Skip to content

Less zip false positives #5640

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 4 commits into from
Mar 16, 2023
Merged

Less zip false positives #5640

merged 4 commits into from
Mar 16, 2023

Conversation

@lhoestq
Copy link
Member

@lhoestq lhoestq commented Mar 15, 2023 •
edited
Loading

zipfile.is_zipfile return false positives for some Parquet files. It causes errors when loading certain parquet datasets, where some files are considered ZIP files by zipfile.is_zipfile

This is a known issue: python/cpython#72680

At first I wanted to rely only on magic numbers, but then I found that someone contributed a fix to is_zipfile - do you think we should use it @albertvillanova or not ?

IMO it's ok to rely on magic numbers only for now, since in streaming mode we've had no issue checking only the magic number so far.

Close #5639

@github-actions
Copy link

github-actions bot commented Mar 15, 2023

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) 0.006998 / 0.011353 (-0.004355) 0.005093 / 0.011008 (-0.005916) 0.100490 / 0.038508 (0.061982) 0.032736 / 0.023109 (0.009627) 0.297738 / 0.275898 (0.021840) 0.322255 / 0.323480 (-0.001225) 0.005583 / 0.007986 (-0.002402) 0.004007 / 0.004328 (-0.000321) 0.075863 / 0.004250 (0.071613) 0.044212 / 0.037052 (0.007159) 0.300033 / 0.258489 (0.041544) 0.341997 / 0.293841 (0.048156) 0.036172 / 0.128546 (-0.092374) 0.012176 / 0.075646 (-0.063471) 0.356052 / 0.419271 (-0.063220) 0.050438 / 0.043533 (0.006905) 0.294677 / 0.255139 (0.039538) 0.318050 / 0.283200 (0.034850) 0.104733 / 0.141683 (-0.036950) 1.435681 / 1.452155 (-0.016474) 1.534793 / 1.492716 (0.042076)

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) 0.242815 / 0.018006 (0.224809) 0.565983 / 0.000490 (0.565494) 0.006800 / 0.000200 (0.006600) 0.000124 / 0.000054 (0.000070)

Benchmark: benchmark_indices_mapping.json

metric select shard shuffle sort train_test_split
new / old (diff) 0.026548 / 0.037411 (-0.010863) 0.104816 / 0.014526 (0.090290) 0.116222 / 0.176557 (-0.060335) 0.172143 / 0.737135 (-0.564992) 0.121631 / 0.296338 (-0.174707)

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) 0.400126 / 0.215209 (0.184917) 4.004538 / 2.077655 (1.926883) 1.798822 / 1.504120 (0.294702) 1.595191 / 1.541195 (0.053996) 1.645777 / 1.468490 (0.177287) 0.705643 / 4.584777 (-3.879134) 3.750887 / 3.745712 (0.005175) 2.136547 / 5.269862 (-3.133315) 1.475881 / 4.565676 (-3.089795) 0.086921 / 0.424275 (-0.337354) 0.012379 / 0.007607 (0.004771) 0.505824 / 0.226044 (0.279779) 5.052364 / 2.268929 (2.783435) 2.279983 / 55.444624 (-53.164641) 1.932253 / 6.876477 (-4.944224) 2.051359 / 2.142072 (-0.090714) 0.851906 / 4.805227 (-3.953321) 0.169566 / 6.500664 (-6.331098) 0.064600 / 0.075469 (-0.010869)

