Repository navigation
Expand file tree
/
Copy pathreleases.html
More file actions
1239 lines (1188 loc) · 155 KB
/
Copy pathreleases.html
File metadata and controls
1239 lines (1188 loc) · 155 KB
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
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html class="writer-html5" lang="en" data-content_root="./">
<head>
<meta charset="utf-8" /><meta name="viewport" content="width=device-width, initial-scale=1" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>Releases — POT Python Optimal Transport 0.9.8.dev documentation</title>
<link rel="stylesheet" type="text/css" href="_static/pygments.css?v=b86133f3" />
<link rel="stylesheet" type="text/css" href="_static/css/theme.css?v=9edc463e" />
<link rel="stylesheet" type="text/css" href="_static/sg_gallery.css?v=c83e6b79" />
<link rel="stylesheet" type="text/css" href="_static/sg_gallery-binder.css?v=f4aeca0c" />
<link rel="stylesheet" type="text/css" href="_static/sg_gallery-dataframe.css?v=2082cf3c" />
<link rel="stylesheet" type="text/css" href="_static/sg_gallery-rendered-html.css?v=1277b6f3" />
<script src="_static/jquery.js?v=5d32c60e"></script>
<script src="_static/_sphinx_javascript_frameworks_compat.js?v=2cd50e6c"></script>
<script src="_static/documentation_options.js?v=dd0b68c4"></script>
<script src="_static/doctools.js?v=fd6eb6e6"></script>
<script src="_static/sphinx_highlight.js?v=6ffebe34"></script>
<script src="_static/sg-tags.js?v=d80f6fa4"></script>
<script src="_static/js/theme.js"></script>
<link rel="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
<link rel="next" title="Contributors" href="contributors.html" />
<link rel="prev" title="ot.weak" href="gen_modules/ot.weak.html" />
</head>
<body class="wy-body-for-nav">
<div class="wy-grid-for-nav">
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
<div class="wy-side-scroll">
<div class="wy-side-nav-search" >
<a href="index.html" class="icon icon-home">
POT Python Optimal Transport
<img src="_static/logo_dark.svg" class="logo" alt="Logo"/>
</a>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="search.html" method="get">
<input type="text" name="q" placeholder="Search docs" aria-label="Search docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div><div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="Navigation menu">
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="index.html">POT: Python Optimal Transport</a></li>
<li class="toctree-l1"><a class="reference internal" href="auto_examples/plot_quickstart_guide.html">Quickstart Guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="auto_examples/index.html">Examples gallery</a></li>
<li class="toctree-l1"><a class="reference internal" href="user_guide.html">User guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="all.html">API and modules</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">Releases</a><ul>
<li class="toctree-l2"><a class="reference internal" href="#dev">0.9.8dev</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#new-features">New features</a></li>
<li class="toctree-l3"><a class="reference internal" href="#closed-issues">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#post1">0.9.7.post1</a></li>
<li class="toctree-l2"><a class="reference internal" href="#id1">0.9.7</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#id2">New features</a></li>
<li class="toctree-l3"><a class="reference internal" href="#id3">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id4">0.9.6.post1</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#id5">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id6">0.9.6</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#id7">New features</a></li>
<li class="toctree-l3"><a class="reference internal" href="#id8">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id9">0.9.5</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#breaking-change">Breaking change</a></li>
<li class="toctree-l3"><a class="reference internal" href="#id10">New features</a></li>
<li class="toctree-l3"><a class="reference internal" href="#id11">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id12">0.9.4</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#id13">New features</a></li>
<li class="toctree-l3"><a class="reference internal" href="#id14">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id15">0.9.3</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#id16">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id17">0.9.2</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#id18">New features</a></li>
<li class="toctree-l3"><a class="reference internal" href="#id19">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id20">0.9.1</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#id21">New features</a></li>
<li class="toctree-l3"><a class="reference internal" href="#id22">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id23">0.8.2</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#id24">New features</a></li>
<li class="toctree-l3"><a class="reference internal" href="#id25">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id26">0.8.1.0</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#id27">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id28">0.8.1</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#id29">New features</a></li>
<li class="toctree-l3"><a class="reference internal" href="#id30">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id31">0.8.0</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#id32">New features</a></li>
<li class="toctree-l3"><a class="reference internal" href="#id33">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id34">0.7.0</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#features">Features</a></li>
<li class="toctree-l3"><a class="reference internal" href="#id35">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id36">0.6.0</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#id37">Features</a></li>
<li class="toctree-l3"><a class="reference internal" href="#id38">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id39">0.5.0</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#id40">Features</a></li>
<li class="toctree-l3"><a class="reference internal" href="#deprecation">Deprecation</a></li>
<li class="toctree-l3"><a class="reference internal" href="#id41">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id42">0.4</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#id43">Features</a></li>
<li class="toctree-l3"><a class="reference internal" href="#id44">Closed issues</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#id45">0.3.1</a></li>
<li class="toctree-l2"><a class="reference internal" href="#id46">0.3</a></li>
<li class="toctree-l2"><a class="reference internal" href="#v0-2">V0.2</a></li>
<li class="toctree-l2"><a class="reference internal" href="#id47">0.1.11</a></li>
<li class="toctree-l2"><a class="reference internal" href="#id48">0.1.10</a></li>
<li class="toctree-l2"><a class="reference internal" href="#id49">0.1.9</a></li>
<li class="toctree-l2"><a class="reference internal" href="#id50">0.1.7</a></li>
<li class="toctree-l2"><a class="reference internal" href="#id51">0.1.3</a></li>
<li class="toctree-l2"><a class="reference internal" href="#first-pre-release">First pre-release</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="contributors.html">Contributors</a></li>
<li class="toctree-l1"><a class="reference internal" href="contributing.html">Contributing to POT</a></li>
<li class="toctree-l1"><a class="reference internal" href="code_of_conduct.html">Code of conduct</a></li>
</ul>
</div>
</div>
</nav>
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap"><nav class="wy-nav-top" aria-label="Mobile navigation menu" >
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
<a href="index.html">POT Python Optimal Transport</a>
</nav>
<div class="wy-nav-content">
<div class="rst-content">
<div role="navigation" aria-label="Page navigation">
<ul class="wy-breadcrumbs">
<li><a href="index.html" class="icon icon-home" aria-label="Home"></a></li>
<li class="breadcrumb-item active">Releases</li>
<li class="wy-breadcrumbs-aside">
<a href="_sources/releases.rst.txt" rel="nofollow"> View page source</a>
</li>
</ul>
<hr/>
</div>
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
<div itemprop="articleBody">
<section id="releases">
<h1>Releases<a class="headerlink" href="#releases" title="Link to this heading"></a></h1>
<section id="dev">
<h2>0.9.8dev<a class="headerlink" href="#dev" title="Link to this heading"></a></h2>
<section id="new-features">
<h3>New features<a class="headerlink" href="#new-features" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">ot.dr.wda</span></code> gains <code class="docutils literal notranslate"><span class="pre">solver='torch'</span></code>, a PyTorch autodiff solver with Riemannian gradient descent, usable on installations without autograd or pymanopt, and accepting torch tensors directly (<a class="reference external" href="https://github.com/PythonOT/POT/pull/858">PR #858</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/806">Issue #806</a>)</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">ot.dr</span></code> dependencies (autograd, pymanopt, scikit-learn, torch) are now imported optionally, so importing <code class="docutils literal notranslate"><span class="pre">ot.dr</span></code> no longer requires all of them; each function raises an explicit <code class="docutils literal notranslate"><span class="pre">ImportError</span></code> naming what it needs (<a class="reference external" href="https://github.com/PythonOT/POT/pull/858">PR #858</a>)</p></li>
<li><p>Use <code class="docutils literal notranslate"><span class="pre">ot.utils.check_marginal</span></code> (and shape-tuple support in <code class="docutils literal notranslate"><span class="pre">ot.utils.unif</span></code>) to fill and validate default marginals consistently across solvers (Gromov, low-rank, stochastic, barycenter, factored) (<a class="reference external" href="https://github.com/PythonOT/POT/pull/856">PR #856</a>)</p></li>
<li><p>Add stereographic spherical sliced Wasserstein distance in <code class="docutils literal notranslate"><span class="pre">ot.sliced.stereographic_sliced_wasserstein_sphere</span></code>, with its rotationally invariant extension (<a class="reference external" href="https://github.com/PythonOT/POT/pull/836">PR #836</a>)</p></li>
<li><p>Add Quasi-Monte Carlo sliced Wasserstein sampling (QSW/RQSW) via generalized
spiral points, selectable with <code class="docutils literal notranslate"><span class="pre">sampling_slices</span></code> in <code class="docutils literal notranslate"><span class="pre">sliced_wasserstein_distance</span></code>,
as described in [95] (<a class="reference external" href="https://github.com/PythonOT/POT/pull/838">PR #838</a>)</p></li>
<li><p>Add UnifOrtho sliced Wasserstein sampling via independent orthogonal
blocks, selectable with <code class="docutils literal notranslate"><span class="pre">sampling_slices</span></code> in <code class="docutils literal notranslate"><span class="pre">sliced_wasserstein_distance</span></code>,
as described in [97] (<a class="reference external" href="https://github.com/PythonOT/POT/pull/853">PR #853</a>)</p></li>
</ul>
</section>
<section id="closed-issues">
<h3>Closed issues<a class="headerlink" href="#closed-issues" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Fix internal arrays created on the torch default device instead of the input device in <code class="docutils literal notranslate"><span class="pre">ot.partial.partial_wasserstein</span></code>, <code class="docutils literal notranslate"><span class="pre">ot.wasserstein_circle</span></code> (and <code class="docutils literal notranslate"><span class="pre">ot.sliced_wasserstein_sphere</span></code>), the high-dimensional Bures-Wasserstein functions in <code class="docutils literal notranslate"><span class="pre">ot.gaussian</span></code>, <code class="docutils literal notranslate"><span class="pre">ot.gmm</span></code>, <code class="docutils literal notranslate"><span class="pre">ot.lowrank_sinkhorn</span></code> with <code class="docutils literal notranslate"><span class="pre">init="deterministic"</span></code>, the semi-relaxed (F)GW barycenters and <code class="docutils literal notranslate"><span class="pre">ot.utils.projection_sparse_simplex</span></code>, which made them fail on GPU inputs. Add a <code class="docutils literal notranslate"><span class="pre">meta_default_device</span></code> test fixture to catch such allocations on CPU (<a class="reference external" href="https://github.com/PythonOT/POT/pull/883">PR #883</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/852">Issue #852</a>)</p></li>
<li><p>Allow <code class="docutils literal notranslate"><span class="pre">NumpyBackend.seed</span></code> to adopt an existing <code class="docutils literal notranslate"><span class="pre">np.random.RandomState</span></code> instance and remove NumPy-specific random sampling paths in sliced utilities (<a class="reference external" href="https://github.com/PythonOT/POT/pull/849">PR #849</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/848">Issue #848</a>)</p></li>
<li><p>Remove a leftover debug <code class="docutils literal notranslate"><span class="pre">print</span></code> from <code class="docutils literal notranslate"><span class="pre">ot.utils.projection_sparse_simplex</span></code> with <code class="docutils literal notranslate"><span class="pre">axis=1</span></code>, and make the <code class="docutils literal notranslate"><span class="pre">ot.datasets.make_gauss_hd</span></code> docstring a raw string so importing <code class="docutils literal notranslate"><span class="pre">ot</span></code> no longer emits a <code class="docutils literal notranslate"><span class="pre">SyntaxWarning</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/860">PR #860</a>)</p></li>
<li><p>Fix the value reported by the entropic unbalanced OT solvers (<code class="docutils literal notranslate"><span class="pre">ot.unbalanced.sinkhorn_unbalanced</span></code>, <code class="docutils literal notranslate"><span class="pre">sinkhorn_unbalanced2</span></code>, <code class="docutils literal notranslate"><span class="pre">sinkhorn_knopp_unbalanced</span></code>, <code class="docutils literal notranslate"><span class="pre">sinkhorn_stabilized_unbalanced</span></code> and <code class="docutils literal notranslate"><span class="pre">sinkhorn_unbalanced_translation_invariant</span></code>, also exposed as <code class="docutils literal notranslate"><span class="pre">ot.solve(...,</span> <span class="pre">reg=...,</span> <span class="pre">unbalanced=...).value</span></code>): every penalization is now evaluated with the generalized KL divergence, and with <code class="docutils literal notranslate"><span class="pre">reg_type="entropy"</span></code> the constant <code class="docutils literal notranslate"><span class="pre">dim_a</span> <span class="pre">*</span> <span class="pre">dim_b</span></code> is removed so that the reported value matches the documented negative entropy regularizer. The transport plans are unchanged. <code class="docutils literal notranslate"><span class="pre">reg_type</span></code> is now handled case insensitively by all the methods and an unknown value raises a <code class="docutils literal notranslate"><span class="pre">ValueError</span></code> instead of silently falling back to <code class="docutils literal notranslate"><span class="pre">"kl"</span></code>. The docstrings now state the generalized KL convention, the regularizer <code class="docutils literal notranslate"><span class="pre">Omega_reg_type(gamma,</span> <span class="pre">c)</span></code> with its two cases, the keys of the <code class="docutils literal notranslate"><span class="pre">log</span></code> dictionary and the multi-histogram behaviour (<a class="reference external" href="https://github.com/PythonOT/POT/pull/874">PR #874</a>)</p></li>
