-
Notifications
You must be signed in to change notification settings - Fork 22
/
Copy path10.21105.joss.04611.jats
783 lines (783 loc) · 39.9 KB
/
10.21105.joss.04611.jats
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
<?xml version="1.0" encoding="utf-8" ?>
<!DOCTYPE article PUBLIC "-//NLM//DTD JATS (Z39.96) Journal Publishing DTD v1.2 20190208//EN"
"JATS-publishing1.dtd">
<article xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" dtd-version="1.2" article-type="other">
<front>
<journal-meta>
<journal-id></journal-id>
<journal-title-group>
<journal-title>Journal of Open Source Software</journal-title>
<abbrev-journal-title>JOSS</abbrev-journal-title>
</journal-title-group>
<issn publication-format="electronic">2475-9066</issn>
<publisher>
<publisher-name>Open Journals</publisher-name>
</publisher>
</journal-meta>
<article-meta>
<article-id pub-id-type="publisher-id">4611</article-id>
<article-id pub-id-type="doi">10.21105/joss.04611</article-id>
<title-group>
<article-title>fseval: A Benchmarking Framework for Feature Selection
and Feature Ranking Algorithms</article-title>
</title-group>
<contrib-group>
<contrib contrib-type="author">
<contrib-id contrib-id-type="orcid">0000-0003-3304-3800</contrib-id>
<name>
<surname>Overschie</surname>
<given-names>Jeroen G. S.</given-names>
</name>
<xref ref-type="aff" rid="aff-1"/>
</contrib>
<contrib contrib-type="author">
<contrib-id contrib-id-type="orcid">0000-0002-0770-1390</contrib-id>
<name>
<surname>Alsahaf</surname>
<given-names>Ahmad</given-names>
</name>
<xref ref-type="aff" rid="aff-2"/>
</contrib>
<contrib contrib-type="author">
<contrib-id contrib-id-type="orcid">0000-0001-6552-2596</contrib-id>
<name>
<surname>Azzopardi</surname>
<given-names>George</given-names>
</name>
<xref ref-type="aff" rid="aff-1"/>
</contrib>
<aff id="aff-1">
<institution-wrap>
<institution>Bernoulli Institute for Mathematics, Computer Science and
Artificial Intelligence, University of Groningen, P.O. Box 407, 9700 AK
Groningen, The Netherlands</institution>
</institution-wrap>
</aff>
<aff id="aff-2">
<institution-wrap>
<institution>Department of Biomedical Sciences of Cells and Systems,
University Medical Center Groningen, University of Groningen, 9713 GZ
Groningen, The Netherlands</institution>
</institution-wrap>
</aff>
</contrib-group>
<pub-date date-type="pub" publication-format="electronic" iso-8601-date="2022-07-06">
<day>6</day>
<month>7</month>
<year>2022</year>
</pub-date>
<volume>7</volume>
<issue>79</issue>
<fpage>4611</fpage>
<permissions>
<copyright-statement>Authors of papers retain copyright and release the
work under a Creative Commons Attribution 4.0 International License (CC
BY 4.0)</copyright-statement>
<copyright-year>2022</copyright-year>
<copyright-holder>The article authors</copyright-holder>
<license license-type="open-access" xlink:href="https://creativecommons.org/licenses/by/4.0/">
<license-p>Authors of papers retain copyright and release the work under
a Creative Commons Attribution 4.0 International License (CC BY
4.0)</license-p>
</license>
</permissions>
<kwd-group kwd-group-type="author">
<kwd>feature ranking</kwd>
<kwd>feature selection</kwd>
<kwd>benchmarking</kwd>
<kwd>machine learning</kwd>
<kwd>open-source</kwd>
<kwd>software</kwd>
<kwd>python</kwd>
</kwd-group>
</article-meta>
</front>
<body>
<sec id="summary">
<title>Summary</title>
<p>The <monospace>fseval</monospace> Python package allows
benchmarking Feature Selection and Feature Ranking algorithms on a
large scale, and facilitates the comparison of multiple algorithms in
a systematic way. In particular, <monospace>fseval</monospace> enables
users to run experiments in parallel and distributed over multiple
machines, and export the results to an SQL database. The execution of
an experiment can be fully determined by a configuration file, which
means the experiment results can be reproduced easily, given only the
configuration file. <monospace>fseval</monospace> has high test
coverage, continuous integration, and rich documentation. The package
is open source and can be installed through PyPI. The source code is