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) 1.165859 / 1.841788 (-0.675929) 15.049950 / 8.074308 (6.975642) 14.095981 / 10.191392 (3.904589) 0.151779 / 0.680424 (-0.528645) 0.017537 / 0.534201 (-0.516664) 0.420164 / 0.579283 (-0.159119) 0.418932 / 0.434364 (-0.015432) 0.488749 / 0.540337 (-0.051588) 0.582359 / 1.386936 (-0.804577)
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) 0.007426 / 0.011353 (-0.003927) 0.005248 / 0.011008 (-0.005761) 0.074118 / 0.038508 (0.035610) 0.034223 / 0.023109 (0.011114) 0.337780 / 0.275898 (0.061882) 0.376300 / 0.323480 (0.052820) 0.006142 / 0.007986 (-0.001843) 0.004246 / 0.004328 (-0.000083) 0.074177 / 0.004250 (0.069926) 0.052698 / 0.037052 (0.015646) 0.340229 / 0.258489 (0.081740) 0.396172 / 0.293841 (0.102331) 0.037293 / 0.128546 (-0.091253) 0.012514 / 0.075646 (-0.063132) 0.087144 / 0.419271 (-0.332128) 0.051922 / 0.043533 (0.008390) 0.333188 / 0.255139 (0.078049) 0.355420 / 0.283200 (0.072220) 0.110273 / 0.141683 (-0.031410) 1.447826 / 1.452155 (-0.004329) 1.561135 / 1.492716 (0.068419)

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) 0.269203 / 0.018006 (0.251197) 0.551997 / 0.000490 (0.551508) 0.001558 / 0.000200 (0.001359) 0.000090 / 0.000054 (0.000035)

Benchmark: benchmark_indices_mapping.json

metric select shard shuffle sort train_test_split
new / old (diff) 0.029511 / 0.037411 (-0.007900) 0.108614 / 0.014526 (0.094089) 0.123438 / 0.176557 (-0.053118) 0.171596 / 0.737135 (-0.565539) 0.126828 / 0.296338 (-0.169511)

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) 0.420520 / 0.215209 (0.205310) 4.175672 / 2.077655 (2.098017) 1.982220 / 1.504120 (0.478101) 1.788575 / 1.541195 (0.247381) 1.860840 / 1.468490 (0.392349) 0.706730 / 4.584777 (-3.878047) 3.858718 / 3.745712 (0.113005) 3.069389 / 5.269862 (-2.200472) 1.827603 / 4.565676 (-2.738073) 0.087893 / 0.424275 (-0.336382) 0.012613 / 0.007607 (0.005006) 0.524177 / 0.226044 (0.298132) 5.177077 / 2.268929 (2.908148) 2.494397 / 55.444624 (-52.950227) 2.189484 / 6.876477 (-4.686992) 2.217626 / 2.142072 (0.075554) 0.846326 / 4.805227 (-3.958901) 0.176558 / 6.500664 (-6.324106) 0.065018 / 0.075469 (-0.010451)

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) 1.268618 / 1.841788 (-0.573170) 15.132711 / 8.074308 (7.058403) 14.585530 / 10.191392 (4.394138) 0.163454 / 0.680424 (-0.516970) 0.017442 / 0.534201 (-0.516759) 0.421746 / 0.579283 (-0.157537) 0.425412 / 0.434364 (-0.008952) 0.499178 / 0.540337 (-0.041159) 0.595458 / 1.386936 (-0.791478)

@HuggingFaceDocBuilderDev
Copy link

HuggingFaceDocBuilderDev commented Mar 15, 2023 •
edited
Loading

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

@lhoestq lhoestq marked this pull request as ready for review March 15, 2023 17:08
@github-actions
Copy link

github-actions bot commented Mar 15, 2023

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) 0.007980 / 0.011353 (-0.003373) 0.005414 / 0.011008 (-0.005594) 0.099226 / 0.038508 (0.060718) 0.035442 / 0.023109 (0.012332) 0.304851 / 0.275898 (0.028952) 0.337144 / 0.323480 (0.013664) 0.006162 / 0.007986 (-0.001823) 0.004151 / 0.004328 (-0.000177) 0.074708 / 0.004250 (0.070458) 0.049690 / 0.037052 (0.012638) 0.307658 / 0.258489 (0.049168) 0.358472 / 0.293841 (0.064631) 0.037181 / 0.128546 (-0.091365) 0.012259 / 0.075646 (-0.063387) 0.335426 / 0.419271 (-0.083846) 0.050790 / 0.043533 (0.007257) 0.301715 / 0.255139 (0.046576) 0.320834 / 0.283200 (0.037634) 0.102357 / 0.141683 (-0.039326) 1.454750 / 1.452155 (0.002596) 1.571994 / 1.492716 (0.079278)

