible * minor fix docstring * update comment

lhoestq added 2 commits August 17, 2023 11:16

use yaml data_files instead of get_data_patterns when possible

7f598ea

minor fix docstring

acac888

thomwolf reviewed Aug 17, 2023

View reviewed changes

update comment

e8dc4b3

lhoestq marked this pull request as ready for review August 17, 2023 09:46

lhoestq requested a review from mariosasko August 17, 2023 09:47

mariosasko approved these changes Aug 17, 2023

View reviewed changes

lhoestq merged commit 5ca2ba0 into main Aug 17, 2023
13 checks passed

lhoestq deleted the use-yaml-instead-of-get_data_patterns-when-possible branch August 17, 2023 20:37

albertvillanova pushed a commit that referenced this pull request Oct 24, 2023

Use yaml instead of get data patterns when possible (#6154)

1de8527

* use yaml data_files instead of get_data_patterns when possible * minor fix docstring * update comment

Use yaml instead of get data patterns when possible #6154

Use yaml instead of get data patterns when possible #6154

Conversation

lhoestq commented Aug 17, 2023 • edited Loading

thomwolf Aug 17, 2023

Choose a reason for hiding this comment

lhoestq Aug 17, 2023

Choose a reason for hiding this comment

HuggingFaceDocBuilderDev commented Aug 17, 2023 • edited Loading

github-actions bot commented Aug 17, 2023

Benchmark: benchmark_array_xd.json

Benchmark: benchmark_getitem_100B.json

Benchmark: benchmark_indices_mapping.json

Benchmark: benchmark_iterating.json

Benchmark: benchmark_map_filter.json

Benchmark: benchmark_array_xd.json

Benchmark: benchmark_getitem_100B.json

Benchmark: benchmark_indices_mapping.json

Benchmark: benchmark_iterating.json

Benchmark: benchmark_map_filter.json

github-actions bot commented Aug 17, 2023

Benchmark: benchmark_array_xd.json

Benchmark: benchmark_getitem_100B.json

Benchmark: benchmark_indices_mapping.json

Benchmark: benchmark_iterating.json

Benchmark: benchmark_map_filter.json

Benchmark: benchmark_array_xd.json

Benchmark: benchmark_getitem_100B.json

Benchmark: benchmark_indices_mapping.json

Benchmark: benchmark_iterating.json

Benchmark: benchmark_map_filter.json

mariosasko commented Aug 17, 2023

lhoestq commented Aug 17, 2023

mariosasko left a comment

Choose a reason for hiding this comment

github-actions bot commented Aug 17, 2023

Benchmark: benchmark_array_xd.json

Benchmark: benchmark_getitem_100B.json

Benchmark: benchmark_indices_mapping.json

Benchmark: benchmark_iterating.json

Benchmark: benchmark_map_filter.json

Benchmark: benchmark_array_xd.json

Benchmark: benchmark_getitem_100B.json

Benchmark: benchmark_indices_mapping.json

Benchmark: benchmark_iterating.json

Benchmark: benchmark_map_filter.json

lhoestq commented Aug 17, 2023 •

edited

Loading

HuggingFaceDocBuilderDev commented Aug 17, 2023 •

edited

Loading