available at:
<ext-link ext-link-type="uri" xlink:href="https://github.com/dunnkers/fseval">https://github.com/dunnkers/fseval</ext-link>.</p>
</sec>
<sec id="statement-of-need">
<title>Statement of Need</title>
<p>Feature Selection (FS) and Feature Ranking (FR) are extensively
researched topics within machine learning
(<xref alt="Guyon & Elisseeff, 2003" rid="ref-guyon_introduction_2003" ref-type="bibr">Guyon
& Elisseeff, 2003</xref>;
<xref alt="Venkatesh & Anuradha, 2019" rid="ref-venkatesh2019review" ref-type="bibr">Venkatesh
& Anuradha, 2019</xref>). FS methods determine subsets of relevant
features in a dataset, whereas FR methods rank the features in a
dataset relative to each other in terms of their relevance. When a new
FS or FR method is developed, a benchmarking scheme is necessary to
empirically validate its effectiveness. Often, the benchmark is
conducted as follows: features are ranked by importance, then the
predictive quality of the feature subsets containing the top ranked
features is evaluated using a validation estimator. Some studies let
the competing FS or FR algorithms pick out a fixed number of top
<inline-formula><alternatives>
<tex-math><![CDATA[k]]></tex-math>
<mml:math display="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mi>k</mml:mi></mml:math></alternatives></inline-formula>
features and validate the performance of that feature subset
(<xref alt="Bradley & Mangasarian, 1998" rid="ref-bradley_feature_1998" ref-type="bibr">Bradley
& Mangasarian, 1998</xref>;
<xref alt="Roffo et al., 2015" rid="ref-roffo_infinite_2015" ref-type="bibr">Roffo
et al., 2015</xref>;
<xref alt="Zhao & Liu, 2007" rid="ref-zhao_searching_2007" ref-type="bibr">Zhao
& Liu, 2007</xref>), whilst others evaluate multiple subsets of
increasing cardinality containing the highest ranked features
(<xref alt="Almuallim & Dietterich, 1991" rid="ref-almuallim_learning_1991" ref-type="bibr">Almuallim
& Dietterich, 1991</xref>;
<xref alt="Bennasar et al., 2015" rid="ref-bennasar_feature_2015" ref-type="bibr">Bennasar
et al., 2015</xref>;
<xref alt="Gu et al., 2012" rid="ref-gu_generalized_2012" ref-type="bibr">Gu
et al., 2012</xref>;
<xref alt="Kira & Rendell, 1992" rid="ref-kira_feature_1992" ref-type="bibr">Kira
& Rendell, 1992</xref>;
<xref alt="Peng et al., 2005" rid="ref-peng_feature_2005" ref-type="bibr">Peng
et al., 2005</xref>;
<xref alt="Wojtas & Chen, 2020" rid="ref-wojtas_feature_2020" ref-type="bibr">Wojtas
& Chen, 2020</xref>). FS algorithms that only make a binary
prediction on which features to keep, are always evaluated in the
former way.</p>
<p>There is a clear case for performing Feature Selection, as it has
been shown to improve classification performance in many tasks,
especially those with a large number of features and limited
observations. In those applications, it is difficult to determine
which FS method is suitable in the general case. Therefore, large
empirical comparisons of several FS methods and classifiers are
routinely performed. For instance, in microarray data
(<xref alt="Cilia et al., 2019" rid="ref-cilia2019experimental" ref-type="bibr">Cilia
et al., 2019</xref>), medical imaging
(<xref alt="Sun et al., 2019" rid="ref-sun2019comparison" ref-type="bibr">Sun
et al., 2019</xref>;
<xref alt="Tohka et al., 2016" rid="ref-tohka2016comparison" ref-type="bibr">Tohka
et al., 2016</xref>), and text classification
(<xref alt="Kou et al., 2020" rid="ref-kou2020evaluation" ref-type="bibr">Kou
et al., 2020</xref>;
<xref alt="Liu et al., 2017" rid="ref-liu2017multi" ref-type="bibr">Liu
et al., 2017</xref>). Therefore, it is valuable to find out
emperically which FR- or FS method works best. This requires running a
benchmark to do so.</p>
<p><monospace>fseval</monospace> is an open-source Python package that
helps researchers perform such benchmarks efficiently by eliminating
the need for implementing benchmarking pipelines from scratch to test
new methods. The pipeline only requires a well-defined configuration
file to run - the rest of the pipeline is automatically executed.