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) 0.218708 / 0.018006 (0.200702) 0.444391 / 0.000490 (0.443901) 0.005717 / 0.000200 (0.005517) 0.000089 / 0.000054 (0.000035)

Benchmark: benchmark_indices_mapping.json

metric select shard shuffle sort train_test_split
new / old (diff) 0.028017 / 0.037411 (-0.009395) 0.112753 / 0.014526 (0.098227) 0.121003 / 0.176557 (-0.055554) 0.181085 / 0.737135 (-0.556050) 0.127211 / 0.296338 (-0.169127)

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) 0.400803 / 0.215209 (0.185594) 4.007315 / 2.077655 (1.929660) 1.826911 / 1.504120 (0.322791) 1.637799 / 1.541195 (0.096605) 1.699754 / 1.468490 (0.231264) 0.709413 / 4.584777 (-3.875364) 4.008904 / 3.745712 (0.263192) 3.916540 / 5.269862 (-1.353322) 1.902102 / 4.565676 (-2.663575) 0.089048 / 0.424275 (-0.335227) 0.012763 / 0.007607 (0.005155) 0.498957 / 0.226044 (0.272913) 4.979865 / 2.268929 (2.710937) 2.301987 / 55.444624 (-53.142637) 1.929404 / 6.876477 (-4.947073) 2.107839 / 2.142072 (-0.034233) 0.857253 / 4.805227 (-3.947974) 0.171935 / 6.500664 (-6.328729) 0.066753 / 0.075469 (-0.008716)

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) 1.186811 / 1.841788 (-0.654977) 15.866319 / 8.074308 (7.792011) 14.738555 / 10.191392 (4.547163) 0.142879 / 0.680424 (-0.537544) 0.017679 / 0.534201 (-0.516522) 0.422840 / 0.579283 (-0.156443) 0.450307 / 0.434364 (0.015943) 0.491802 / 0.540337 (-0.048536) 0.588837 / 1.386936 (-0.798099)
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) 0.007659 / 0.011353 (-0.003694) 0.005331 / 0.011008 (-0.005678) 0.075360 / 0.038508 (0.036852) 0.034011 / 0.023109 (0.010902) 0.354488 / 0.275898 (0.078590) 0.401781 / 0.323480 (0.078301) 0.005806 / 0.007986 (-0.002179) 0.004029 / 0.004328 (-0.000300) 0.073822 / 0.004250 (0.069572) 0.049067 / 0.037052 (0.012015) 0.364483 / 0.258489 (0.105994) 0.405637 / 0.293841 (0.111796) 0.037166 / 0.128546 (-0.091380) 0.012397 / 0.075646 (-0.063249) 0.087346 / 0.419271 (-0.331926) 0.050888 / 0.043533 (0.007355) 0.334796 / 0.255139 (0.079657) 0.387681 / 0.283200 (0.104481) 0.105056 / 0.141683 (-0.036627) 1.471630 / 1.452155 (0.019475) 1.554764 / 1.492716 (0.062047)

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) 0.231825 / 0.018006 (0.213819) 0.449746 / 0.000490 (0.449256) 0.000888 / 0.000200 (0.000688) 0.000078 / 0.000054 (0.000023)

Benchmark: benchmark_indices_mapping.json

metric select shard shuffle sort train_test_split
new / old (diff) 0.030363 / 0.037411 (-0.007049) 0.115234 / 0.014526 (0.100708) 0.123005 / 0.176557 (-0.053551) 0.172772 / 0.737135 (-0.564363) 0.127818 / 0.296338 (-0.168520)