<li><p>Fix <code class="docutils literal notranslate"><span class="pre">ot.dist</span></code> ignoring the weights <code class="docutils literal notranslate"><span class="pre">w</span></code> for <code class="docutils literal notranslate"><span class="pre">metric="cityblock"</span></code>, which returned the unweighted distance although the weights are documented for this metric (<a class="reference external" href="https://github.com/PythonOT/POT/pull/859">PR #859</a>)</p></li>
<li><p>Fix swapped arguments to <code class="docutils literal notranslate"><span class="pre">div_to_product</span></code> in <code class="docutils literal notranslate"><span class="pre">ot.gromov.fused_unbalanced_across_spaces_cost</span></code>: with <code class="docutils literal notranslate"><span class="pre">reg_type="independent"</span></code> (UCOOT) the entropic terms used the plan marginals as the reference measures and vice versa (<a class="reference external" href="https://github.com/PythonOT/POT/pull/855">PR #855</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/854">Issue #854</a>)</p></li>
<li><p>Fix device placement in <code class="docutils literal notranslate"><span class="pre">ot.batch.bregman_projection_batch</span></code> so <code class="docutils literal notranslate"><span class="pre">ot.solve_batch(...,</span> <span class="pre">method="sinkhorn")</span></code> no longer crashes on GPU when the torch default device is CPU (<a class="reference external" href="https://github.com/PythonOT/POT/pull/851">PR #851</a>)</p></li>
<li><p>Preserve input dtype and device for expected sliced plans, avoid materializing dense distance matrices for sparse plans, and fix weighted sparse-distance ordering (<a class="reference external" href="https://github.com/PythonOT/POT/pull/846">PR #846</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/845">Issue #845</a>)</p></li>
<li><p>Build the CUDA generator and the CUDA entries of <code class="docutils literal notranslate"><span class="pre">TorchBackend.__type_list__</span></code> lazily, so that using POT with CPU-only torch tensors no longer initialises a CUDA context and claims device memory (<a class="reference external" href="https://github.com/PythonOT/POT/pull/847">PR #847</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/612">Issue #612</a>)</p></li>
<li><p>Fix the sign issue in updates of the previous transport plan in <code class="docutils literal notranslate"><span class="pre">ot.batch.proximal_bregman_log_plan_batch</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/issues/842">Issue #842</a>)</p></li>
<li><p>Load triton before TensorFlow in <code class="docutils literal notranslate"><span class="pre">ot.backend</span></code> so that building a torch optimizer no longer segfaults the interpreter, and remove the <code class="docutils literal notranslate"><span class="pre">torch<2.12</span></code> pin from the doctest and documentation requirements (<a class="reference external" href="https://github.com/PythonOT/POT/pull/839">PR #839</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/816">Issue #816</a>)</p></li>
<li><p>Fix mean centering in <code class="docutils literal notranslate"><span class="pre">ot.dr.fda</span></code> and <code class="docutils literal notranslate"><span class="pre">ot.dr.wda</span></code>: <code class="docutils literal notranslate"><span class="pre">np.mean(X)</span></code> returned a scalar instead of the per-feature mean, so <code class="docutils literal notranslate"><span class="pre">proj</span></code> did not center the data as documented. In <code class="docutils literal notranslate"><span class="pre">ot.dr.fda</span></code> the same pattern in the class means made the between-class scatter matrix independent of which features separate the classes, and FDA returned a non-discriminant direction (<a class="reference external" href="https://github.com/PythonOT/POT/pull/840">PR #840</a>)</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">ot.dr.fda</span></code> and <code class="docutils literal notranslate"><span class="pre">ot.dr.wda</span></code> no longer modify the input array <code class="docutils literal notranslate"><span class="pre">X</span></code> in place (<a class="reference external" href="https://github.com/PythonOT/POT/pull/840">PR #840</a>)</p></li>
<li><p>Fix <code class="docutils literal notranslate"><span class="pre">UnbalancedSinkhornTransport</span></code> <code class="docutils literal notranslate"><span class="pre">transform</span></code> failing with <code class="docutils literal notranslate"><span class="pre">AttributeError:</span> <span class="pre">'NoneType'</span> <span class="pre">object</span> <span class="pre">has</span> <span class="pre">no</span> <span class="pre">attribute</span> <span class="pre">'array_equal'</span></code> when <code class="docutils literal notranslate"><span class="pre">fit</span></code> was called with missing parameters (<a class="reference external" href="https://github.com/PythonOT/POT/pull/837">PR #837</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/650">Issue #650</a>)</p></li>
<li><p>Fix quantized (F)GW solvers that ordered OT based on clusters and not initial node ordering (<a class="reference external" href="https://github.com/PythonOT/POT/pull/857">PR #857</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/786">Issue #786</a>)</p></li>
<li><p>Update CircleCI config to use <code class="docutils literal notranslate"><span class="pre">version:</span> <span class="pre">2.1</span></code> instead of the deprecated <code class="docutils literal notranslate"><span class="pre">version:</span> <span class="pre">2.</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/882">PR #882</a>)</p></li>
<li><p>Fix <code class="docutils literal notranslate"><span class="pre">ot.gromov.fused_unbalanced_across_spaces_divergence</span></code> crashing with a <code class="docutils literal notranslate"><span class="pre">TypeError</span></code> when only one of the sample or feature dual pairs was passed in <code class="docutils literal notranslate"><span class="pre">init_duals</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/888">PR #888</a>)</p></li>
<li><p>Raise errors instead of printing a message and returning <code class="docutils literal notranslate"><span class="pre">None</span></code> or dummy values in <code class="docutils literal notranslate"><span class="pre">ot.datasets.make_data_classif</span></code> (unknown dataset), <code class="docutils literal notranslate"><span class="pre">ot.stochastic.solve_semi_dual_entropic</span></code> (unknown method) and <code class="docutils literal notranslate"><span class="pre">ot.mapping.nearest_brenier_potential_*</span></code> (missing CVXPY), and emit warnings instead of prints in <code class="docutils literal notranslate"><span class="pre">ot.gaussian.bures_barycenter_fixpoint</span></code>, <code class="docutils literal notranslate"><span class="pre">ot.regpath</span></code> and <code class="docutils literal notranslate"><span class="pre">ot.partial.entropic_partial_wasserstein</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/888">PR #888</a>)</p></li>
<li><p>Fix docstring errors (wrong or duplicated parameters, typos) and fix the file patterns of the ruff hooks in <code class="docutils literal notranslate"><span class="pre">.pre-commit-config.yaml</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/888">PR #888</a>)</p></li>
</ul>
</section>
</section>
<section id="post1">
<h2>0.9.7.post1<a class="headerlink" href="#post1" title="Link to this heading"></a></h2>
<p>This release is identical to 0.9.7 but will allow the upload of a source distribution to PyPI and release on conda-forge (that requires a source distribution).</p>
</section>
<section id="id1">
<h2>0.9.7<a class="headerlink" href="#id1" title="Link to this heading"></a></h2>
<p>This release contains several bug fixes and performance improvements, as well as updates to the documentation and examples and tests. We provide below a summary of the main new features and closed issues.</p>
<p><strong>New exact OT solvers.</strong> This new release adds many new variants and updates for the exact OT solver. First a new lazy exact OT solver that re-computes distances on-the-fly from coordinates, reducing memory usage from O(n×m) to O(n+m) ( available in <a class="reference external" href="https://pythonot.github.io/all.html#ot.solve_sample"><code class="docutils literal notranslate"><span class="pre">ot.solve_sample</span></code></a> with <code class="docutils literal notranslate"><span class="pre">lazy=True)</span></code>). Another major feature is the addition of a warmstart feature for the EMD solver, allowing users to provide initial potentials to speed up convergence (with <code class="docutils literal notranslate"><span class="pre">init_potentials</span></code> in <a class="reference external" href="https://pythonot.github.io/all.html#ot.solve"><code class="docutils literal notranslate"><span class="pre">ot.solve</span></code></a> and <a class="reference external" href="https://pythonot.github.io/all.html#ot.solve_sample"><code class="docutils literal notranslate"><span class="pre">ot.solve_sample</span></code></a>). Finally the release also include a sparse solver and <a class="reference external" href="https://pythonot.github.io/all.html#ot.solve"><code class="docutils literal notranslate"><span class="pre">ot.solve</span></code></a> now accept sparse cost matrices and provides an OT plan whose support is included in the support of the cost matrix. All implementations are backend-agnostic and preserve gradient computation for automatic differentiation (but are solved on CPU).</p>
<p>The computational time for different solvers (on a laptop CPU) are shown below:</p>
<table class="docutils align-default">
<thead>
<tr class="row-odd"><th class="head"><p>n</p></th>
<th class="head text-center"><p>100</p></th>
<th class="head text-center"><p>500</p></th>
<th class="head text-center"><p>1000</p></th>
<th class="head text-center"><p>5000</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>Solve</p></td>
<td class="text-center"><p>7.259e-04</p></td>
<td class="text-center"><p>1.501e-02</p></td>
<td class="text-center"><p>1.055e-01</p></td>
<td class="text-center"><p>2.343e+00</p></td>
</tr>
<tr class="row-odd"><td><p>Solve Warm Start</p></td>
<td class="text-center"><p>6.970e-04</p></td>
<td class="text-center"><p>5.413e-03</p></td>
<td class="text-center"><p>2.291e-02</p></td>
<td class="text-center"><p>6.036e-01</p></td>
</tr>
<tr class="row-even"><td><p>Solve Lazy</p></td>
<td class="text-center"><p>1.280e-03</p></td>
<td class="text-center"><p>3.098e-02</p></td>
<td class="text-center"><p>1.421e-01</p></td>
<td class="text-center"><p>3.377e+00</p></td>
</tr>
<tr class="row-odd"><td><p>Solve Sparse 10%</p></td>
<td class="text-center"><p>3.453e-04</p></td>
<td class="text-center"><p>6.716e-04</p></td>
<td class="text-center"><p>1.535e-03</p></td>
<td class="text-center"><p>3.534e-02</p></td>
</tr>
</tbody>
</table>
<p><strong>BSPOT solver.</strong> A new solver relying on <a class="reference external" href="https://pythonot.github.io/master/auto_examples/plot_bsp_ot.html">Binary Space Partitioning (BSP)</a> has been added to compute sparse transport plans between discrete measures in loglinear time which allows for very large problems to be solved. The release also includes a new <a class="reference external" href="https://pythonot.github.io/master/gen_modules/ot.unbalanced.html#ot.unbalanced.uot_1d"><code class="docutils literal notranslate"><span class="pre">ot.unbalanced.uot_1d</span></code></a> solver with a Frank-Wolfe solver for unbalanced optimal transport in 1D.</p>
<p><strong>Sliced OT pans</strong> Finally we now have a sliced OT plan solver that can be used to compute sliced transport plans (<a class="reference external" href="https://pythonot.github.io/master/gen_modules/ot.sliced.html#ot.sliced.min_sliced_transport_plan">min-pivot sliced</a> and <a class="reference external" href="https://pythonot.github.io/master/gen_modules/ot.sliced.html#ot.sliced.expected_sliced_plan">expected sliced</a>) between two measures.</p>
<p><strong>OT between dynamical systems.</strong> A novel <a class="reference external" href="https://pythonot.github.io/master/auto_examples/others/plot_sgot.html">Spectral-Grassmann Wasserstein metric for operator representations of dynamical systems</a> has been implemented in <code class="docutils literal notranslate"><span class="pre">ot.sgot</span></code>.</p>
<p><strong>Unified API for barycenter solvers in <code class="docutils literal notranslate"><span class="pre">ot.solve_bary_sample</span></code>.</strong> A new free support solver for barycenter solvers has been added in <a class="reference external" href="https://pythonot.github.io/master/all.html#ot.solve_bary_sample"><code class="docutils literal notranslate"><span class="pre">ot.solve_bary_sample</span></code></a>. You can see <a class="reference external" href="https://pythonot.github.io/master/auto_examples/barycenters/plot_solve_barycenter_variants.html">examples of how to use it here</a>.</p>
<p><strong>OT between high dimensional Gaussian distributions.</strong> New methods to compute the linear transport map and the related 2-Wasserstein distance between high-dimensional (HD) Gaussian distributions have been added in <a class="reference external" href="https://pythonot.github.io/master/gen_modules/ot.gaussian.html#id56"><code class="docutils literal notranslate"><span class="pre">ot.gaussian.bures_wasserstein_mapping_hd</span></code></a> and <a class="reference external" href="https://pythonot.github.io/master/gen_modules/ot.gaussian.html#ot.gaussian.bures_wasserstein_distance_hd"><code class="docutils literal notranslate"><span class="pre">ot.gaussian.bures_wasserstein_distance_hd</span></code></a>, respectively with empirical versions that estimate the distance from sample.</p>