Because the entire experiment setup is deterministic and captured in a
configuration file, results of any experiment can be reproduced given
the configuration file. This can be very convenient to researchers in
order to prove the integrity of their benchmarks.</p>
<p>To the best of our knowledge, there is only one package that aims
to accomplish a similar goal (<monospace>featsel</monospace>,
(<xref alt="Reis et al., 2017" rid="ref-reis_featsel_2017" ref-type="bibr">Reis
et al., 2017</xref>)). Compared to this tool,
<monospace>fseval</monospace> is easier to install and use, has better
documentation, and is better maintained. <monospace>fseval</monospace>
also has more extensive functionalities compared to
<monospace>featsel</monospace>: with support for easily configurable
and reproducible pipeline configuration using either YAML or Python
and distributed-processing support. Due to the lack of functionality
and the fact that the refered-to library is out-of-date, we consider
there to be a gap in the field, which our library aims to fill.</p>
<list list-type="bullet">
<list-item>
<p>The <bold>target audiences</bold> are researchers in the
domains of Feature Selection and Feature Ranking, as well as
businesses that are looking for the best FR- or FS method to use
for their use case.</p>
</list-item>
<list-item>
<p>The <bold>scope</bold> of <monospace>fseval</monospace> is
limited to handle tabular datasets for the classification and
regression objectives.</p>
</list-item>
</list>
</sec>
<sec id="key-features">
<title>Key Features</title>
<p><monospace>fseval</monospace> is a flexible and unbiased framework
which provides as much useful functionality as possible. Most features
are optional, and can be enabled or disabled according to what the
user deems fit. The aim of the package is to accommodate the most
common benchmarking settings and protocols that feature selection
researchers use.</p>
<list list-type="bullet">
<list-item>
<p><bold>Algorithm support</bold>. FR or FS algorithms that
estimate the importance of features in various ways are supported,
including the following output attributes: (1) a
<sc>feature_importances_</sc> vector in
<inline-formula><alternatives>
<tex-math><![CDATA[\mathbb{R}]]></tex-math>
<mml:math display="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mstyle mathvariant="double-struck"><mml:mi>ℝ</mml:mi></mml:mstyle></mml:math></alternatives></inline-formula>,
(2) a <sc>ranking_</sc> vector in <inline-formula><alternatives>
<tex-math><![CDATA[\mathbb{Z}]]></tex-math>
<mml:math display="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mstyle mathvariant="double-struck"><mml:mi>ℤ</mml:mi></mml:mstyle></mml:math></alternatives></inline-formula>
and (3) a <sc>support_</sc> vector in
<inline-formula><alternatives>
<tex-math><![CDATA[\mathbb{B}]]></tex-math>
<mml:math display="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mstyle mathvariant="double-struck"><mml:mi>𝔹</mml:mi></mml:mstyle></mml:math></alternatives></inline-formula>.
An estimator might support any combination of the output
attributes. Once estimators are fit there is the option to save a
<italic>cached</italic> version.</p>
</list-item>
<list-item>
<p><bold>Dataset</bold> adapters. Datasets can be loaded from
multiple sources using <italic>adapters</italic>. Users can
implement adapters themselves by implementing a given interface,
or use a built-in adapter class to load datasets from OpenML
(<xref alt="Vanschoren et al., 2013" rid="ref-vanschoren_openml_2013" ref-type="bibr">Vanschoren
et al., 2013</xref>). Adapters might also be functions, which, for
example, allow users to directly use the sklearn functions
<monospace>make_classification</monospace> or
<monospace>make_regression</monospace> as adapters to create
<bold>synthetic</bold> datasets. Datasets might also define
dataset feature importance <italic>ground truths</italic>, which
can be used to compute metrics in the scoring stage (Section
‘<xref alt="Scoring" rid="sectionU003Ascoring">Scoring</xref>’).</p>
</list-item>
<list-item>
<p><bold>Built-in integrations</bold>.
<monospace>fseval</monospace> allows exporting benchmark results
directly to various SQL databases using SQLAlchemy
(<xref alt="Bayer, 2012" rid="ref-bayer_sqlalchemy_2012" ref-type="bibr">Bayer,
2012</xref>), or to the Weights and Biases experiment tracker
platform
(<xref alt="Biewald, 2020" rid="ref-biewald_experiment_2020" ref-type="bibr">Biewald,
2020</xref>). Users can create custom metrics and perform
aggregations over the bootstrap results.</p>
</list-item>
<list-item>
<p><bold>Scalable and distributed</bold> computing. Besides that
the process of running multiple bootstraps can be distributed over
the CPU, <monospace>fseval</monospace> also allows executing
experiments on SLURM clusters
(<xref alt="Yoo et al., 2003" rid="ref-yoo_slurm_2003" ref-type="bibr">Yoo
et al., 2003</xref>) or on the cloud platform AWS. This is
possible because all configuration regarding the execution of the
pipeline can be captured in a configuration file.</p>
</list-item>
<list-item>
<p><bold>Reproducible</bold> experiments. Because the entire
execution state of the pipeline can be expressed in a single
configuration file, it is easy to reproduce experiment results.