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) 0.425761 / 0.215209 (0.210552) 4.237950 / 2.077655 (2.160295) 1.992045 / 1.504120 (0.487925) 1.801622 / 1.541195 (0.260427) 1.918477 / 1.468490 (0.449987) 0.722730 / 4.584777 (-3.862047) 4.015968 / 3.745712 (0.270256) 3.720412 / 5.269862 (-1.549450) 1.763111 / 4.565676 (-2.802566) 0.089041 / 0.424275 (-0.335234) 0.012608 / 0.007607 (0.005001) 0.522645 / 0.226044 (0.296601) 5.227108 / 2.268929 (2.958180) 2.444714 / 55.444624 (-52.999910) 2.109745 / 6.876477 (-4.766732) 2.194042 / 2.142072 (0.051969) 0.871781 / 4.805227 (-3.933447) 0.173149 / 6.500664 (-6.327515) 0.066192 / 0.075469 (-0.009277)

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) 1.312051 / 1.841788 (-0.529737) 16.024315 / 8.074308 (7.950007) 15.123823 / 10.191392 (4.932431) 0.163997 / 0.680424 (-0.516427) 0.017595 / 0.534201 (-0.516606) 0.426379 / 0.579283 (-0.152904) 0.467709 / 0.434364 (0.033345) 0.498308 / 0.540337 (-0.042030) 0.591426 / 1.386936 (-0.795510)

@lhoestq lhoestq requested a review from albertvillanova March 15, 2023 17:13
@lhoestq
Copy link
Member Author

lhoestq commented Mar 15, 2023 •
edited
Loading

CI is failing due to unrelated issues, hopefully #5642 fixes it

albertvillanova
albertvillanova approved these changes Mar 16, 2023
View reviewed changes
Copy link
Member

@albertvillanova albertvillanova left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the improvement.

I agree we should tweak the zipfile.is_zipfile default implementation if that is flaky.

src/datasets/utils/extract.py
if centdir[_CD_SIGNATURE] == stringCentralDir:
return True # First central directory entry has correct magic number
except Exception: # catch all errors in case future python versions change the zipfile internals
return False
Copy link
Member

@albertvillanova albertvillanova Mar 16, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please note that this function as it is could return None. To fix this:

Suggested change
return False
return False
return False

tests/test_extract.py
with not_a_zip_file.open("wb") as f:
f.write(data)
assert zipfile.is_zipfile(str(not_a_zip_file)) # is a false positive for `zipfile`
assert not ZipExtractor.is_extractable(not_a_zip_file) # but we're right
Copy link
Member

@albertvillanova albertvillanova Mar 16, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The test passes because not None is True.

albertvillanova
albertvillanova approved these changes Mar 16, 2023
View reviewed changes
@lhoestq @albertvillanova
Update src/datasets/utils/extract.py
e2f8e17 
Co-authored-by: Albert Villanova del Moral <[email protected]>
@github-actions
Copy link

github-actions bot commented Mar 16, 2023

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) 0.006478 / 0.011353 (-0.004875) 0.004347 / 0.011008 (-0.006661) 0.097103 / 0.038508 (0.058595) 0.027650 / 0.023109 (0.004541) 0.372355 / 0.275898 (0.096457) 0.408794 / 0.323480 (0.085314) 0.005034 / 0.007986 (-0.002952) 0.003252 / 0.004328 (-0.001076) 0.074068 / 0.004250 (0.069818) 0.035542 / 0.037052 (-0.001510) 0.367392 / 0.258489 (0.108903) 0.409644 / 0.293841 (0.115803) 0.031745 / 0.128546 (-0.096801) 0.011501 / 0.075646 (-0.064145) 0.323355 / 0.419271 (-0.095917) 0.043065 / 0.043533 (-0.000467) 0.377313 / 0.255139 (0.122174) 0.395326 / 0.283200 (0.112127) 0.087101 / 0.141683 (-0.054582) 1.461228 / 1.452155 (0.009073) 1.529413 / 1.492716 (0.036696)

