Fix WebDatasets KeyError for user-defined Features when a field is missing in an example

[ProGamerGov]

Fixes: #6900

Not sure if this needs any addition stuff before merging

[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]

Great ! I added a small suggestion to handle the case where the missing field is an image or an audio sample.

Could you also add a small test in test_webdataset.py ?
I believe you can reuse the same code as in test_image_webdataset and simply pass features= to Webdataset() and check if the examples have the extra features as None. Then we'll be good to merge !

[lhoestq]

Great ! I updated the test :) the rest of the CI failures are unrelated to this 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)	0.005188 / 0.011353 (-0.006165)	0.003812 / 0.011008 (-0.007196)	0.062408 / 0.038508 (0.023900)	0.030734 / 0.023109 (0.007625)	0.236528 / 0.275898 (-0.039370)	0.267684 / 0.323480 (-0.055796)	0.003182 / 0.007986 (-0.004804)	0.004009 / 0.004328 (-0.000319)	0.048966 / 0.004250 (0.044715)	0.045259 / 0.037052 (0.008207)	0.250586 / 0.258489 (-0.007903)	0.287079 / 0.293841 (-0.006762)	0.029235 / 0.128546 (-0.099311)	0.012216 / 0.075646 (-0.063431)	0.203864 / 0.419271 (-0.215408)	0.036324 / 0.043533 (-0.007209)	0.245180 / 0.255139 (-0.009959)	0.270327 / 0.283200 (-0.012872)	0.018676 / 0.141683 (-0.123007)	1.115568 / 1.452155 (-0.336586)	1.183267 / 1.492716 (-0.309449)

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)	0.094307 / 0.018006 (0.076301)	0.299071 / 0.000490 (0.298581)	0.000219 / 0.000200 (0.000019)	0.000042 / 0.000054 (-0.000012)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.018336 / 0.037411 (-0.019076)	0.062973 / 0.014526 (0.048447)	0.074137 / 0.176557 (-0.102420)	0.120553 / 0.737135 (-0.616582)	0.075411 / 0.296338 (-0.220927)

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)	0.284751 / 0.215209 (0.069542)	2.789294 / 2.077655 (0.711640)	1.457789 / 1.504120 (-0.046331)	1.339140 / 1.541195 (-0.202055)	1.341685 / 1.468490 (-0.126805)	0.714928 / 4.584777 (-3.869849)	2.361197 / 3.745712 (-1.384516)	2.791569 / 5.269862 (-2.478293)	1.892261 / 4.565676 (-2.673416)	0.077954 / 0.424275 (-0.346321)	0.005454 / 0.007607 (-0.002153)	0.350766 / 0.226044 (0.124721)	3.416749 / 2.268929 (1.147820)	1.835377 / 55.444624 (-53.609247)	1.506456 / 6.876477 (-5.370020)	1.642728 / 2.142072 (-0.499344)	0.791543 / 4.805227 (-4.013684)	0.133102 / 6.500664 (-6.367562)	0.042572 / 0.075469 (-0.032897)

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)	0.977958 / 1.841788 (-0.863830)	11.438271 / 8.074308 (3.363963)	9.305719 / 10.191392 (-0.885673)	0.141239 / 0.680424 (-0.539185)	0.014330 / 0.534201 (-0.519871)	0.302201 / 0.579283 (-0.277082)	0.261688 / 0.434364 (-0.172676)	0.338752 / 0.540337 (-0.201586)	0.468466 / 1.386936 (-0.918470)

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)	0.005629 / 0.011353 (-0.005723)	0.003997 / 0.011008 (-0.007011)	0.050447 / 0.038508 (0.011939)	0.031694 / 0.023109 (0.008585)	0.277392 / 0.275898 (0.001494)	0.290440 / 0.323480 (-0.033040)	0.004403 / 0.007986 (-0.003583)	0.002851 / 0.004328 (-0.001478)	0.048593 / 0.004250 (0.044343)	0.040622 / 0.037052 (0.003570)	0.282640 / 0.258489 (0.024151)	0.309390 / 0.293841 (0.015549)	0.031453 / 0.128546 (-0.097094)	0.012424 / 0.075646 (-0.063223)	0.060311 / 0.419271 (-0.358960)	0.033195 / 0.043533 (-0.010338)	0.266867 / 0.255139 (0.011728)	0.281966 / 0.283200 (-0.001234)	0.018026 / 0.141683 (-0.123657)	1.136273 / 1.452155 (-0.315882)	1.141643 / 1.492716 (-0.351073)

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)	0.095011 / 0.018006 (0.077005)	0.300571 / 0.000490 (0.300082)	0.000220 / 0.000200 (0.000020)	0.000044 / 0.000054 (-0.000011)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.022903 / 0.037411 (-0.014508)	0.077130 / 0.014526 (0.062604)	0.089576 / 0.176557 (-0.086980)	0.127156 / 0.737135 (-0.609980)	0.090008 / 0.296338 (-0.206331)

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)	0.289270 / 0.215209 (0.074061)	2.848239 / 2.077655 (0.770585)	1.546788 / 1.504120 (0.042668)	1.417275 / 1.541195 (-0.123920)	1.456214 / 1.468490 (-0.012276)	0.716688 / 4.584777 (-3.868088)	0.940242 / 3.745712 (-2.805470)	2.911956 / 5.269862 (-2.357906)	1.871358 / 4.565676 (-2.694318)	0.077553 / 0.424275 (-0.346722)	0.005279 / 0.007607 (-0.002328)	0.343338 / 0.226044 (0.117294)	3.368694 / 2.268929 (1.099766)	1.896765 / 55.444624 (-53.547859)	1.612414 / 6.876477 (-5.264063)	1.615934 / 2.142072 (-0.526138)	0.794016 / 4.805227 (-4.011212)	0.131821 / 6.500664 (-6.368843)	0.041495 / 0.075469 (-0.033975)

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)	1.003418 / 1.841788 (-0.838370)	12.073906 / 8.074308 (3.999598)	10.166291 / 10.191392 (-0.025101)	0.131224 / 0.680424 (-0.549200)	0.015246 / 0.534201 (-0.518955)	0.299835 / 0.579283 (-0.279448)	0.124308 / 0.434364 (-0.310056)	0.336414 / 0.540337 (-0.203924)	0.429569 / 1.386936 (-0.957367)

[ProGamerGov]

@lhoestq Thank you!