Given that a scientist uses estimates that are deterministic
(e.g. by fixing a <sc>random_state</sc> variable), others can
reproduce the results, improving the scientific integrity of the
work.</p>
</list-item>
</list>
</sec>
<sec id="the-pipeline">
<title>The Pipeline</title>
<p><monospace>fseval</monospace> executes a predefined sequence of
steps, as can be seen in Figure
<xref alt="1" rid="figU003Apipeline">1</xref>.</p>
<fig id="figU003Apipeline">
<caption><p>A schematic of the benchmarking pipeline. The input of
the pipeline is at all times a <monospace>PipelineConfig</monospace>
object, processed from YAML or Python by Hydra. After steps 1-6,
steps a-d are executed for both the fitting step and scoring
step.</p></caption>
<graphic mimetype="image" mime-subtype="svg+xml" xlink:href="media/pipeline.svg" xlink:title="" />
</fig>
<p>First, in step 1, the pipeline configuration is processed using
Hydra
(<xref alt="Yadan, 2019" rid="ref-yadan_hydra_2019" ref-type="bibr">Yadan,
2019</xref>). Hydra is a powerful tool for creating Command Line
Interfaces in Python, allowing hierarchical representation of the
configuration. Configuration can be defined in either YAML or Python
files, or a combination of the two. The top-level config is enforced
to be of the <monospace>PipelineConfig</monospace> interface, allowing
Hydra to perform type-checking. The config is then passed to the
<monospace>run_pipeline</monospace> function in step 2. Then, after
the dataset is loaded in step 3, the splits for cross validation are
determined in step 4. Each cross validation fold is executed in a
separate run of the pipeline. The training and testing subsets are
then given to the <italic>fitting</italic> and
<italic>scoring</italic> steps, steps 5 and 6, respectively.</p>
<sec id="fitting">
<title>Fitting</title>
<p>In the fitting step, the Feature- Ranker or Selector and
validation estimators are fit on the given training set. The
validation estimator is fit on all feature subsets that are desired
to be evaluated. For every bootstrap <inline-formula><alternatives>
<tex-math><![CDATA[b \in \{1, \dots, \texttt{PipelineConfig.n\_bootstraps}\}]]></tex-math>
<mml:math display="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mrow><mml:mi>b</mml:mi><mml:mo>∈</mml:mo><mml:mo stretchy="false" form="prefix">{</mml:mo><mml:mn>1</mml:mn><mml:mo>,</mml:mo><mml:mi>…</mml:mi><mml:mo>,</mml:mo><mml:mtext mathvariant="monospace">𝙿𝚒𝚙𝚎𝚕𝚒𝚗𝚎𝙲𝚘𝚗𝚏𝚒𝚐.𝚗_𝚋𝚘𝚘𝚝𝚜𝚝𝚛𝚊𝚙𝚜</mml:mtext><mml:mo stretchy="false" form="postfix">}</mml:mo></mml:mrow></mml:math></alternatives></inline-formula>,
a fit sequence is run. The bootstraps can be distributed over CPUs
by setting <monospace>PipelineConfig.n_jobs</monospace>
<inline-formula><alternatives>
<tex-math><![CDATA[> 1]]></tex-math>
<mml:math display="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mrow><mml:mo>></mml:mo><mml:mn>1</mml:mn></mml:mrow></mml:math></alternatives></inline-formula>.
The fit process consists of the following steps.</p>
<list list-type="alpha-lower">
<list-item>
<label>(a)</label>
<p>The dataset is resampled according to the
<monospace>PipelineConfig.resample</monospace> config, using
<monospace>random_state</monospace>
<inline-formula><alternatives>
<tex-math><![CDATA[= b]]></tex-math>
<mml:math display="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mrow><mml:mo>=</mml:mo><mml:mi>b</mml:mi></mml:mrow></mml:math></alternatives></inline-formula>.</p>
</list-item>
<list-item>
<label>(b)</label>
<p>The FR or FS algorithm is fit. Then, the estimator can be
cached as a pickle file.</p>
</list-item>
<list-item>
<label>(c)</label>
<p>If the ranker estimates <sc>support_</sc> (<italic>Feature
Selection</italic>): The selected feature subset is validated
using the validation estimator.</p>
</list-item>
<list-item>
<label>(d)</label>
<p>If the ranker estimates <sc>feature_importances_</sc> or
<sc>ranking_</sc> (<italic>Feature Ranking</italic>) then every
number in the list
<monospace>PipelineConfig.all_features_to_select</monospace> is
used to take the <inline-formula><alternatives>
<tex-math><![CDATA[k]]></tex-math>
<mml:math display="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mi>k</mml:mi></mml:math></alternatives></inline-formula>
best features in the ranking, and fitting the validation
estimator on the subset.</p>
</list-item>
</list>
</sec>
<sec id="sectionU003Ascoring">
<title>Scoring</title>
<p>After the estimators have been fit, a scoring step is executed on
the test set. By default, the validation estimator score function is
triggered and its results are stored. Depending on the estimator,
this often means <italic>classification accuracy</italic> for
classifiers and the <inline-formula><alternatives>
<tex-math><![CDATA[R^2]]></tex-math>
<mml:math display="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:msup><mml:mi>R</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:math></alternatives></inline-formula>
score for regressors. Besides the built-in metrics, users can
install custom metrics.</p>
<p>To install custom metrics, programmatic hooks are available. This
enables, for example, a user aggregate over the various validated
feature subsets and bootstrapped datasets. A user could compute the
average accuracy over all bootstraps, or compute various stability
metrics
(<xref alt="Nogueira et al., 2018" rid="ref-nogueira_stability_2018" ref-type="bibr">Nogueira
et al., 2018</xref>). Another example of a custom metric would be to
compare the dataset ground-truth feature importances to the
estimated importances, which information would be available when
using <italic>synthetic</italic> datasets.</p>
</sec>
</sec>
<sec id="conclusion-and-future-work">
<title>Conclusion and Future Work</title>
<p><monospace>fseval</monospace> is a comprehensive and feature rich
Python library for benchmarking Feature Ranking and Feature Selection
algorithms. It allows authors to focus on their empirical research
instead of having to implement another benchmarking pipeline -
exploiting <monospace>fseval</monospace>’s support for parallel
processing, distributed computing and export possibilities.