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) 0.199245 / 0.018006 (0.181239) 0.409978 / 0.000490 (0.409488) 0.002655 / 0.000200 (0.002455) 0.000070 / 0.000054 (0.000016)

Benchmark: benchmark_indices_mapping.json

metric select shard shuffle sort train_test_split
new / old (diff) 0.023903 / 0.037411 (-0.013508) 0.097855 / 0.014526 (0.083330) 0.106405 / 0.176557 (-0.070152) 0.166889 / 0.737135 (-0.570247) 0.110256 / 0.296338 (-0.186082)

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) 0.440351 / 0.215209 (0.225142) 4.382848 / 2.077655 (2.305194) 2.049602 / 1.504120 (0.545482) 1.824638 / 1.541195 (0.283443) 1.850519 / 1.468490 (0.382029) 0.702652 / 4.584777 (-3.882125) 3.394571 / 3.745712 (-0.351141) 1.940608 / 5.269862 (-3.329254) 1.263961 / 4.565676 (-3.301716) 0.083985 / 0.424275 (-0.340290) 0.013046 / 0.007607 (0.005439) 0.538272 / 0.226044 (0.312228) 5.407563 / 2.268929 (3.138634) 2.519207 / 55.444624 (-52.925418) 2.153379 / 6.876477 (-4.723098) 2.394512 / 2.142072 (0.252439) 0.812840 / 4.805227 (-3.992387) 0.152868 / 6.500664 (-6.347796) 0.067823 / 0.075469 (-0.007646)

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) 1.220031 / 1.841788 (-0.621757) 13.781237 / 8.074308 (5.706929) 14.203975 / 10.191392 (4.012583) 0.141077 / 0.680424 (-0.539347) 0.016518 / 0.534201 (-0.517682) 0.379079 / 0.579283 (-0.200204) 0.378916 / 0.434364 (-0.055448) 0.434589 / 0.540337 (-0.105749) 0.521129 / 1.386936 (-0.865807)
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) 0.006997 / 0.011353 (-0.004356) 0.004599 / 0.011008 (-0.006410) 0.078700 / 0.038508 (0.040192) 0.027902 / 0.023109 (0.004793) 0.344406 / 0.275898 (0.068508) 0.392918 / 0.323480 (0.069438) 0.005175 / 0.007986 (-0.002811) 0.004755 / 0.004328 (0.000427) 0.077707 / 0.004250 (0.073457) 0.039409 / 0.037052 (0.002357) 0.343250 / 0.258489 (0.084761) 0.405544 / 0.293841 (0.111703) 0.032286 / 0.128546 (-0.096260) 0.011674 / 0.075646 (-0.063972) 0.087633 / 0.419271 (-0.331639) 0.043346 / 0.043533 (-0.000186) 0.355076 / 0.255139 (0.099937) 0.382155 / 0.283200 (0.098955) 0.090914 / 0.141683 (-0.050769) 1.518369 / 1.452155 (0.066215) 1.583530 / 1.492716 (0.090813)

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) 0.160369 / 0.018006 (0.142362) 0.406844 / 0.000490 (0.406354) 0.002651 / 0.000200 (0.002451) 0.000080 / 0.000054 (0.000025)

Benchmark: benchmark_indices_mapping.json

metric select shard shuffle sort train_test_split
new / old (diff) 0.025295 / 0.037411 (-0.012116) 0.101490 / 0.014526 (0.086964) 0.108825 / 0.176557 (-0.067732) 0.161673 / 0.737135 (-0.575462) 0.113610 / 0.296338 (-0.182729)