<p><strong>Batched solver for exact OT.</strong> The batch implementations have also been improved and you can now solve exact OT problems in parallel on CPU or GPU using the new proximal point solver in functions <a class="reference external" href="https://pythonot.github.io/master/gen_modules/ot.batch.html#ot.batch.solve_batch"><code class="docutils literal notranslate"><span class="pre">ot.solve_batch</span></code></a> and <a class="reference external" href="https://pythonot.github.io/master/gen_modules/ot.batch.html#ot.batch.solve_sample_batch"><code class="docutils literal notranslate"><span class="pre">ot.solve_sample_batch</span></code></a> when <code class="docutils literal notranslate"><span class="pre">reg=0</span></code> or no provided. A new batch <a class="reference external" href="https://pythonot.github.io/master/gen_modules/ot.batch.html#ot.batch.loss_quadratic_batch">loss for Fused unbalanced Gromov-Wasserstein</a> is also now available in <code class="docutils literal notranslate"><span class="pre">ot.batch</span></code>.</p>
<p><strong>New methods in unified API <a class="reference external" href="https://pythonot.github.io/all.html#ot.solve_sample"><code class="docutils literal notranslate"><span class="pre">ot.solve_sample</span></code></a>.</strong> The unified API function
<a class="reference external" href="https://pythonot.github.io/all.html#ot.solve_sample"><code class="docutils literal notranslate"><span class="pre">ot.solve_sample</span></code></a> has also been updated to allows solving of specific problems such as BSP-OT, distance between high-dimensional (HD) Gaussian distributions and sliced and max-sliced distances.</p>
<p><strong>Data normalization for sliced and <a class="reference external" href="https://pythonot.github.io/all.html#ot.solve_sample"><code class="docutils literal notranslate"><span class="pre">ot.solve_sample</span></code></a> solvers.</strong> The release also includes tools for data normalization and scaling which can not be used in sliced Wasserstein distance computations. A simple normalization class <a class="reference external" href="https://pythonot.github.io/master/gen_modules/ot.utils.html#id46"><code class="docutils literal notranslate"><span class="pre">ot.utils.DataScaler</span></code></a> has been added and supports all backends for <code class="docutils literal notranslate"><span class="pre">'standard'</span></code>, <code class="docutils literal notranslate"><span class="pre">'minmax'</span></code>, and <code class="docutils literal notranslate"><span class="pre">'l2'</span></code> methods. Finally an optional <code class="docutils literal notranslate"><span class="pre">scaler</span></code> parameter has been added to <a class="reference external" href="https://pythonot.github.io/master/all.html#ot.sliced_wasserstein_distance"><code class="docutils literal notranslate"><span class="pre">ot.sliced_wasserstein_distance</span></code></a>, <a class="reference external" href="https://pythonot.github.io/master/all.html#ot.max_sliced_wasserstein_distance"><code class="docutils literal notranslate"><span class="pre">ot.max_sliced_wasserstein_distance</span></code></a> and <a class="reference external" href="https://pythonot.github.io/all.html#ot.solve_sample"><code class="docutils literal notranslate"><span class="pre">ot.solve_sample</span></code></a>.</p>
<section id="id2">
<h3>New features<a class="headerlink" href="#id2" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Refactor lazy EMD network simplex storage to avoid dense per-arc cost,
endpoint, flow, and state storage where possible, and return sparse lazy
transport plans instead of materializing dense plans internally (<a class="reference external" href="https://github.com/PythonOT/POT/pull/813">PR #813</a>)</p></li>
<li><p>Removed OpenMP implementation in Simplex since no time gains are observed #820</p></li>
<li><p>Add sliced transport plans (min-pivot sliced and expected sliced) solvers (<a class="reference external" href="https://github.com/PythonOT/POT/pull/767">PR #767</a>)</p></li>
<li><p>Add lazy EMD solver with on-the-fly distance computation from coordinates (<a class="reference external" href="https://github.com/PythonOT/POT/pull/788">PR #788</a>)</p></li>
<li><p>Add Warmstart feature to the EMD solver for existing potentials (<a class="reference external" href="https://github.com/PythonOT/POT/pull/793">PR #793</a>)</p></li>
<li><p>Add Warmstart potentials feature to the EMD solver for lazy and sparse solver (<a class="reference external" href="https://github.com/PythonOT/POT/pull/795">PR #795</a>)</p></li>
<li><p>Faster init and result retrieval for EMD solver (<a class="reference external" href="https://github.com/PythonOT/POT/pull/796">PR #796</a>)</p></li>
<li><p>Migrate backend from deprecated <code class="docutils literal notranslate"><span class="pre">scipy.sparse.coo_matrix</span></code> to modern <code class="docutils literal notranslate"><span class="pre">scipy.sparse.coo_array</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/782">PR #782</a>)</p></li>
<li><p>Geomloss function now handles both scalar and slice indices for i and j (<a class="reference external" href="https://github.com/PythonOT/POT/pull/785">PR #785</a>)</p></li>
<li><p>Add support for sparse cost matrices in EMD solver (<a class="reference external" href="https://github.com/PythonOT/POT/pull/778">PR #778</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/397">Issue #397</a>)</p></li>
<li><p>Add “BSP-OT: Sparse transport plans between discrete measures in loglinear time” (<a class="reference external" href="https://github.com/PythonOT/POT/pull/768">PR #768</a>)</p></li>
<li><p>Added UOT1D with Frank-Wolfe in <code class="docutils literal notranslate"><span class="pre">ot.unbalanced.uot_1d</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/765">PR #765</a>)</p></li>
<li><p>Add Sliced UOT and Unbalanced Sliced OT in <code class="docutils literal notranslate"><span class="pre">ot/unbalanced/_sliced.py</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/765">PR #765</a>)</p></li>
<li><p>Add <code class="docutils literal notranslate"><span class="pre">ot.utils.DataScaler</span></code> class for backend-aware joint normalization of input distributions, with sklearn-compatible <code class="docutils literal notranslate"><span class="pre">fit</span></code>/<code class="docutils literal notranslate"><span class="pre">transform</span></code>/<code class="docutils literal notranslate"><span class="pre">fit_transform</span></code> API and support for <code class="docutils literal notranslate"><span class="pre">'standard'</span></code>, <code class="docutils literal notranslate"><span class="pre">'minmax'</span></code>, and <code class="docutils literal notranslate"><span class="pre">'l2'</span></code> methods (<a class="reference external" href="https://github.com/PythonOT/POT/pull/808">PR #808</a>)</p></li>
<li><p>Add <code class="docutils literal notranslate"><span class="pre">ot.utils.apply_scaler</span></code> helper that dispatches preprocessing to a scaler object,
a callable, or a no-op (<a class="reference external" href="https://github.com/PythonOT/POT/pull/808">PR #808</a>)</p></li>
<li><p>Add optional <code class="docutils literal notranslate"><span class="pre">scaler</span></code> parameter to <code class="docutils literal notranslate"><span class="pre">sliced_wasserstein_distance</span></code> and <code class="docutils literal notranslate"><span class="pre">max_sliced_wasserstein_distance</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/808">PR #808</a>)</p></li>
<li><p>Add SGD based semi-discrete OT solver in <code class="docutils literal notranslate"><span class="pre">ot.semidiscrete</span></code> and a gallery example. (<a class="reference external" href="https://github.com/PythonOT/POT/pull/812">PR #812</a>)</p></li>
<li><p>Add a numerically stable log-domain solver for entropic partial Wasserstein, selectable via the new <code class="docutils literal notranslate"><span class="pre">method</span></code> parameter of <code class="docutils literal notranslate"><span class="pre">entropic_partial_wasserstein</span></code> (<code class="docutils literal notranslate"><span class="pre">method='sinkhorn_log'</span></code>) or directly through <code class="docutils literal notranslate"><span class="pre">entropic_partial_wasserstein_logscale</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/issues/723">Issue #723</a>)</p></li>
<li><p>Add cost functions between linear operators following <a class="reference external" href="https://arxiv.org/pdf/2509.24920">A Spectral-Grassmann Wasserstein metric for operator representations of dynamical systems</a>, implemented in <code class="docutils literal notranslate"><span class="pre">ot.sgot</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/792">PR #792</a>, PR #830)</p></li>
<li><p>Add batch FUGW loss to <code class="docutils literal notranslate"><span class="pre">ot.batch</span></code> and fix issues in some default parameters in the batch module (<a class="reference external" href="https://github.com/PythonOT/POT/pull/775">PR #775</a>)</p></li>
<li><p>Wrapper for barycenter solvers with free support <code class="docutils literal notranslate"><span class="pre">ot.solvers.bary_free_support</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/730">PR #730</a>)</p></li>
<li><p>Build wheels on ubuntu ARM to avoid QEMU emulation (<a class="reference external" href="https://github.com/PythonOT/POT/pull/818">PR #818</a>)</p></li>
<li><p>Add new methods to compute the linear transport map and the related 2-Wasserstein distance betweeen high-dimensional (HD) Gaussian distributions as described in [88], implemented in <code class="docutils literal notranslate"><span class="pre">ot.gaussian.bures_wasserstein_mapping_hd</span></code> and <code class="docutils literal notranslate"><span class="pre">ot.gaussian.bures_wasserstein_distance_hd</span></code>, respectively. Two additional methods estimate the same quantities from the source and destination observed data and are implemented in <code class="docutils literal notranslate"><span class="pre">ot.gaussian.empirical_bures_wasserstein_mapping_hd</span></code> and <code class="docutils literal notranslate"><span class="pre">ot.gaussian.empirical_bures_wasserstein_distance_hd</span></code>, respectively (<a class="reference external" href="https://github.com/PythonOT/POT/pull/814">PR #814</a>)</p></li>
<li><p>Update the geomloss wrapper to the new version and API (<a class="reference external" href="https://github.com/PythonOT/POT/pull/826">PR #826</a>)</p></li>
<li><p>Fix docstrings for <code class="docutils literal notranslate"><span class="pre">lowrank_gromov_wasserstein_samples</span></code> and <code class="docutils literal notranslate"><span class="pre">lowrank_sinkhorn</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/823">PR #823</a>)</p></li>
<li><p>Reorganize all tests per backend (<a class="reference external" href="https://github.com/PythonOT/POT/pull/828">PR #828</a>)</p></li>
<li><p>Implemented batch proximal point solver for OT problems <code class="docutils literal notranslate"><span class="pre">ot.batch.proximal_bregman_log_plan_batch</span></code> function and updated wrapper functions <code class="docutils literal notranslate"><span class="pre">ot.batch.solve_batch</span></code> and <code class="docutils literal notranslate"><span class="pre">ot.batch.solve_sample_batch</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/832">PR #832</a>)</p></li>
<li><p>Implement debiased OT solvers in <code class="docutils literal notranslate"><span class="pre">ot.solve_sample</span></code>.</p></li>
</ul>
</section>
<section id="id3">
<h3>Closed issues<a class="headerlink" href="#id3" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Fix label-aware cost correction in <code class="docutils literal notranslate"><span class="pre">ot.da</span></code> transport classes: the large cost was applied to unlabeled pairs instead of labeled pairs with different labels, so <code class="docutils literal notranslate"><span class="pre">ys</span></code>/<code class="docutils literal notranslate"><span class="pre">yt</span></code> had no effect when all labels were known (<a class="reference external" href="https://github.com/PythonOT/POT/pull/833">PR #833</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/664">Issue #664</a>)</p></li>
<li><p>Mitigate NaN regime of <code class="docutils literal notranslate"><span class="pre">entropic_partial_wasserstein</span></code> at small <code class="docutils literal notranslate"><span class="pre">reg</span></code> via a new log-domain solver, reachable with <code class="docutils literal notranslate"><span class="pre">entropic_partial_wasserstein(...,</span> <span class="pre">method='sinkhorn_log')</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/issues/723">Issue #723</a>; the default <code class="docutils literal notranslate"><span class="pre">method='sinkhorn'</span></code> path is unchanged — callers opt into the log-domain variant)</p></li>
<li><p>Fix NumPy 2.x compatibility in Brenier potential bounds (<a class="reference external" href="https://github.com/PythonOT/POT/pull/788">PR #788</a>)</p></li>
<li><p>Fix MSVC Windows build by removing <strong>restrict</strong> keyword (<a class="reference external" href="https://github.com/PythonOT/POT/pull/788">PR #788</a>)</p></li>
<li><p>Fix O(n³) performance bottleneck in sparse bipartite graph arc iteration (<a class="reference external" href="https://github.com/PythonOT/POT/pull/785">PR #785</a>)</p></li>
<li><p>Fix deprecated JAX function in <code class="docutils literal notranslate"><span class="pre">ot.backend.JaxBackend</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/771">PR #771</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/770">Issue #770</a>)</p></li>
<li><p>Add test for build from source (<a class="reference external" href="https://github.com/PythonOT/POT/pull/772">PR #772</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/764">Issue #764</a>)</p></li>
<li><p>Fix device for batch Ot solver in <code class="docutils literal notranslate"><span class="pre">ot.batch</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/784">PR #784</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/783">Issue #783</a>)</p></li>
<li><p>Fix openmp flags on macOS (<a class="reference external" href="https://github.com/PythonOT/POT/pull/789">PR #789</a>)</p></li>
<li><p>Clean documentation (<a class="reference external" href="https://github.com/PythonOT/POT/pull/787">PR #787</a>)</p></li>
<li><p>Fix code coverage (<a class="reference external" href="https://github.com/PythonOT/POT/pull/791">PR #791</a>)</p></li>
<li><p>Fix test of the version of jax in <code class="docutils literal notranslate"><span class="pre">ot.backend</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/794">PR #794</a>)</p></li>
<li><p>Reverting the openmp fix on macOS (<a class="reference external" href="https://github.com/PythonOT/POT/pull/789">PR #789</a>) for macOS (PR #797)</p></li>
<li><p>Align documentation build dependencies and doc extras (<a class="reference external" href="https://github.com/PythonOT/POT/pull/801">PR #801</a>)</p></li>
<li><p>Debug Debug linux test core dump (<a class="reference external" href="https://github.com/PythonOT/POT/pull/815">PR #815</a>)</p></li>