<monospace>fseval</monospace> is open source and published on the PyPi
platform. Next steps are to include more built-in dataset adapters,
metrics and export possibilities.</p>
</sec>
</body>
<back>
<ref-list>
<ref id="ref-zhao_searching_2007">
<element-citation publication-type="paper-conference">
<person-group person-group-type="author">
<name><surname>Zhao</surname><given-names>Zheng</given-names></name>
<name><surname>Liu</surname><given-names>Huan</given-names></name>
</person-group>
<article-title>Searching for interacting features</article-title>
<source>Proceedings of the 20th international joint conference on Artifical intelligence</source>
<publisher-name>Morgan Kaufmann Publishers Inc.</publisher-name>
<publisher-loc>San Francisco, CA, USA</publisher-loc>
<year iso-8601-date="2007-01">2007</year><month>01</month>
<date-in-citation content-type="access-date"><year iso-8601-date="2020-10-21">2020</year><month>10</month><day>21</day></date-in-citation>
<uri>https://www.semanticscholar.org/paper/Searching-for-Interacting-Features-Zhao-Liu/d2debe138a9b67d838b11d622651383322934aee</uri>
<fpage>1156</fpage>
<lpage>1161</lpage>
</element-citation>
</ref>
<ref id="ref-kira_feature_1992">
<element-citation publication-type="paper-conference">
<person-group person-group-type="author">
<name><surname>Kira</surname><given-names>K.</given-names></name>
<name><surname>Rendell</surname><given-names>L.</given-names></name>
</person-group>
<article-title>The feature selection problem: Traditional methods and a new algorithm</article-title>
<source>AAAI</source>
<year iso-8601-date="1992">1992</year>
</element-citation>
</ref>
<ref id="ref-bennasar_feature_2015">
<element-citation publication-type="article-journal">
<person-group person-group-type="author">
<name><surname>Bennasar</surname><given-names>Mohamed</given-names></name>
<name><surname>Hicks</surname><given-names>Yulia</given-names></name>
<name><surname>Setchi</surname><given-names>Rossitza</given-names></name>
</person-group>
<article-title>Feature selection using Joint Mutual Information Maximisation</article-title>
<source>Expert Systems with Applications</source>
<year iso-8601-date="2015-12">2015</year><month>12</month>
<date-in-citation content-type="access-date"><year iso-8601-date="2020-09-23">2020</year><month>09</month><day>23</day></date-in-citation>
<volume>42</volume>
<issue>22</issue>
<issn>0957-4174</issn>
<uri>http://www.sciencedirect.com/science/article/pii/S0957417415004674</uri>
<pub-id pub-id-type="doi">10.1016/j.eswa.2015.07.007</pub-id>
<fpage>8520</fpage>
<lpage>8532</lpage>
</element-citation>
</ref>
<ref id="ref-roffo_infinite_2015">
<element-citation publication-type="paper-conference">
<person-group person-group-type="author">
<name><surname>Roffo</surname><given-names>Giorgio</given-names></name>
<name><surname>Melzi</surname><given-names>Simone</given-names></name>
<name><surname>Cristani</surname><given-names>Marco</given-names></name>
</person-group>
<article-title>Infinite feature selection</article-title>
<source>2015 IEEE International Conference on Computer Vision (ICCV)</source>
<publisher-name>IEEE</publisher-name>
<publisher-loc>Santiago, Chile</publisher-loc>
<year iso-8601-date="2015-12">2015</year><month>12</month>
<date-in-citation content-type="access-date"><year iso-8601-date="2020-09-23">2020</year><month>09</month><day>23</day></date-in-citation>
<isbn>978-1-4673-8391-2</isbn>
<uri>http://ieeexplore.ieee.org/document/7410835/</uri>
<pub-id pub-id-type="doi">10.1109/ICCV.2015.478</pub-id>
<fpage>4202</fpage>
<lpage>4210</lpage>
</element-citation>
</ref>
<ref id="ref-almuallim_learning_1991">
<element-citation publication-type="paper-conference">
<person-group person-group-type="author">
<name><surname>Almuallim</surname><given-names>Hussein</given-names></name>
<name><surname>Dietterich</surname><given-names>Thomas G.</given-names></name>
</person-group>
<article-title>Learning with many irrelevant features</article-title>
<source>In Proceedings of the Ninth National Conference on Artificial Intelligence</source>
<publisher-name>AAAI Press</publisher-name>
<year iso-8601-date="1991">1991</year>
<fpage>547</fpage>
<lpage>552</lpage>
</element-citation>
</ref>
<ref id="ref-gu_generalized_2012">
<element-citation publication-type="article-journal">
<person-group person-group-type="author">
<name><surname>Gu</surname><given-names>Quanquan</given-names></name>
<name><surname>Li</surname><given-names>Zhenhui</given-names></name>