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) 0.443514 / 0.215209 (0.228305) 4.436722 / 2.077655 (2.359067) 2.144008 / 1.504120 (0.639888) 2.005324 / 1.541195 (0.464129) 2.123356 / 1.468490 (0.654866) 0.697217 / 4.584777 (-3.887560) 3.401105 / 3.745712 (-0.344607) 1.874621 / 5.269862 (-3.395240) 1.165069 / 4.565676 (-3.400608) 0.082799 / 0.424275 (-0.341476) 0.012806 / 0.007607 (0.005199) 0.542688 / 0.226044 (0.316644) 5.420963 / 2.268929 (3.152034) 2.579034 / 55.444624 (-52.865590) 2.240201 / 6.876477 (-4.636276) 2.261309 / 2.142072 (0.119237) 0.800246 / 4.805227 (-4.004981) 0.150380 / 6.500664 (-6.350285) 0.066880 / 0.075469 (-0.008589)

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) 1.281721 / 1.841788 (-0.560067) 13.906361 / 8.074308 (5.832053) 14.135336 / 10.191392 (3.943944) 0.128865 / 0.680424 (-0.551559) 0.016452 / 0.534201 (-0.517749) 0.373563 / 0.579283 (-0.205720) 0.385321 / 0.434364 (-0.049043) 0.437198 / 0.540337 (-0.103139) 0.530720 / 1.386936 (-0.856216)

@lhoestq lhoestq merged commit 11cd0f7 into main Mar 16, 2023
@lhoestq lhoestq deleted the less-zip-false-positives branch March 16, 2023 13:40
@github-actions
Copy link

github-actions bot commented Mar 16, 2023

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) 0.008099 / 0.011353 (-0.003254) 0.005093 / 0.011008 (-0.005916) 0.106258 / 0.038508 (0.067750) 0.037051 / 0.023109 (0.013942) 0.347960 / 0.275898 (0.072062) 0.370849 / 0.323480 (0.047369) 0.006122 / 0.007986 (-0.001863) 0.004094 / 0.004328 (-0.000235) 0.079549 / 0.004250 (0.075299) 0.046563 / 0.037052 (0.009510) 0.332735 / 0.258489 (0.074246) 0.417061 / 0.293841 (0.123220) 0.038105 / 0.128546 (-0.090441) 0.011886 / 0.075646 (-0.063760) 0.342103 / 0.419271 (-0.077169) 0.053233 / 0.043533 (0.009700) 0.344754 / 0.255139 (0.089615) 0.355354 / 0.283200 (0.072155) 0.101059 / 0.141683 (-0.040624) 1.518561 / 1.452155 (0.066406) 1.558652 / 1.492716 (0.065935)

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) 0.225919 / 0.018006 (0.207913) 0.518539 / 0.000490 (0.518049) 0.006230 / 0.000200 (0.006030) 0.000124 / 0.000054 (0.000070)

Benchmark: benchmark_indices_mapping.json

metric select shard shuffle sort train_test_split
new / old (diff) 0.026782 / 0.037411 (-0.010629) 0.108457 / 0.014526 (0.093931) 0.125203 / 0.176557 (-0.051353) 0.175726 / 0.737135 (-0.561409) 0.127051 / 0.296338 (-0.169287)

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) 0.416427 / 0.215209 (0.201217) 4.168851 / 2.077655 (2.091196) 1.962238 / 1.504120 (0.458118) 1.825224 / 1.541195 (0.284029) 1.831200 / 1.468490 (0.362710) 0.765526 / 4.584777 (-3.819250) 4.303957 / 3.745712 (0.558245) 2.193467 / 5.269862 (-3.076395) 1.654605 / 4.565676 (-2.911071) 0.096709 / 0.424275 (-0.327566) 0.013792 / 0.007607 (0.006185) 0.537862 / 0.226044 (0.311818) 5.152230 / 2.268929 (2.883302) 2.520938 / 55.444624 (-52.923686) 2.108422 / 6.876477 (-4.768054) 2.214220 / 2.142072 (0.072147) 0.834320 / 4.805227 (-3.970907) 0.170635 / 6.500664 (-6.330029) 0.063131 / 0.075469 (-0.012338)