<li><p>Fix entropic regularization in <code class="docutils literal notranslate"><span class="pre">gcg</span></code>(<a class="reference external" href="https://github.com/PythonOT/POT/pull/817">PR #817</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/758">Issue #758</a>)</p></li>
<li><p>Fix documentation build on master with submodules (<a class="reference external" href="https://github.com/PythonOT/POT/pull/818">PR #818</a>)</p></li>
<li><p>Fix failing test for unbalanced solver with generic regularization (<a class="reference external" href="https://github.com/PythonOT/POT/pull/824">PR #824</a>)</p></li>
<li><p>Fix reference number error introduced in <a class="reference external" href="https://github.com/PythonOT/POT/pull/767">PR #767</a> (PR #819)</p></li>
<li><p>Fix docstrings for <code class="docutils literal notranslate"><span class="pre">lowrank_gromov_wasserstein_samples</span></code> and <code class="docutils literal notranslate"><span class="pre">lowrank_sinkhorn</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/823">PR #823</a>)</p></li>
<li><p>Update sgot cost function and example (<a class="reference external" href="https://github.com/PythonOT/POT/pull/830">PR #830</a>)</p></li>
</ul>
</section>
</section>
<section id="id4">
<h2>0.9.6.post1<a class="headerlink" href="#id4" title="Link to this heading"></a></h2>
<p><em>September 2025</em></p>
<section id="id5">
<h3>Closed issues<a class="headerlink" href="#id5" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Fix missing cython file in MANIFEST.in (<a class="reference external" href="https://github.com/PythonOT/POT/pull/763">PR #763</a>)</p></li>
</ul>
</section>
</section>
<section id="id6">
<h2>0.9.6<a class="headerlink" href="#id6" title="Link to this heading"></a></h2>
<p><em>September 2025</em></p>
<p>This new release contains several new features and bug fixes. Among the new features we have a new submodule <code class="docutils literal notranslate"><span class="pre">ot.batch</span></code> that contains batch parallel solvers for several OT problems including <a class="reference external" href="https://pythonot.github.io/master/auto_examples/backends/plot_ot_batch.html">Sinkhorn, Gromov-Wasserstein and Fused Gromov-Wasserstein</a>. This new submodule can be used to solve many independent OT problems in parallel on CPU or GPU with shared source and target support sizes. We also implemented a new Nystrom kernel approximation for the Sinkhorn solver that can be used to speed up the computation of the Sinkhorn divergence on large datasets. We also added new 1D solvers for <a class="reference external" href="https://pythonot.github.io/master/auto_examples/sliced-wasserstein/plot_compute_wasserstein_circle.html">Linear circular OT</a> and new solvers for free support <a class="reference external" href="https://pythonot.github.io/master/auto_examples/barycenters/plot_free_support_barycenter_generic_cost.html">OT barycenters with generic cost functions</a> and for <a class="reference external" href="https://pythonot.github.io/master/auto_examples/barycenters/plot_gmm_barycenter.html">barycenters between Gaussian Mixture Models (GMMs)</a>. We also implemented two solvers for partial Fused Gromov-Wasserstein problems based on <a class="reference external" href="https://pythonot.github.io/master/gen_modules/ot.gromov.html#ot.gromov.partial_fused_gromov_wasserstein">conditional gradient</a> and <a class="reference external" href="https://pythonot.github.io/master/gen_modules/ot.gromov.html#ot.gromov.entropic_partial_fused_gromov_wasserstein">projected gradient</a> descents.</p>
<p>Finally we have updated the documentation to reflect the new generic API and reorganized the <a class="reference external" href="https://pythonot.github.io/auto_examples/index.html">examples gallery</a>.</p>
<section id="id7">
<h3>New features<a class="headerlink" href="#id7" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Implement CG solvers for partial FGW (<a class="reference external" href="https://github.com/PythonOT/POT/pull/687">PR #687</a>)</p></li>
<li><p>Added feature <code class="docutils literal notranslate"><span class="pre">grad=last_step</span></code> for <code class="docutils literal notranslate"><span class="pre">ot.solvers.solve</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/693">PR #693</a>)</p></li>
<li><p>Automatic PR labeling and release file update check (<a class="reference external" href="https://github.com/PythonOT/POT/pull/704">PR #704</a>)</p></li>
<li><p>Reorganize sub-module <code class="docutils literal notranslate"><span class="pre">ot/lp/__init__.py</span></code> into separate files (<a class="reference external" href="https://github.com/PythonOT/POT/pull/714">PR #714</a>)</p></li>
<li><p>Implement fixed-point solver for OT barycenters with generic cost functions
(generalizes <code class="docutils literal notranslate"><span class="pre">ot.lp.free_support_barycenter</span></code>), with example. (<a class="reference external" href="https://github.com/PythonOT/POT/pull/715">PR #715</a>)</p></li>
<li><p>Implement fixed-point solver for barycenters between GMMs (<a class="reference external" href="https://github.com/PythonOT/POT/pull/715">PR #715</a>), with example.</p></li>
<li><p>Fix warning raise when import the library (<a class="reference external" href="https://github.com/PythonOT/POT/pull/716">PR #716</a>)</p></li>
<li><p>Implement projected gradient descent solvers for entropic partial FGW (<a class="reference external" href="https://github.com/PythonOT/POT/pull/702">PR #702</a>)</p></li>
<li><p>Fix documentation in the module <code class="docutils literal notranslate"><span class="pre">ot.gaussian</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/718">PR #718</a>)</p></li>
<li><p>Refactored <code class="docutils literal notranslate"><span class="pre">ot.bregman._convolutional</span></code> to improve readability (<a class="reference external" href="https://github.com/PythonOT/POT/pull/709">PR #709</a>)</p></li>
<li><p>Added <code class="docutils literal notranslate"><span class="pre">ot.gaussian.bures_barycenter_gradient_descent</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/680">PR #680</a>)</p></li>
<li><p>Added <code class="docutils literal notranslate"><span class="pre">ot.gaussian.bures_wasserstein_distance</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/680">PR #680</a>)</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">ot.gaussian.bures_wasserstein_distance</span></code> can be batched (<a class="reference external" href="https://github.com/PythonOT/POT/pull/680">PR #680</a>)</p></li>
<li><p>Backend implementation of <code class="docutils literal notranslate"><span class="pre">ot.dist</span></code> for (<a class="reference external" href="https://github.com/PythonOT/POT/pull/701">PR #701</a>)</p></li>
<li><p>Updated documentation Quickstart guide and User guide with new API (<a class="reference external" href="https://github.com/PythonOT/POT/pull/726">PR #726</a>)</p></li>
<li><p>Fix jax version for auto-grad (<a class="reference external" href="https://github.com/PythonOT/POT/pull/732">PR #732</a>)</p></li>
<li><p>Add Nystrom kernel approximation for Sinkhorn (<a class="reference external" href="https://github.com/PythonOT/POT/pull/742">PR #742</a>)</p></li>
<li><p>Added <code class="docutils literal notranslate"><span class="pre">ot.solver_1d.linear_circular_ot</span></code> and <code class="docutils literal notranslate"><span class="pre">ot.sliced.linear_sliced_wasserstein_sphere</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/736">PR #736</a>)</p></li>
<li><p>Implement 1d solver for partial optimal transport (<a class="reference external" href="https://github.com/PythonOT/POT/pull/741">PR #741</a>)</p></li>
<li><p>Fix reg_div function compatibility with numpy in <code class="docutils literal notranslate"><span class="pre">ot.unbalanced.lbfgsb_unbalanced</span></code> via new function <code class="docutils literal notranslate"><span class="pre">ot.utils.fun_to_numpy</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/731">PR #731</a>)</p></li>
<li><p>Added to each example in the examples gallery the information about the release version in which it was introduced (<a class="reference external" href="https://github.com/PythonOT/POT/pull/743">PR #743</a>)</p></li>
<li><p>Removed release information from quickstart guide (<a class="reference external" href="https://github.com/PythonOT/POT/pull/744">PR #744</a>)</p></li>
<li><p>Implement batch parallel solvers in ot.batch (<a class="reference external" href="https://github.com/PythonOT/POT/pull/745">PR #745</a>)</p></li>
<li><p>Update REAMDE with new API and reorganize examples (<a class="reference external" href="https://github.com/PythonOT/POT/pull/754">PR #754</a>)</p></li>
<li><p>Speedup and update tests and wheels (<a class="reference external" href="https://github.com/PythonOT/POT/pull/759">PR #759</a>)</p></li>
</ul>
</section>
<section id="id8">
<h3>Closed issues<a class="headerlink" href="#id8" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Fixed <code class="docutils literal notranslate"><span class="pre">ot.mapping</span></code> solvers which depended on deprecated <code class="docutils literal notranslate"><span class="pre">cvxpy</span></code> <code class="docutils literal notranslate"><span class="pre">ECOS</span></code> solver (<a class="reference external" href="https://github.com/PythonOT/POT/pull/692">PR #692</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/668">Issue #668</a>)</p></li>
<li><p>Fixed numerical errors in <code class="docutils literal notranslate"><span class="pre">ot.gmm</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/690">PR #690</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/689">Issue #689</a>)</p></li>
<li><p>Add version number to the documentation (<a class="reference external" href="https://github.com/PythonOT/POT/pull/696">PR #696</a>)</p></li>
<li><p>Update doc for default regularization in <code class="docutils literal notranslate"><span class="pre">ot.unbalanced</span></code> sinkhorn solvers (<a class="reference external" href="https://github.com/PythonOT/POT/issues/691">Issue #691</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/700">PR #700</a>)</p></li>
<li><p>Clean documentation for <code class="docutils literal notranslate"><span class="pre">gromov</span></code>, <code class="docutils literal notranslate"><span class="pre">lp</span></code> and <code class="docutils literal notranslate"><span class="pre">unbalanced</span></code> folders (<a class="reference external" href="https://github.com/PythonOT/POT/pull/710">PR #710</a>)</p></li>
<li><p>Clean references in documentation (<a class="reference external" href="https://github.com/PythonOT/POT/pull/722">PR #722</a>)</p></li>
<li><p>Clean documentation for <code class="docutils literal notranslate"><span class="pre">ot.gromov.gromov_wasserstein</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/737">PR #737</a>)</p></li>
<li><p>Debug wheels building (<a class="reference external" href="https://github.com/PythonOT/POT/pull/739">PR #739</a>)</p></li>
<li><p>Fix doc for projection sparse simplex (<a class="reference external" href="https://github.com/PythonOT/POT/pull/734">PR #734</a>, PR #746)</p></li>
<li><p>Changed the default behavior of <code class="docutils literal notranslate"><span class="pre">ot.lp.solver_1d.wasserstein_circle</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/issues/738">Issue #738</a>)</p></li>
<li><p>Avoid raising unnecessary warnings in <code class="docutils literal notranslate"><span class="pre">ot.lp.solver_1d.binary_search_circle</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/issues/738">Issue #738</a>)</p></li>
<li><p>Avoid deprecation warning in <code class="docutils literal notranslate"><span class="pre">ot.lp.solver_1d.wasserstein_1d</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/issues/760">Issue #760</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/761">PR #761</a>)</p></li>
</ul>
</section>
</section>
<section id="id9">
<h2>0.9.5<a class="headerlink" href="#id9" title="Link to this heading"></a></h2>
<p><em>November 2024</em></p>
<p>This new release contains several new features, starting with
a novel <a class="reference external" href="https://pythonot.github.io/master/gen_modules/ot.gmm.html#examples-using-ot-gmm-gmm-ot-apply-map">Gaussian Mixture Model Optimal Transport (GMM-OT)</a> solver to compare GMM while enforcing the transport plan to remain a GMM, that benefits from a closed-form solution making it practical for high-dimensional matching problems. We also extended our general unbalanced OT solvers to support any non-negative reference measure in the regularization terms, before adding the novel <a class="reference external" href="https://pythonot.github.io/master/auto_examples/unbalanced-partial/plot_conv_sinkhorn_ti.html">translation invariant UOT</a> solver showcasing a higher convergence speed. We also implemented several new solvers and enhanced existing ones to perform OT across spaces. These include a <a class="reference external" href="https://pythonot.github.io/master/auto_examples/gromov/plot_semirelaxed_gromov_wasserstein_barycenter.html">semi-relaxed FGW barycenter</a> solver, coupled with new initialization heuristics for the inner divergence computation, to perform graph partitioning or dictionary learning. Followed by novel <a class="reference external" href="https://pythonot.github.io/master/auto_examples/others/plot_outlier_detection_with_COOT_and_unbalanced_COOT.html">unbalanced FGW and Co-optimal transport</a> solvers to promote robustness to outliers in such matching problems. And we finally updated the implementation of partial GW now supporting asymmetric structures and the KL divergence, while leveraging a new generic conditional gradient solver for partial transport problems enabling significant speed improvements. These latest updates required some modifications to the line search functions of our generic conditional gradient solver, paving the way for future improvements to other GW-based solvers. Last but not least, we implemented a pre-commit scheme to automatically correct common programming mistakes likely to be made by our future contributors.</p>
<p>This release also contains few bug fixes, concerning the support of any metric in <code class="docutils literal notranslate"><span class="pre">ot.emd_1d</span></code> / <code class="docutils literal notranslate"><span class="pre">ot.emd2_1d</span></code>, and the support of any weights in <code class="docutils literal notranslate"><span class="pre">ot.gaussian</span></code>.</p>
<section id="breaking-change">