<name><surname>Han</surname><given-names>Jiawei</given-names></name>
</person-group>
<article-title>Generalized Fisher score for feature selection</article-title>
<source>arXiv:1202.3725 [cs, stat]</source>
<year iso-8601-date="2012-02">2012</year><month>02</month>
<date-in-citation content-type="access-date"><year iso-8601-date="2021-04-20">2021</year><month>04</month><day>20</day></date-in-citation>
<uri>http://arxiv.org/abs/1202.3725</uri>
</element-citation>
</ref>
<ref id="ref-bradley_feature_1998">
<element-citation publication-type="paper-conference">
<person-group person-group-type="author">
<name><surname>Bradley</surname><given-names>P. S.</given-names></name>
<name><surname>Mangasarian</surname><given-names>O. L.</given-names></name>
</person-group>
<article-title>Feature selection via concave minimization and support vector machines</article-title>
<source>Machine Learning Proceedings of the Fifteenth International Conference(ICML ’98)</source>
<publisher-name>Morgan Kaufmann</publisher-name>
<year iso-8601-date="1998">1998</year>
<fpage>82</fpage>
<lpage>90</lpage>
</element-citation>
</ref>
<ref id="ref-wojtas_feature_2020">
<element-citation publication-type="article-journal">
<person-group person-group-type="author">
<name><surname>Wojtas</surname><given-names>Maksymilian</given-names></name>
<name><surname>Chen</surname><given-names>Ke</given-names></name>
</person-group>
<article-title>Feature importance ranking for deep learning</article-title>
<source>arXiv:2010.08973 [cs]</source>
<year iso-8601-date="2020-10">2020</year><month>10</month>
<date-in-citation content-type="access-date"><year iso-8601-date="2021-04-24">2021</year><month>04</month><day>24</day></date-in-citation>
<uri>http://arxiv.org/abs/2010.08973</uri>
</element-citation>
</ref>
<ref id="ref-peng_feature_2005">
<element-citation publication-type="article-journal">
<person-group person-group-type="author">
<name><surname>Peng</surname><given-names>Hanchuan</given-names></name>
<name><surname>Long</surname><given-names>Fuhui</given-names></name>
<name><surname>Ding</surname><given-names>C.</given-names></name>
</person-group>
<article-title>Feature selection based on mutual information criteria of max-dependency, max-relevance, and min-redundancy</article-title>
<source>IEEE Transactions on Pattern Analysis and Machine Intelligence</source>
<year iso-8601-date="2005-08">2005</year><month>08</month>
<volume>27</volume>
<issue>8</issue>
<issn>1939-3539</issn>
<pub-id pub-id-type="doi">10.1109/TPAMI.2005.159</pub-id>
<fpage>1226</fpage>
<lpage>1238</lpage>
</element-citation>
</ref>
<ref id="ref-guyon_introduction_2003">
<element-citation publication-type="article-journal">
<person-group person-group-type="author">
<name><surname>Guyon</surname><given-names>Isabelle</given-names></name>
<name><surname>Elisseeff</surname><given-names>André</given-names></name>
</person-group>
<article-title>An introduction to variable and feature selection</article-title>
<source>The Journal of Machine Learning Research</source>
<year iso-8601-date="2003-03">2003</year><month>03</month>
<volume>3</volume>
<issue>null</issue>
<issn>1532-4435</issn>
<fpage>1157</fpage>
<lpage>1182</lpage>
</element-citation>
</ref>
<ref id="ref-bayer_sqlalchemy_2012">
<element-citation publication-type="chapter">
<person-group person-group-type="author">
<name><surname>Bayer</surname><given-names>Michael</given-names></name>
</person-group>
<article-title>SQLAlchemy</article-title>
<source>The Architecture of Open Source Applications Volume II: Structure, Scale, and a Few More Fearless Hacks</source>
<person-group person-group-type="editor">
<name><surname>Brown</surname><given-names>Amy</given-names></name>
<name><surname>Wilson</surname><given-names>Greg</given-names></name>
</person-group>
<publisher-name>aosabook.org</publisher-name>
<year iso-8601-date="2012">2012</year>
<uri>http://aosabook.org/en/sqlalchemy.html</uri>
</element-citation>
</ref>
<ref id="ref-biewald_experiment_2020">
<element-citation>
<person-group person-group-type="author">
<name><surname>Biewald</surname><given-names>Lukas</given-names></name>
</person-group>
<article-title>Experiment tracking with weights and biases</article-title>
<year iso-8601-date="2020">2020</year>
<uri>https://www.wandb.com/</uri>
</element-citation>
</ref>
<ref id="ref-yoo_slurm_2003">
<element-citation publication-type="paper-conference">
<person-group person-group-type="author">