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) 1.215767 / 1.841788 (-0.626020) 15.254781 / 8.074308 (7.180473) 14.360764 / 10.191392 (4.169372) 0.172511 / 0.680424 (-0.507913) 0.020161 / 0.534201 (-0.514040) 0.426936 / 0.579283 (-0.152347) 0.438771 / 0.434364 (0.004407) 0.486973 / 0.540337 (-0.053364) 0.584238 / 1.386936 (-0.802698)
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) 0.006777 / 0.011353 (-0.004576) 0.005304 / 0.011008 (-0.005704) 0.073717 / 0.038508 (0.035209) 0.033604 / 0.023109 (0.010494) 0.340448 / 0.275898 (0.064550) 0.351861 / 0.323480 (0.028381) 0.005786 / 0.007986 (-0.002199) 0.005013 / 0.004328 (0.000685) 0.071263 / 0.004250 (0.067012) 0.048189 / 0.037052 (0.011137) 0.339457 / 0.258489 (0.080968) 0.384383 / 0.293841 (0.090542) 0.035563 / 0.128546 (-0.092983) 0.011509 / 0.075646 (-0.064137) 0.083722 / 0.419271 (-0.335550) 0.048886 / 0.043533 (0.005353) 0.350184 / 0.255139 (0.095045) 0.361037 / 0.283200 (0.077837) 0.105191 / 0.141683 (-0.036492) 1.503247 / 1.452155 (0.051093) 1.582298 / 1.492716 (0.089581)

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) 0.221687 / 0.018006 (0.203681) 0.466489 / 0.000490 (0.465999) 0.000484 / 0.000200 (0.000284) 0.000069 / 0.000054 (0.000015)

Benchmark: benchmark_indices_mapping.json

metric select shard shuffle sort train_test_split
new / old (diff) 0.027978 / 0.037411 (-0.009434) 0.119572 / 0.014526 (0.105047) 0.133530 / 0.176557 (-0.043026) 0.177892 / 0.737135 (-0.559243) 0.127045 / 0.296338 (-0.169294)

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) 0.430198 / 0.215209 (0.214989) 4.435512 / 2.077655 (2.357858) 2.007183 / 1.504120 (0.503063) 1.799230 / 1.541195 (0.258036) 1.884750 / 1.468490 (0.416260) 0.745232 / 4.584777 (-3.839545) 4.088069 / 3.745712 (0.342357) 4.114669 / 5.269862 (-1.155193) 2.374086 / 4.565676 (-2.191590) 0.089154 / 0.424275 (-0.335121) 0.012938 / 0.007607 (0.005331) 0.505954 / 0.226044 (0.279909) 5.194226 / 2.268929 (2.925298) 2.487230 / 55.444624 (-52.957394) 2.163353 / 6.876477 (-4.713124) 2.177879 / 2.142072 (0.035807) 0.828728 / 4.805227 (-3.976499) 0.171157 / 6.500664 (-6.329507) 0.062883 / 0.075469 (-0.012586)

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) 1.275906 / 1.841788 (-0.565882) 15.235484 / 8.074308 (7.161176) 14.467396 / 10.191392 (4.276004) 0.198994 / 0.680424 (-0.481430) 0.020203 / 0.534201 (-0.513998) 0.447904 / 0.579283 (-0.131380) 0.454210 / 0.434364 (0.019846) 0.528062 / 0.540337 (-0.012275) 0.619311 / 1.386936 (-0.767625)

@lhoestq lhoestq mentioned this pull request May 23, 2023
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@albertvillanova albertvillanova albertvillanova approved these changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

Parquet file wrongly recognized as zip prevents loading a dataset

4 participants

@lhoestq @HuggingFaceDocBuilderDev @albertvillanova