<h3>Breaking change<a class="headerlink" href="#breaking-change" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Custom functions provided as parameter <code class="docutils literal notranslate"><span class="pre">line_search</span></code> to <code class="docutils literal notranslate"><span class="pre">ot.optim.generic_conditional_gradient</span></code> must now have the signature <code class="docutils literal notranslate"><span class="pre">line_search(cost,</span> <span class="pre">G,</span> <span class="pre">deltaG,</span> <span class="pre">Mi,</span> <span class="pre">cost_G,</span> <span class="pre">df_G,</span> <span class="pre">**kwargs)</span></code>, adding as input <code class="docutils literal notranslate"><span class="pre">df_G</span></code> the gradient of the regularizer evaluated at the transport plan <code class="docutils literal notranslate"><span class="pre">G</span></code>. This change aims at improving speed of solvers having quadratic polynomial functions as regularizer such as the Gromov-Wassertein loss (<a class="reference external" href="https://github.com/PythonOT/POT/pull/663">PR #663</a>).</p></li>
</ul>
</section>
<section id="id10">
<h3>New features<a class="headerlink" href="#id10" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>New linter based on pre-commit using ruff, codespell and yamllint (<a class="reference external" href="https://github.com/PythonOT/POT/pull/681">PR #681</a>)</p></li>
<li><p>Added feature <code class="docutils literal notranslate"><span class="pre">mass=True</span></code> for <code class="docutils literal notranslate"><span class="pre">nx.kl_div</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/654">PR #654</a>)</p></li>
<li><p>Implemented Gaussian Mixture Model OT <code class="docutils literal notranslate"><span class="pre">ot.gmm</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/649">PR #649</a>)</p></li>
<li><p>Added feature <code class="docutils literal notranslate"><span class="pre">semirelaxed_fgw_barycenters</span></code> and generic FGW-related barycenter updates <code class="docutils literal notranslate"><span class="pre">update_barycenter_structure</span></code> and <code class="docutils literal notranslate"><span class="pre">update_barycenter_feature</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/659">PR #659</a>)</p></li>
<li><p>Added initialization heuristics for sr(F)GW problems via <code class="docutils literal notranslate"><span class="pre">semirelaxed_init_plan</span></code>, integrated in all sr(F)GW solvers (<a class="reference external" href="https://github.com/PythonOT/POT/pull/659">PR #659</a>)</p></li>
<li><p>Improved <code class="docutils literal notranslate"><span class="pre">ot.plot.plot1D_mat</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/649">PR #649</a>)</p></li>
<li><p>Added <code class="docutils literal notranslate"><span class="pre">nx.det</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/649">PR #649</a>)</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">nx.sqrtm</span></code> is now broadcastable (takes …, d, d) inputs (<a class="reference external" href="https://github.com/PythonOT/POT/pull/649">PR #649</a>)</p></li>
<li><p>Restructured <code class="docutils literal notranslate"><span class="pre">ot.unbalanced</span></code> module (<a class="reference external" href="https://github.com/PythonOT/POT/pull/658">PR #658</a>)</p></li>
<li><p>Added <code class="docutils literal notranslate"><span class="pre">ot.unbalanced.lbfgsb_unbalanced2</span></code> and add flexible reference measure <code class="docutils literal notranslate"><span class="pre">c</span></code> in all unbalanced solvers (<a class="reference external" href="https://github.com/PythonOT/POT/pull/658">PR #658</a>)</p></li>
<li><p>Implemented Fused unbalanced Gromov-Wasserstein and unbalanced Co-Optimal Transport (<a class="reference external" href="https://github.com/PythonOT/POT/pull/677">PR #677</a>)</p></li>
<li><p>Notes before depreciating partial Gromov-Wasserstein function in <code class="docutils literal notranslate"><span class="pre">ot.partial</span></code> moved to ot.gromov (<a class="reference external" href="https://github.com/PythonOT/POT/pull/663">PR #663</a>)</p></li>
<li><p>Create <code class="docutils literal notranslate"><span class="pre">ot.gromov._partial</span></code> add new features <code class="docutils literal notranslate"><span class="pre">loss_fun</span> <span class="pre">=</span> <span class="pre">"kl_loss"</span></code> and <code class="docutils literal notranslate"><span class="pre">symmetry=False</span></code> to all solvers while increasing speed + updating adequatly <code class="docutils literal notranslate"><span class="pre">ot.solvers</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/663">PR #663</a>)</p></li>
<li><p>Added <code class="docutils literal notranslate"><span class="pre">ot.unbalanced.sinkhorn_unbalanced_translation_invariant</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/676">PR #676</a>)</p></li>
</ul>
</section>
<section id="id11">
<h3>Closed issues<a class="headerlink" href="#id11" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Fixed <code class="docutils literal notranslate"><span class="pre">ot.gaussian</span></code> ignoring weights when computing means (<a class="reference external" href="https://github.com/PythonOT/POT/pull/649">PR #649</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/648">Issue #648</a>)</p></li>
<li><p>Fixed <code class="docutils literal notranslate"><span class="pre">ot.emd_1d</span></code> and <code class="docutils literal notranslate"><span class="pre">ot.emd2_1d</span></code> incorrectly allowing any metric (<a class="reference external" href="https://github.com/PythonOT/POT/pull/670">PR #670</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/669">Issue #669</a>)</p></li>
</ul>
</section>
</section>
<section id="id12">
<h2>0.9.4<a class="headerlink" href="#id12" title="Link to this heading"></a></h2>
<p><em>June 2024</em></p>
<p>This new release contains several new features and bug fixes. Among the new features
we have novel <a class="reference external" href="https://pythonot.github.io/auto_examples/gromov/plot_quantized_gromov_wasserstein.html">Quantized FGW</a> solvers that can be used to speed up the computation of the FGW loss on large datasets or to promote a structure on the pairwise matrices. We also updated the continuous entropic mapping to provide efficient out-of-sample continuous mapping thanks to entropic regularization. We also have a new general unbalanced solvers for <code class="docutils literal notranslate"><span class="pre">ot.solve</span></code> and BFGS solver and illustrative example. Finally we have a new solver for the <a class="reference external" href="https://pythonot.github.io/auto_examples/others/plot_lowrank_GW.html">Low Rank Gromov-Wasserstein</a> that can be used to compute the GW distance between two large scale datasets with a low rank approximation.</p>
<p>From a maintenance point of view, we now have a new option to install optional dependencies with <code class="docutils literal notranslate"><span class="pre">pip</span> <span class="pre">install</span> <span class="pre">POT[all]</span></code> and the specific backends or submodules’ dependencies may also be installed individually. The pip options are: <code class="docutils literal notranslate"><span class="pre">backend-jax,</span> <span class="pre">backend-tf,</span> <span class="pre">backend-torch,</span> <span class="pre">cvxopt,</span> <span class="pre">dr,</span> <span class="pre">gnn,</span> <span class="pre">plot,</span> <span class="pre">all</span></code>. We also provide with this release support for NumPy 2.0 (the wheels should now be compatible with NumPy 2.0 and below). We also fixed several issues such as gradient sign errors for FGW solvers, empty weights for <code class="docutils literal notranslate"><span class="pre">ot.emd2</span></code>, and line-search in partial GW. We also split the <code class="docutils literal notranslate"><span class="pre">test/test_gromov.py</span></code> into <code class="docutils literal notranslate"><span class="pre">test/gromov/</span></code> to make the tests more manageable.</p>
<section id="id13">
<h3>New features<a class="headerlink" href="#id13" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>NumPy 2.0 support is added (<a class="reference external" href="https://github.com/PythonOT/POT/pull/629">PR #629</a>)</p></li>
<li><p>New quantized FGW solvers <code class="docutils literal notranslate"><span class="pre">ot.gromov.quantized_fused_gromov_wasserstein</span></code>, <code class="docutils literal notranslate"><span class="pre">ot.gromov.quantized_fused_gromov_wasserstein_samples</span></code> and <code class="docutils literal notranslate"><span class="pre">ot.gromov.quantized_fused_gromov_wasserstein_partitioned</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/603">PR #603</a>)</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">ot.gromov._gw.solve_gromov_linesearch</span></code> now has an argument to specify if the matrices are symmetric in which case the computation can be done faster (<a class="reference external" href="https://github.com/PythonOT/POT/pull/607">PR #607</a>).</p></li>
<li><p>Continuous entropic mapping (<a class="reference external" href="https://github.com/PythonOT/POT/pull/613">PR #613</a>)</p></li>
<li><p>New general unbalanced solvers for <code class="docutils literal notranslate"><span class="pre">ot.solve</span></code> and BFGS solver and illustrative example (<a class="reference external" href="https://github.com/PythonOT/POT/pull/620">PR #620</a>)</p></li>
<li><p>Add gradient computation with envelope theorem to sinkhorn solver of <code class="docutils literal notranslate"><span class="pre">ot.solve</span></code> with <code class="docutils literal notranslate"><span class="pre">grad='envelope'</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/605">PR #605</a>).</p></li>
<li><p>Added support for <a class="reference external" href="https://proceedings.mlr.press/v162/scetbon22b/scetbon22b.pdf">Low rank Gromov-Wasserstein</a> with <code class="docutils literal notranslate"><span class="pre">ot.gromov.lowrank_gromov_wasserstein_samples</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/614">PR #614</a>)</p></li>
<li><p>Optional dependencies may now be installed with <code class="docutils literal notranslate"><span class="pre">pip</span> <span class="pre">install</span> <span class="pre">POT[all]</span></code> The specific backends or submodules’ dependencies may also be installed individually. The pip options are: <code class="docutils literal notranslate"><span class="pre">backend-jax,</span> <span class="pre">backend-tf,</span> <span class="pre">backend-torch,</span> <span class="pre">cvxopt,</span> <span class="pre">dr,</span> <span class="pre">gnn,</span> <span class="pre">all</span></code>. The installation of the <code class="docutils literal notranslate"><span class="pre">cupy</span></code> backend should be done with conda.</p></li>
</ul>
</section>
<section id="id14">
<h3>Closed issues<a class="headerlink" href="#id14" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Fix gpu compatibility of sr(F)GW solvers when <code class="docutils literal notranslate"><span class="pre">G0</span> <span class="pre">is</span> <span class="pre">not</span> <span class="pre">None</span></code>(<a class="reference external" href="https://github.com/PythonOT/POT/pull/596">PR #596</a>)</p></li>
<li><p>Fix doc and example for lowrank sinkhorn (<a class="reference external" href="https://github.com/PythonOT/POT/pull/601">PR #601</a>)</p></li>
<li><p>Fix issue with empty weights for <code class="docutils literal notranslate"><span class="pre">ot.emd2</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/606">PR #606</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/534">Issue #534</a>)</p></li>
<li><p>Fix a sign error regarding the gradient of <code class="docutils literal notranslate"><span class="pre">ot.gromov._gw.fused_gromov_wasserstein2</span></code> and <code class="docutils literal notranslate"><span class="pre">ot.gromov._gw.gromov_wasserstein2</span></code> for the kl loss (<a class="reference external" href="https://github.com/PythonOT/POT/pull/610">PR #610</a>)</p></li>
<li><p>Fix same sign error for sr(F)GW conditional gradient solvers (<a class="reference external" href="https://github.com/PythonOT/POT/pull/611">PR #611</a>)</p></li>
<li><p>Split <code class="docutils literal notranslate"><span class="pre">test/test_gromov.py</span></code> into <code class="docutils literal notranslate"><span class="pre">test/gromov/</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/619">PR #619</a>)</p></li>
<li><p>Fix (F)GW barycenter functions to support computing barycenter on 1 input + deprecate structures as lists (<a class="reference external" href="https://github.com/PythonOT/POT/pull/628">PR #628</a>)</p></li>
<li><p>Fix line-search in partial GW and change default init to the interior of partial transport plans (<a class="reference external" href="https://github.com/PythonOT/POT/pull/602">PR #602</a>)</p></li>
<li><p>Fix <code class="docutils literal notranslate"><span class="pre">ot.da.sinkhorn_lpl1_mm</span></code> compatibility with JAX (<a class="reference external" href="https://github.com/PythonOT/POT/pull/592">PR #592</a>)</p></li>
<li><p>Fiw linesearch import error on Scipy 1.14 (<a class="reference external" href="https://github.com/PythonOT/POT/pull/642">PR #642</a>, <a class="reference external" href="https://github.com/PythonOT/POT/issues/641">Issue #641</a>)</p></li>
<li><p>Upgrade supported JAX versions from jax<=0.4.24 to jax<=0.4.30 (<a class="reference external" href="https://github.com/PythonOT/POT/pull/643">PR #643</a>)</p></li>
</ul>
</section>
</section>
<section id="id15">
<h2>0.9.3<a class="headerlink" href="#id15" title="Link to this heading"></a></h2>
<p><em>January 2024</em></p>
<section id="id16">
<h3>Closed issues<a class="headerlink" href="#id16" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Fixed an issue with cost correction for mismatched labels in <code class="docutils literal notranslate"><span class="pre">ot.da.BaseTransport</span></code> fit methods. This fix addresses the original issue introduced <a class="reference external" href="https://github.com/PythonOT/POT/pull/587">PR #587</a> (PR #593)</p></li>
</ul>
</section>
</section>
<section id="id17">
<h2>0.9.2<a class="headerlink" href="#id17" title="Link to this heading"></a></h2>
<p><em>December 2023</em></p>
<p>This new release contains several new features and bug fixes. Among the new features
we have a new solver for estimation of nearest Brenier potentials (SSNB) that can be used for OT mapping estimation (on small problems), new Bregman Alternated Projected Gradient solvers for GW and FGW, and new solvers for Bures-Wasserstein barycenters. We also provide a first solver for Low Rank Sinkhorn that will be ussed to provide low rak OT extensions in the next releases. Finally we have a new exact line-search for (F)GW solvers with KL loss that can be used to improve the convergence of the solvers.</p>
<p>We also have a new <code class="docutils literal notranslate"><span class="pre">LazyTensor</span></code> class that can be used to model OT plans and low rank tensors in large scale OT. This class is used to return the plan for the new wrapper for <code class="docutils literal notranslate"><span class="pre">geomloss</span></code> Sinkhorn solver on empirical samples that can lead to x10/x100 speedups on CPU or GPU and have a lazy implementation that allows solving very large problems of a few millions samples.</p>