<name><surname>Yoo</surname><given-names>Andy B.</given-names></name>
<name><surname>Jette</surname><given-names>Morris A.</given-names></name>
<name><surname>Grondona</surname><given-names>Mark</given-names></name>
</person-group>
<article-title>SLURM: Simple Linux utility for resource management</article-title>
<source>Job Scheduling Strategies for Parallel Processing</source>
<person-group person-group-type="editor">
<name><surname>Feitelson</surname><given-names>Dror</given-names></name>
<name><surname>Rudolph</surname><given-names>Larry</given-names></name>
<name><surname>Schwiegelshohn</surname><given-names>Uwe</given-names></name>
</person-group>
<publisher-name>Springer</publisher-name>
<publisher-loc>Berlin, Heidelberg</publisher-loc>
<year iso-8601-date="2003">2003</year>
<isbn>978-3-540-39727-4</isbn>
<pub-id pub-id-type="doi">10.1007/10968987_3</pub-id>
<fpage>44</fpage>
<lpage>60</lpage>
</element-citation>
</ref>
<ref id="ref-yadan_hydra_2019">
<element-citation>
<person-group person-group-type="author">
<name><surname>Yadan</surname><given-names>Omry</given-names></name>
</person-group>
<article-title>Hydra - A framework for elegantly configuring complex applications</article-title>
<year iso-8601-date="2019">2019</year>
<uri>https://github.com/facebookresearch/hydra</uri>
</element-citation>
</ref>
<ref id="ref-vanschoren_openml_2013">
<element-citation publication-type="article-journal">
<person-group person-group-type="author">
<name><surname>Vanschoren</surname><given-names>Joaquin</given-names></name>
<name><surname>Rijn</surname><given-names>Jan N. van</given-names></name>
<name><surname>Bischl</surname><given-names>Bernd</given-names></name>
<name><surname>Torgo</surname><given-names>Luis</given-names></name>
</person-group>
<article-title>OpenML: Networked science in machine learning</article-title>
<source>SIGKDD Explorations</source>
<year iso-8601-date="2013">2013</year>
<volume>15</volume>
<issue>2</issue>
<uri>http://doi.acm.org/10.1145/2641190.2641198</uri>
<pub-id pub-id-type="doi">10.1145/2641190.2641198</pub-id>
<fpage>49</fpage>
<lpage>60</lpage>
</element-citation>
</ref>
<ref id="ref-nogueira_stability_2018">
<element-citation publication-type="article-journal">
<person-group person-group-type="author">
<name><surname>Nogueira</surname><given-names>Sarah</given-names></name>
<name><surname>Sechidis</surname><given-names>Konstantinos</given-names></name>
<name><surname>Brown</surname><given-names>Gavin</given-names></name>
</person-group>
<article-title>On the stability of feature selection algorithms</article-title>
<source>Journal of Machine Learning Research</source>
<year iso-8601-date="2018">2018</year>
<date-in-citation content-type="access-date"><year iso-8601-date="2022-01-29">2022</year><month>01</month><day>29</day></date-in-citation>
<volume>18</volume>
<issue>174</issue>
<issn>1533-7928</issn>
<uri>http://jmlr.org/papers/v18/17-514.html</uri>
<fpage>1</fpage>
<lpage>54</lpage>
</element-citation>
</ref>
<ref id="ref-reis_featsel_2017">
<element-citation publication-type="article-journal">
<person-group person-group-type="author">
<name><surname>Reis</surname><given-names>Marcelo S.</given-names></name>
<name><surname>Estrela</surname><given-names>Gustavo</given-names></name>
<name><surname>Ferreira</surname><given-names>Carlos Eduardo</given-names></name>
<name><surname>Barrera</surname><given-names>Junior</given-names></name>
</person-group>
<article-title>Featsel: A framework for benchmarking of feature selection algorithms and cost functions</article-title>
<source>SoftwareX</source>
<year iso-8601-date="2017">2017</year>
<volume>6</volume>
<issn>2352-7110</issn>
<uri>https://www.sciencedirect.com/science/article/pii/S2352711017300286</uri>
<pub-id pub-id-type="doi">10.1016/j.softx.2017.07.005</pub-id>
<fpage>193</fpage>
<lpage>197</lpage>
</element-citation>
</ref>
<ref id="ref-cilia2019experimental">
<element-citation publication-type="article-journal">
<person-group person-group-type="author">
<name><surname>Cilia</surname><given-names>Nicole Dalia</given-names></name>
<name><surname>De Stefano</surname><given-names>Claudio</given-names></name>
<name><surname>Fontanella</surname><given-names>Francesco</given-names></name>
<name><surname>Raimondo</surname><given-names>Stefano</given-names></name>
<name><surname>Scotto di Freca</surname><given-names>Alessandra</given-names></name>
</person-group>