<p>We also have a new API for solving OT problems from empirical samples with <code class="docutils literal notranslate"><span class="pre">ot.solve_sample</span></code> Finally we have a new API for Gromov-Wasserstein solvers with <code class="docutils literal notranslate"><span class="pre">ot.solve_gromov</span></code> function that centralizes most of the (F)GW methods with unified notation. Some example of how to use the new API below:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="c1"># Generate random data</span>
<span class="n">xs</span><span class="p">,</span> <span class="n">xt</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">random</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">100</span><span class="p">,</span> <span class="mi">2</span><span class="p">),</span> <span class="n">np</span><span class="o">.</span><span class="n">random</span><span class="o">.</span><span class="n">randn</span><span class="p">(</span><span class="mi">50</span><span class="p">,</span> <span class="mi">2</span><span class="p">)</span>
<span class="c1"># Solve OT problem with empirical samples</span>
<span class="n">sol</span> <span class="o">=</span> <span class="n">ot</span><span class="o">.</span><span class="n">solve_sample</span><span class="p">(</span><span class="n">xs</span><span class="p">,</span> <span class="n">xt</span><span class="p">)</span> <span class="c1"># Exact OT betwen smaples with uniform weights</span>
<span class="n">sol</span> <span class="o">=</span> <span class="n">ot</span><span class="o">.</span><span class="n">solve_sample</span><span class="p">(</span><span class="n">xs</span><span class="p">,</span> <span class="n">xt</span><span class="p">,</span> <span class="n">wa</span><span class="p">,</span> <span class="n">wb</span><span class="p">)</span> <span class="c1"># Exact OT with weights given by user</span>
<span class="n">sol</span> <span class="o">=</span> <span class="n">ot</span><span class="o">.</span><span class="n">solve_sample</span><span class="p">(</span><span class="n">xs</span><span class="p">,</span> <span class="n">xt</span><span class="p">,</span> <span class="n">reg</span><span class="o">=</span> <span class="mi">1</span><span class="p">,</span> <span class="n">metric</span><span class="o">=</span><span class="s1">'euclidean'</span><span class="p">)</span> <span class="c1"># sinkhorn with euclidean metric</span>
<span class="n">sol</span> <span class="o">=</span> <span class="n">ot</span><span class="o">.</span><span class="n">solve_sample</span><span class="p">(</span><span class="n">xs</span><span class="p">,</span> <span class="n">xt</span><span class="p">,</span> <span class="n">reg</span><span class="o">=</span> <span class="mi">1</span><span class="p">,</span> <span class="n">method</span><span class="o">=</span><span class="s1">'geomloss'</span><span class="p">)</span> <span class="c1"># faster sinkhorn solver on CPU/GPU</span>
<span class="n">sol</span> <span class="o">=</span> <span class="n">ot</span><span class="o">.</span><span class="n">solve_sample</span><span class="p">(</span><span class="n">x</span><span class="p">,</span><span class="n">x2</span><span class="p">,</span> <span class="n">method</span><span class="o">=</span><span class="s1">'factored'</span><span class="p">,</span> <span class="n">rank</span><span class="o">=</span><span class="mi">10</span><span class="p">)</span> <span class="c1"># compute factored OT</span>
<span class="n">sol</span> <span class="o">=</span> <span class="n">ot</span><span class="o">.</span><span class="n">solve_sample</span><span class="p">(</span><span class="n">x</span><span class="p">,</span><span class="n">x2</span><span class="p">,</span> <span class="n">method</span><span class="o">=</span><span class="s1">'lowrank'</span><span class="p">,</span> <span class="n">rank</span><span class="o">=</span><span class="mi">10</span><span class="p">)</span> <span class="c1"># compute lowrank sinkhorn OT</span>
<span class="n">value_bw</span> <span class="o">=</span> <span class="n">ot</span><span class="o">.</span><span class="n">solve_sample</span><span class="p">(</span><span class="n">xs</span><span class="p">,</span> <span class="n">xt</span><span class="p">,</span> <span class="n">method</span><span class="o">=</span><span class="s1">'gaussian'</span><span class="p">)</span><span class="o">.</span><span class="n">value</span> <span class="c1"># Bures-Wasserstein distance</span>
<span class="c1"># Solve GW problem</span>
<span class="n">Cs</span><span class="p">,</span> <span class="n">Ct</span> <span class="o">=</span> <span class="n">ot</span><span class="o">.</span><span class="n">dist</span><span class="p">(</span><span class="n">xs</span><span class="p">,</span> <span class="n">xs</span><span class="p">),</span> <span class="n">ot</span><span class="o">.</span><span class="n">dist</span><span class="p">(</span><span class="n">xt</span><span class="p">,</span> <span class="n">xt</span><span class="p">)</span> <span class="c1"># compute cost matrices</span>
<span class="n">sol</span> <span class="o">=</span> <span class="n">ot</span><span class="o">.</span><span class="n">solve_gromov</span><span class="p">(</span><span class="n">Cs</span><span class="p">,</span><span class="n">Ct</span><span class="p">)</span> <span class="c1"># Exact GW between samples with uniform weights</span>
<span class="c1"># Solve FGW problem</span>
<span class="n">M</span> <span class="o">=</span> <span class="n">ot</span><span class="o">.</span><span class="n">dist</span><span class="p">(</span><span class="n">xs</span><span class="p">,</span> <span class="n">xt</span><span class="p">)</span> <span class="c1"># compute cost matrix</span>
<span class="c1"># Exact FGW between samples with uniform weights</span>
<span class="n">sol</span> <span class="o">=</span> <span class="n">ot</span><span class="o">.</span><span class="n">solve_gromov</span><span class="p">(</span><span class="n">Cs</span><span class="p">,</span> <span class="n">Ct</span><span class="p">,</span> <span class="n">M</span><span class="p">,</span> <span class="n">loss</span><span class="o">=</span><span class="s1">'KL'</span><span class="p">,</span> <span class="n">alpha</span><span class="o">=</span><span class="mf">0.7</span><span class="p">)</span> <span class="c1"># FGW with KL data fitting</span>
<span class="c1"># recover solutions objects</span>
<span class="n">P</span> <span class="o">=</span> <span class="n">sol</span><span class="o">.</span><span class="n">plan</span> <span class="c1"># OT plan</span>
<span class="n">u</span><span class="p">,</span> <span class="n">v</span> <span class="o">=</span> <span class="n">sol</span><span class="o">.</span><span class="n">potentials</span> <span class="c1"># dual variables</span>
<span class="n">value</span> <span class="o">=</span> <span class="n">sol</span><span class="o">.</span><span class="n">value</span> <span class="c1"># OT value</span>
<span class="c1"># for GW and FGW</span>
<span class="n">value_linear</span> <span class="o">=</span> <span class="n">sol</span><span class="o">.</span><span class="n">value_linear</span> <span class="c1"># linear part of the loss</span>
<span class="n">value_quad</span> <span class="o">=</span> <span class="n">sol</span><span class="o">.</span><span class="n">value_quad</span> <span class="c1"># quadratic part of the loss</span>
</pre></div>
</div>
<p>Users are encouraged to use the new API (it is much simpler) but it might still be subjects to small changes before the release of POT 1.0 .</p>
<p>We also fixed a number of issues, the most pressing being a problem of GPU memory allocation when pytorch is installed that will not happen now thanks to Lazy initialization of the backends. We now also have the possibility to deactivate some backends using environment which prevents POT from importing them and can lead to large import speedup.</p>
<section id="id18">
<h3>New features<a class="headerlink" href="#id18" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Added support for <a class="reference external" href="http://proceedings.mlr.press/v108/paty20a/paty20a.pdf">Nearest Brenier Potentials (SSNB)</a> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/526">PR #526</a>) + minor fix (PR #535)</p></li>
<li><p>Tweaked <code class="docutils literal notranslate"><span class="pre">get_backend</span></code> to ignore <code class="docutils literal notranslate"><span class="pre">None</span></code> inputs (<a class="reference external" href="https://github.com/PythonOT/POT/pull/525">PR #525</a>)</p></li>
<li><p>Callbacks for generalized conditional gradient in <code class="docutils literal notranslate"><span class="pre">ot.da.sinkhorn_l1l2_gl</span></code> are now vectorized to improve performance (<a class="reference external" href="https://github.com/PythonOT/POT/pull/507">PR #507</a>)</p></li>
<li><p>The <code class="docutils literal notranslate"><span class="pre">linspace</span></code> method of the backends now has the <code class="docutils literal notranslate"><span class="pre">type_as</span></code> argument to convert to the same dtype and device. (<a class="reference external" href="https://github.com/PythonOT/POT/pull/533">PR #533</a>)</p></li>
<li><p>The <code class="docutils literal notranslate"><span class="pre">convolutional_barycenter2d</span></code> and <code class="docutils literal notranslate"><span class="pre">convolutional_barycenter2d_debiased</span></code> functions now work with different devices.. (<a class="reference external" href="https://github.com/PythonOT/POT/pull/533">PR #533</a>)</p></li>
<li><p>New API for Gromov-Wasserstein solvers with <code class="docutils literal notranslate"><span class="pre">ot.solve_gromov</span></code> function (<a class="reference external" href="https://github.com/PythonOT/POT/pull/536">PR #536</a>)</p></li>
<li><p>New LP solvers from scipy used by default for LP barycenter (<a class="reference external" href="https://github.com/PythonOT/POT/pull/537">PR #537</a>)</p></li>
<li><p>Update wheels to Python 3.12 and remove old i686 arch that do not have scipy wheels (<a class="reference external" href="https://github.com/PythonOT/POT/pull/543">PR #543</a>)</p></li>
<li><p>Upgraded unbalanced OT solvers for more flexibility (<a class="reference external" href="https://github.com/PythonOT/POT/pull/539">PR #539</a>)</p></li>
<li><p>Add LazyTensor for modeling plans and low rank tensor in large scale OT (<a class="reference external" href="https://github.com/PythonOT/POT/pull/544">PR #544</a>)</p></li>
<li><p>Add exact line-search for <code class="docutils literal notranslate"><span class="pre">gromov_wasserstein</span></code> and <code class="docutils literal notranslate"><span class="pre">fused_gromov_wasserstein</span></code> with KL loss (<a class="reference external" href="https://github.com/PythonOT/POT/pull/556">PR #556</a>)</p></li>
<li><p>Add KL loss to all semi-relaxed (Fused) Gromov-Wasserstein solvers (<a class="reference external" href="https://github.com/PythonOT/POT/pull/559">PR #559</a>)</p></li>
<li><p>Further upgraded unbalanced OT solvers for more flexibility and future use (<a class="reference external" href="https://github.com/PythonOT/POT/pull/551">PR #551</a>)</p></li>
<li><p>New API function <code class="docutils literal notranslate"><span class="pre">ot.solve_sample</span></code> for solving OT problems from empirical samples (<a class="reference external" href="https://github.com/PythonOT/POT/pull/563">PR #563</a>)</p></li>
<li><p>Wrapper for `geomloss`` solver on empirical samples (<a class="reference external" href="https://github.com/PythonOT/POT/pull/571">PR #571</a>)</p></li>
<li><p>Add <code class="docutils literal notranslate"><span class="pre">stop_criterion</span></code> feature to (un)regularized (f)gw barycenter solvers (<a class="reference external" href="https://github.com/PythonOT/POT/pull/578">PR #578</a>)</p></li>
<li><p>Add <code class="docutils literal notranslate"><span class="pre">fixed_structure</span></code> and <code class="docutils literal notranslate"><span class="pre">fixed_features</span></code> to entropic fgw barycenter solver (<a class="reference external" href="https://github.com/PythonOT/POT/pull/578">PR #578</a>)</p></li>
<li><p>Add new BAPG solvers with KL projections for GW and FGW (<a class="reference external" href="https://github.com/PythonOT/POT/pull/581">PR #581</a>)</p></li>
<li><p>Add Bures-Wasserstein barycenter in <code class="docutils literal notranslate"><span class="pre">ot.gaussian</span></code> and example (<a class="reference external" href="https://github.com/PythonOT/POT/pull/582">PR #582</a>, PR #584)</p></li>
<li><p>Domain adaptation method <code class="docutils literal notranslate"><span class="pre">SinkhornL1l2Transport</span></code> now supports JAX backend (<a class="reference external" href="https://github.com/PythonOT/POT/pull/587">PR #587</a>)</p></li>
<li><p>Added support for <a class="reference external" href="https://arxiv.org/pdf/2103.04737.pdf">Low-Rank Sinkhorn Factorization</a> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/568">PR #568</a>)</p></li>
</ul>
</section>
<section id="id19">
<h3>Closed issues<a class="headerlink" href="#id19" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Fix line search evaluating cost outside of the interpolation range (<a class="reference external" href="https://github.com/PythonOT/POT/issues/502">Issue #502</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/504">PR #504</a>)</p></li>
<li><p>Lazily instantiate backends to avoid unnecessary GPU memory pre-allocations on package import (<a class="reference external" href="https://github.com/PythonOT/POT/issues/516">Issue #516</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/520">PR #520</a>)</p></li>
<li><p>Handle documentation and warnings when integers are provided to (f)gw solvers based on cg (<a class="reference external" href="https://github.com/PythonOT/POT/issues/530">Issue #530</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/559">PR #559</a>)</p></li>
<li><p>Correct independence of <code class="docutils literal notranslate"><span class="pre">fgw_barycenters</span></code> to <code class="docutils literal notranslate"><span class="pre">init_C</span></code> and <code class="docutils literal notranslate"><span class="pre">init_X</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/issues/547">Issue #547</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/566">PR #566</a>)</p></li>
<li><p>Avoid precision change when computing norm using PyTorch backend (Discussion #570, <a class="reference external" href="https://github.com/PythonOT/POT/pull/572">PR #572</a>)</p></li>
<li><p>Create <code class="docutils literal notranslate"><span class="pre">ot/bregman/</span></code>repository (<a class="reference external" href="https://github.com/PythonOT/POT/issues/567">Issue #567</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/569">PR #569</a>)</p></li>