<article-title>An experimental comparison of feature-selection and classification methods for microarray datasets</article-title>
<source>Information</source>
<year iso-8601-date="2019">2019</year>
<volume>10</volume>
<issue>3</issue>
<issn>2078-2489</issn>
<uri>https://www.mdpi.com/2078-2489/10/3/109</uri>
<pub-id pub-id-type="doi">10.3390/info10030109</pub-id>
</element-citation>
</ref>
<ref id="ref-sun2019comparison">
<element-citation publication-type="article-journal">
<person-group person-group-type="author">
<name><surname>Sun</surname><given-names>Pan</given-names></name>
<name><surname>Wang</surname><given-names>Defeng</given-names></name>
<name><surname>Mok</surname><given-names>Vincent Ct</given-names></name>
<name><surname>Shi</surname><given-names>Lin</given-names></name>
</person-group>
<article-title>Comparison of feature selection methods and machine learning classifiers for radiomics analysis in glioma grading</article-title>
<source>IEEE Access</source>
<year iso-8601-date="2019">2019</year>
<volume>7</volume>
<issue></issue>
<pub-id pub-id-type="doi">10.1109/ACCESS.2019.2928975</pub-id>
<fpage>102010</fpage>
<lpage>102020</lpage>
</element-citation>
</ref>
<ref id="ref-tohka2016comparison">
<element-citation publication-type="article-journal">
<person-group person-group-type="author">
<name><surname>Tohka</surname><given-names>Jussi</given-names></name>
<name><surname>Moradi</surname><given-names>Elaheh</given-names></name>
<name><surname>Huttunen</surname><given-names>Heikki</given-names></name>
</person-group>
<article-title>Comparison of feature selection techniques in machine learning for anatomical brain MRI in dementia</article-title>
<source>Neuroinformatics</source>
<publisher-name>Springer</publisher-name>
<year iso-8601-date="2016">2016</year>
<volume>14</volume>
<issue>3</issue>
<pub-id pub-id-type="doi">10.1007/s12021-015-9292-3</pub-id>
<fpage>279</fpage>
<lpage>296</lpage>
</element-citation>
</ref>
<ref id="ref-liu2017multi">
<element-citation publication-type="article-journal">
<person-group person-group-type="author">
<name><surname>Liu</surname><given-names>Yang</given-names></name>
<name><surname>Bi</surname><given-names>Jian-Wu</given-names></name>
<name><surname>Fan</surname><given-names>Zhi-Ping</given-names></name>
</person-group>
<article-title>Multi-class sentiment classification: The experimental comparisons of feature selection and machine learning algorithms</article-title>
<source>Expert Systems with Applications</source>
<year iso-8601-date="2017">2017</year>
<volume>80</volume>
<issn>0957-4174</issn>
<uri>https://www.sciencedirect.com/science/article/pii/S0957417417301951</uri>
<pub-id pub-id-type="doi">10.1016/j.eswa.2017.03.042</pub-id>
<fpage>323</fpage>
<lpage>339</lpage>
</element-citation>
</ref>
<ref id="ref-kou2020evaluation">
<element-citation publication-type="article-journal">
<person-group person-group-type="author">
<name><surname>Kou</surname><given-names>Gang</given-names></name>
<name><surname>Yang</surname><given-names>Pei</given-names></name>
<name><surname>Peng</surname><given-names>Yi</given-names></name>
<name><surname>Xiao</surname><given-names>Feng</given-names></name>
<name><surname>Chen</surname><given-names>Yang</given-names></name>
<name><surname>Alsaadi</surname><given-names>Fawaz E.</given-names></name>
</person-group>
<article-title>Evaluation of feature selection methods for text classification with small datasets using multiple criteria decision-making methods</article-title>
<source>Applied Soft Computing</source>
<year iso-8601-date="2020">2020</year>
<volume>86</volume>
<issn>1568-4946</issn>
<uri>https://www.sciencedirect.com/science/article/pii/S1568494619306179</uri>
<pub-id pub-id-type="doi">10.1016/j.asoc.2019.105836</pub-id>
<fpage>105836</fpage>
<lpage></lpage>
</element-citation>
</ref>
<ref id="ref-venkatesh2019review">
<element-citation publication-type="article-journal">
<person-group person-group-type="author">
<name><surname>Venkatesh</surname><given-names>B.</given-names></name>
<name><surname>Anuradha</surname><given-names>J.</given-names></name>
</person-group>
<article-title>A review of feature selection and its methods</article-title>
<source>Cybernetics and Information Technologies</source>
<year iso-8601-date="2019">2019</year>
<volume>19</volume>
<issue>1</issue>
<uri>https://doi.org/10.2478/cait-2019-0001</uri>
<pub-id pub-id-type="doi">10.2478/cait-2019-0001</pub-id>
<fpage>3</fpage>
<lpage>26</lpage>
</element-citation>
</ref>
</ref-list>
</back>
</article>