<li><p>Fix matrix feature shape in <code class="docutils literal notranslate"><span class="pre">entropic_fused_gromov_barycenters</span></code>(<a class="reference external" href="https://github.com/PythonOT/POT/issues/574">Issue #574</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/573">PR #573</a>)</p></li>
<li><p>Fix (fused) gromov-wasserstein barycenter solvers to support <code class="docutils literal notranslate"><span class="pre">kl_loss</span></code>(<a class="reference external" href="https://github.com/PythonOT/POT/pull/576">PR #576</a>)</p></li>
</ul>
</section>
</section>
<section id="id20">
<h2>0.9.1<a class="headerlink" href="#id20" title="Link to this heading"></a></h2>
<p><em>August 2023</em></p>
<p>This new release contains several new features and bug fixes.</p>
<p>New features include a new submodule <code class="docutils literal notranslate"><span class="pre">ot.gnn</span></code> that contains two new Graph neural network layers (compatible with <a class="reference external" href="https://pytorch-geometric.readthedocs.io/">Pytorch Geometric</a>) for template-based pooling of graphs with an example on <a class="reference external" href="https://pythonot.github.io/master/auto_examples/gromov/plot_gnn_TFGW.html">graph classification</a>. Related to this, we also now provide FGW and semi relaxed FGW solvers for which the resulting loss is differentiable w.r.t. the parameter <code class="docutils literal notranslate"><span class="pre">alpha</span></code>. Other contributions on the (F)GW front include a new solver for the Proximal Point algorithm <a class="reference external" href="https://pythonot.github.io/master/auto_examples/gromov/plot_fgw_solvers.html">that can be used to solve entropic GW problems</a> (using the parameter <code class="docutils literal notranslate"><span class="pre">solver="PPA"</span></code>), new solvers for entropic FGW barycenters, novels Sinkhorn-based solvers for entropic semi-relaxed (F)GW, the possibility to provide a warm-start to the solvers, and optional marginal weights of the samples (uniform weights ar used by default). Finally we added in the submodule <code class="docutils literal notranslate"><span class="pre">ot.gaussian</span></code> and <code class="docutils literal notranslate"><span class="pre">ot.da</span></code> new loss and mapping estimators for the Gaussian Gromov-Wasserstein that can be used as a fast alternative to GW and estimates linear mappings between unregistered spaces that can potentially have different size (See the update <a class="reference external" href="https://pythonot.github.io/master/auto_examples/domain-adaptation/plot_otda_linear_mapping.html">linear mapping example</a> for an illustration).</p>
<p>We also provide a new solver for the <a class="reference external" href="https://pythonot.github.io/master/auto_examples/others/plot_EWCA.html">Entropic Wasserstein Component Analysis</a> that is a generalization of the celebrated PCA taking into account the local neighborhood of the samples. We also now have a new solver in <code class="docutils literal notranslate"><span class="pre">ot.smooth</span></code> for the <a class="reference external" href="https://pythonot.github.io/master/auto_examples/plot_OT_1D_smooth.html">sparsity-constrained OT (last plot)</a> that can be used to find regularized OT plans with sparsity constraints. Finally we have a first multi-marginal solver for regular 1D distributions with a Monge loss (see <a class="reference external" href="https://pythonot.github.io/master/auto_examples/others/plot_dmmot.html">here</a>).</p>
<p>The documentation and testings have also been updated. We now have nearly 95% code coverage with the tests. The documentation has been updated and some examples have been streamlined to build more quickly and avoid timeout problems with CircleCI. We also added an optional CI on GPU for the master branch and approved PRs that can be used when a GPU runner is online.</p>
<p>Many other bugs and issues have been fixed and we want to thank all the contributors, old and new, who made this release possible. More details below.</p>
<section id="id21">
<h3>New features<a class="headerlink" href="#id21" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Added Bures Wasserstein distance in <code class="docutils literal notranslate"><span class="pre">ot.gaussian</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/">PR #</a>#428)</p></li>
<li><p>Added Generalized Wasserstein Barycenter solver + example (<a class="reference external" href="https://github.com/PythonOT/POT/pull/372">PR #372</a>), fixed graphical details on the example (PR #376)</p></li>
<li><p>Added Free Support Sinkhorn Barycenter + example (<a class="reference external" href="https://github.com/PythonOT/POT/pull/387">PR #387</a>)</p></li>
<li><p>New API for OT solver using function <code class="docutils literal notranslate"><span class="pre">ot.solve</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/388">PR #388</a>)</p></li>
<li><p>Backend version of <code class="docutils literal notranslate"><span class="pre">ot.partial</span></code> and <code class="docutils literal notranslate"><span class="pre">ot.smooth</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/388">PR #388</a>)</p></li>
<li><p>Added argument for warmstart of dual vectors in Sinkhorn-based methods in <code class="docutils literal notranslate"><span class="pre">ot.bregman</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/437">PR #437</a>)</p></li>
</ul>
</section>
<section id="id22">
<h3>Closed issues<a class="headerlink" href="#id22" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Fixed an issue with the documentation gallery sections (<a class="reference external" href="https://github.com/PythonOT/POT/pull/395">PR #395</a>)</p></li>
<li><p>Fixed an issue where sinkhorn divergence did not have a gradients (<a class="reference external" href="https://github.com/PythonOT/POT/issues/393">Issue #393</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/394">PR #394</a>)</p></li>
<li><p>Fixed an issue where we could not ask TorchBackend to place a random tensor on GPU
(<a class="reference external" href="https://github.com/PythonOT/POT/issues/371">Issue #371</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/373">PR #373</a>)</p></li>
<li><p>Fixed an issue where Sinkhorn solver assumed a symmetric cost matrix (<a class="reference external" href="https://github.com/PythonOT/POT/issues/374">Issue #374</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/375">PR #375</a>)</p></li>
<li><p>Fixed an issue where hitting iteration limits would be reported to stderr by std::cerr regardless of Python’s stderr stream status (<a class="reference external" href="https://github.com/PythonOT/POT/pull/377">PR #377</a>)</p></li>
<li><p>Fixed an issue where the metric argument in ot.dist did not allow a callable parameter (<a class="reference external" href="https://github.com/PythonOT/POT/issues/378">Issue #378</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/379">PR #379</a>)</p></li>
<li><p>Fixed an issue where the max number of iterations in ot.emd was not allowed to go beyond 2^31 (<a class="reference external" href="https://github.com/PythonOT/POT/pull/380">PR #380</a>)</p></li>
<li><p>Fixed an issue where pointers would overflow in the EMD solver, returning an
incomplete transport plan above a certain size (slightly above 46k, its square being
roughly 2^31) (<a class="reference external" href="https://github.com/PythonOT/POT/pull/381">PR #381</a>)</p></li>
<li><p>Error raised when mass mismatch in emd2 (<a class="reference external" href="https://github.com/PythonOT/POT/pull/386">PR #386</a>)</p></li>
<li><p>Fixed an issue where a pytorch example would throw an error if executed on a GPU (<a class="reference external" href="https://github.com/PythonOT/POT/issues/389">Issue #389</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/391">PR #391</a>)</p></li>
<li><p>Added a work-around for scipy’s bug, where you cannot compute the Hamming distance with a “None” weight attribute. (<a class="reference external" href="https://github.com/PythonOT/POT/issues/400">Issue #400</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/402">PR #402</a>)</p></li>
<li><p>Fixed an issue where the doc could not be built due to some changes in matplotlib’s API (<a class="reference external" href="https://github.com/PythonOT/POT/issues/403">Issue #403</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/402">PR #402</a>)</p></li>
<li><p>Replaced Numpy C Compiler with Setuptools C Compiler due to deprecation issues (<a class="reference external" href="https://github.com/PythonOT/POT/issues/408">Issue #408</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/409">PR #409</a>)</p></li>
<li><p>Fixed weak optimal transport docstring (<a class="reference external" href="https://github.com/PythonOT/POT/issues/404">Issue #404</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/410">PR #410</a>)</p></li>
<li><p>Fixed error with parameter <code class="docutils literal notranslate"><span class="pre">log=True</span></code>for <code class="docutils literal notranslate"><span class="pre">SinkhornLpl1Transport</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/issues/412">Issue #412</a>,
<a class="reference external" href="https://github.com/PythonOT/POT/pull/413">PR #413</a>)</p></li>
<li><p>Fixed an issue about <code class="docutils literal notranslate"><span class="pre">warn</span></code> parameter in <code class="docutils literal notranslate"><span class="pre">sinkhorn2</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/417">PR #417</a>)</p></li>
<li><p>Fix an issue where the parameter <code class="docutils literal notranslate"><span class="pre">stopThr</span></code> in <code class="docutils literal notranslate"><span class="pre">empirical_sinkhorn_divergence</span></code> was rendered useless by subcalls
that explicitly specified <code class="docutils literal notranslate"><span class="pre">stopThr=1e-9</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/issues/421">Issue #421</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/422">PR #422</a>).</p></li>
<li><p>Fixed a bug breaking an example where we would try to make an array of arrays of different shapes (<a class="reference external" href="https://github.com/PythonOT/POT/issues/424">Issue #424</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/425">PR #425</a>)</p></li>
</ul>
</section>
</section>
<section id="id23">
<h2>0.8.2<a class="headerlink" href="#id23" title="Link to this heading"></a></h2>
<p>This releases introduces several new notable features. The less important
but most exiting one being that we now have a logo for the toolbox (color
and dark background) :</p>
<p><img alt="" src="https://pythonot.github.io/master/_images/logo.svg" /><img alt="" src="https://pythonot.github.io/master/_static/logo_dark.svg" /></p>
<p>This logo is generated using with matplotlib and using the solution of an OT
problem provided by POT (with <code class="docutils literal notranslate"><span class="pre">ot.emd</span></code>). Generating the logo can be done with a
simple python script also provided in the <a class="reference external" href="https://pythonot.github.io/auto_examples/others/plot_logo.html#sphx-glr-auto-examples-others-plot-logo-py">documentation gallery</a>.</p>
<p>New OT solvers include <a class="reference external" href="https://pythonot.github.io/gen_modules/ot.weak.html#ot.weak.weak_optimal_transport">Weak
OT</a>
and <a class="reference external" href="https://pythonot.github.io/gen_modules/ot.factored.html#ot.factored.factored_optimal_transport">OT with factored
coupling</a>
that can be used on large datasets. The <a class="reference external" href="https://pythonot.github.io/gen_modules/ot.unbalanced.html?highlight=mm_#ot.unbalanced.mm_unbalanced">Majorization Minimization</a> solvers for
non-regularized Unbalanced OT are now also available. We also now provide an
implementation of <a class="reference external" href="https://pythonot.github.io/gen_modules/ot.gromov.html#ot.gromov.gromov_wasserstein_linear_unmixing">GW and FGW unmixing</a> and <a class="reference external" href="https://pythonot.github.io/gen_modules/ot.gromov.html#ot.gromov.gromov_wasserstein_dictionary_learning">dictionary learning</a>. It is now
possible to use autodiff to solve entropic an quadratic regularized OT in the
dual for full or stochastic optimization thanks to the new functions to compute
the dual loss for <a class="reference external" href="https://pythonot.github.io/gen_modules/ot.stochastic.html#ot.stochastic.loss_dual_entropic">entropic</a> and <a class="reference external" href="https://pythonot.github.io/gen_modules/ot.stochastic.html#ot.stochastic.loss_dual_quadratic">quadratic</a> regularized OT and reconstruct the <a class="reference external" href="https://pythonot.github.io/gen_modules/ot.stochastic.html#ot.stochastic.plan_dual_entropic">OT
plan</a> on part or all of the data. They can be used for instance to solve OT
problems with stochastic gradient or for estimating the <a class="reference external" href="https://pythonot.github.io/auto_examples/backends/plot_stoch_continuous_ot_pytorch.html#sphx-glr-auto-examples-backends-plot-stoch-continuous-ot-pytorch-py">dual potentials as
neural networks</a>.</p>
<p>On the backend front, we now have backend compatible functions and classes in
the domain adaptation <a class="reference external" href="https://pythonot.github.io/gen_modules/ot.da.html#module-ot.da"><code class="docutils literal notranslate"><span class="pre">ot.da</span></code></a> and unbalanced OT <a class="reference external" href="https://pythonot.github.io/gen_modules/ot.unbalanced.html"><code class="docutils literal notranslate"><span class="pre">ot.unbalanced</span></code></a> modules. This
means that the DA classes can be used on tensors from all compatible backends.
The <a class="reference external" href="https://pythonot.github.io/gen_modules/ot.lp.html?highlight=free%20support#ot.lp.free_support_barycenter">free support Wasserstein barycenter</a> solver is now also backend compatible.</p>
<p>Finally we have worked on the documentation to provide an update of existing
examples in the gallery and and several new examples including <a class="reference external" href="https://pythonot.github.io/auto_examples/gromov/plot_gromov_wasserstein_dictionary_learning.html#sphx-glr-auto-examples-gromov-plot-gromov-wasserstein-dictionary-learning-py">GW dictionary
learning</a>
<a class="reference external" href="https://pythonot.github.io/auto_examples/others/plot_WeakOT_VS_OT.html#sphx-glr-auto-examples-others-plot-weakot-vs-ot-py">weak Optimal
Transport</a>,
<a class="reference external" href="https://pythonot.github.io/auto_examples/backends/plot_stoch_continuous_ot_pytorch.html#sphx-glr-auto-examples-backends-plot-stoch-continuous-ot-pytorch-py">NN based dual potentials
estimation</a>
and <a class="reference external" href="https://pythonot.github.io/auto_examples/others/plot_factored_coupling.html#sphx-glr-auto-examples-others-plot-factored-coupling-py">Factored coupling OT</a>.
.</p>
<section id="id24">
<h3>New features<a class="headerlink" href="#id24" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Remove deprecated <code class="docutils literal notranslate"><span class="pre">ot.gpu</span></code> submodule (<a class="reference external" href="https://github.com/PythonOT/POT/pull/361">PR #361</a>)</p></li>
<li><p>Update examples in the gallery (<a class="reference external" href="https://github.com/PythonOT/POT/pull/359">PR #359</a>)</p></li>
<li><p>Add stochastic loss and OT plan computation for regularized OT and
backend examples(<a class="reference external" href="https://github.com/PythonOT/POT/pull/360">PR #360</a>)</p></li>
<li><p>Implementation of factored OT with emd and sinkhorn (<a class="reference external" href="https://github.com/PythonOT/POT/pull/358">PR #358</a>)</p></li>
<li><p>A brand new logo for POT (<a class="reference external" href="https://github.com/PythonOT/POT/pull/357">PR #357</a>)</p></li>
<li><p>Better list of related examples in quick start guide with <code class="docutils literal notranslate"><span class="pre">minigallery</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/334">PR #334</a>)</p></li>
<li><p>Add optional log-domain Sinkhorn implementation in WDA to support smaller values
of the regularization parameter (<a class="reference external" href="https://github.com/PythonOT/POT/pull/336">PR #336</a>)</p></li>
<li><p>Backend implementation for <code class="docutils literal notranslate"><span class="pre">ot.lp.free_support_barycenter</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/340">PR #340</a>)</p></li>
<li><p>Add weak OT solver + example (<a class="reference external" href="https://github.com/PythonOT/POT/pull/341">PR #341</a>)</p></li>
<li><p>Add backend support for Domain Adaptation and Unbalanced solvers (<a class="reference external" href="https://github.com/PythonOT/POT/pull/343">PR #343</a>)</p></li>
<li><p>Add (F)GW linear dictionary learning solvers + example (<a class="reference external" href="https://github.com/PythonOT/POT/pull/319">PR #319</a>)</p></li>
<li><p>Add links to related PR and Issues in the doc release page (<a class="reference external" href="https://github.com/PythonOT/POT/pull/350">PR #350</a>)</p></li>
<li><p>Add new minimization-maximization algorithms for solving exact Unbalanced OT + example (<a class="reference external" href="https://github.com/PythonOT/POT/pull/362">PR #362</a>)</p></li>
</ul>
</section>
<section id="id25">
<h3>Closed issues<a class="headerlink" href="#id25" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Fix mass gradient of <code class="docutils literal notranslate"><span class="pre">ot.emd2</span></code> and <code class="docutils literal notranslate"><span class="pre">ot.gromov_wasserstein2</span></code> so that they are
centered (<a class="reference external" href="https://github.com/PythonOT/POT/issues/364">Issue #364</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/363">PR #363</a>)</p></li>
<li><p>Fix bug in instantiating an <code class="docutils literal notranslate"><span class="pre">autograd</span></code> function <code class="docutils literal notranslate"><span class="pre">ValFunction</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/issues/337">Issue #337</a>,
<a class="reference external" href="https://github.com/PythonOT/POT/pull/338">PR #338</a>)</p></li>
<li><p>Fix POT ABI compatibility with old and new numpy (<a class="reference external" href="https://github.com/PythonOT/POT/issues/346">Issue #346</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/349">PR #349</a>)</p></li>
<li><p>Warning when feeding integer cost matrix to EMD solver resulting in an integer transport plan (<a class="reference external" href="https://github.com/PythonOT/POT/issues/345">Issue #345</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/343">PR #343</a>)</p></li>
<li><p>Fix bug where gromov_wasserstein2 does not perform backpropagation with CUDA
tensors (<a class="reference external" href="https://github.com/PythonOT/POT/issues/351">Issue #351</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/352">PR #352</a>)</p></li>
</ul>
</section>
</section>
<section id="id26">
<h2>0.8.1.0<a class="headerlink" href="#id26" title="Link to this heading"></a></h2>
<p><em>December 2021</em></p>
<p>This is a bug fix release that will remove the <code class="docutils literal notranslate"><span class="pre">benchmarks</span></code> module form the
installation and correct the documentation generation.</p>
<section id="id27">
<h3>Closed issues<a class="headerlink" href="#id27" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Bug in documentation generation (tag VS master push, <a class="reference external" href="https://github.com/PythonOT/POT/pull/332">PR #332</a>)</p></li>
<li><p>Remove installation of the benchmarks in global namespace (<a class="reference external" href="https://github.com/PythonOT/POT/issues/331">Issue #331</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/333">PR #333</a>)</p></li>
</ul>
</section>
</section>
<section id="id28">
<h2>0.8.1<a class="headerlink" href="#id28" title="Link to this heading"></a></h2>
<p><em>December 2021</em></p>
<p>This release fixes several bugs and introduces two new backends: Cupy
and Tensorflow. Note that the tensorflow backend will work only when tensorflow
has enabled the Numpy behavior (for transpose that is not by default in
tensorflow). We also introduce a simple benchmark on CPU GPU for the sinkhorn
solver that will be provided in the
<a class="reference external" href="https://pythonot.github.io/gen_modules/ot.backend.html">backend</a> documentation.</p>
<p>This release also brings a few changes in dependencies and compatibility. First
we removed tests for Python 3.6 that will not be updated in the future.
Also note that POT now depends on Numpy (>= 1.20) because a recent change in ABI is making the
wheels non-compatible with older numpy versions. If you really need an older
numpy POT will work with no problems but you will need to build it from source.</p>
<p>As always we want to that the contributors who helped make POT better (and bug free).</p>
<section id="id29">
<h3>New features<a class="headerlink" href="#id29" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>New benchmark for sinkhorn solver on CPU/GPU and between backends (<a class="reference external" href="https://github.com/PythonOT/POT/pull/316">PR #316</a>)</p></li>
<li><p>New tensorflow backend (<a class="reference external" href="https://github.com/PythonOT/POT/pull/316">PR #316</a>)</p></li>
<li><p>New Cupy backend (<a class="reference external" href="https://github.com/PythonOT/POT/pull/315">PR #315</a>)</p></li>
<li><p>Documentation always up-to-date with README, RELEASES, CONTRIBUTING and
CODE_OF_CONDUCT files (<a class="reference external" href="https://github.com/PythonOT/POT/pull/316">PR #316</a>, PR #322).</p></li>
</ul>
</section>
<section id="id30">
<h3>Closed issues<a class="headerlink" href="#id30" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Fix bug in older Numpy ABI (<1.20) (<a class="reference external" href="https://github.com/PythonOT/POT/issues/308">Issue #308</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/326">PR #326</a>)</p></li>
<li><p>Fix bug in <code class="docutils literal notranslate"><span class="pre">ot.dist</span></code> function when non euclidean distance (<a class="reference external" href="https://github.com/PythonOT/POT/issues/305">Issue #305</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/306">PR #306</a>)</p></li>
<li><p>Fix gradient scaling for functions using <code class="docutils literal notranslate"><span class="pre">nx.set_gradients</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/issues/309">Issue #309</a>,
<a class="reference external" href="https://github.com/PythonOT/POT/pull/310">PR #310</a>)</p></li>
<li><p>Fix bug in generalized Conditional gradient solver and SinkhornL1L2
(<a class="reference external" href="https://github.com/PythonOT/POT/issues/311">Issue #311</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/313">PR #313</a>)</p></li>
<li><p>Fix log error in <code class="docutils literal notranslate"><span class="pre">gromov_barycenters</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/issues/317">Issue #317</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/3018">PR #3018</a>)</p></li>
</ul>
</section>
</section>
<section id="id31">
<h2>0.8.0<a class="headerlink" href="#id31" title="Link to this heading"></a></h2>
<p><em>November 2021</em></p>
<p>This new stable release introduces several important features.</p>
<p>First we now have
an OpenMP compatible exact ot solver in <code class="docutils literal notranslate"><span class="pre">ot.emd</span></code>. The OpenMP version is used
when the parameter <code class="docutils literal notranslate"><span class="pre">numThreads</span></code> is greater than one and can lead to nice
speedups on multi-core machines.</p>
<p>Second we have introduced a backend mechanism that allows to use standard POT
function seamlessly on Numpy, Pytorch and Jax arrays. Other backends are coming
but right now POT can be used seamlessly for training neural networks in
Pytorch. Notably we propose the first differentiable computation of the exact OT
loss with <code class="docutils literal notranslate"><span class="pre">ot.emd2</span></code> (can be differentiated w.r.t. both cost matrix and sample
weights), but also for the classical Sinkhorn loss with <code class="docutils literal notranslate"><span class="pre">ot.sinkhorn2</span></code>, the
Wasserstein distance in 1D with <code class="docutils literal notranslate"><span class="pre">ot.wasserstein_1d</span></code>, sliced Wasserstein with
<code class="docutils literal notranslate"><span class="pre">ot.sliced_wasserstein_distance</span></code> and Gromov-Wasserstein with <code class="docutils literal notranslate"><span class="pre">ot.gromov_wasserstein2</span></code>. Examples of how
this new feature can be used are now available in the documentation where the
Pytorch backend is used to estimate a <a class="reference external" href="https://PythonOT.github.io/auto_examples/backends/plot_unmix_optim_torch.html">minimal Wasserstein
estimator</a>,
a <a class="reference external" href="https://PythonOT.github.io/auto_examples/backends/plot_wass2_gan_torch.html">Generative Network
(GAN)</a>,
for a <a class="reference external" href="https://PythonOT.github.io/auto_examples/backends/plot_sliced_wass_grad_flow_pytorch.html">sliced Wasserstein gradient
flow</a>
and <a class="reference external" href="https://PythonOT.github.io/auto_examples/backends/plot_optim_gromov_pytorch.html">optimizing the Gromov-Wassersein distance</a>. Note that the Jax backend is still in early development and quite
slow at the moment, we strongly recommend for Jax users to use the <a class="reference external" href="https://github.com/google-research/ott">OTT
toolbox</a> when possible.
As a result of this new feature,
the old <code class="docutils literal notranslate"><span class="pre">ot.gpu</span></code> submodule is now deprecated since GPU
implementations can be done using GPU arrays on the torch backends.</p>
<p>Other novel features include implementation for <a class="reference external" href="https://PythonOT.github.io/auto_examples/gromov/plot_gromov.html#compute-gw-with-a-scalable-stochastic-method-with-any-loss-function">Sampled Gromov Wasserstein and
Pointwise Gromov
Wasserstein</a>,
Sinkhorn in log space with <code class="docutils literal notranslate"><span class="pre">method='sinkhorn_log'</span></code>, <a class="reference external" href="https://PythonOT.github.io/gen_modules/ot.dr.html?highlight=robust#ot.dr.projection_robust_wasserstein">Projection Robust
Wasserstein</a>,
ans <a class="reference external" href="https://PythonOT.github.ioauto_examples/barycenters/plot_debiased_barycenter.html">deviased Sinkorn barycenters</a>.</p>
<p>This release will also simplify the installation process. We have now a
<code class="docutils literal notranslate"><span class="pre">pyproject.toml</span></code> that defines the build dependency and POT should now build even
when cython is not installed yet. Also we now provide pe-compiled wheels for
linux <code class="docutils literal notranslate"><span class="pre">aarch64</span></code> that is used on Raspberry PI and android phones and for MacOS on
ARM processors.</p>
<p>Finally POT was accepted for publication in the Journal of Machine Learning
Research (JMLR) open source software track and we ask the POT users to cite <a class="reference external" href="https://www.jmlr.org/papers/v22/20-451.html">this
paper</a> from now on. The documentation has been improved in particular by adding a
“Why OT?” section to the quick start guide and several new examples illustrating
the new features. The documentation now has two version : the stable version
<a class="reference external" href="https://pythonot.github.io/">https://pythonot.github.io/</a>
corresponding to the last release and the master version <a class="reference external" href="https://pythonot.github.io/master">https://pythonot.github.io/master</a> that corresponds to the
current master branch on GitHub.</p>
<p>As usual, we want to thank all the POT contributors (now 37 people have
contributed to the toolbox). But for this release we thank in particular Nathan
Cassereau and Kamel Guerda from the AI support team at
<a class="reference external" href="http://www.idris.fr/">IDRIS</a> for their support to the development of the
backend and OpenMP implementations.</p>
<section id="id32">
<h3>New features<a class="headerlink" href="#id32" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>OpenMP support for exact OT solvers (<a class="reference external" href="https://github.com/PythonOT/POT/pull/260">PR #260</a>)</p></li>
<li><p>Backend for running POT in numpy/torch + exact solver (<a class="reference external" href="https://github.com/PythonOT/POT/pull/249">PR #249</a>)</p></li>
<li><p>Backend implementation of most functions in <code class="docutils literal notranslate"><span class="pre">ot.bregman</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/280">PR #280</a>)</p></li>
<li><p>Backend implementation of most functions in <code class="docutils literal notranslate"><span class="pre">ot.optim</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/282">PR #282</a>)</p></li>
<li><p>Backend implementation of most functions in <code class="docutils literal notranslate"><span class="pre">ot.gromov</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/294">PR #294</a>, PR #302)</p></li>
<li><p>Test for arrays of different type and device (CPU/GPU) (<a class="reference external" href="https://github.com/PythonOT/POT/pull/304">PR #304</a>, #303)</p></li>
<li><p>Implementation of Sinkhorn in log space with <code class="docutils literal notranslate"><span class="pre">method='sinkhorn_log'</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/290">PR #290</a>)</p></li>
<li><p>Implementation of regularization path for L2 Unbalanced OT (<a class="reference external" href="https://github.com/PythonOT/POT/pull/274">PR #274</a>)</p></li>
<li><p>Implementation of Projection Robust Wasserstein (<a class="reference external" href="https://github.com/PythonOT/POT/pull/267">PR #267</a>)</p></li>
<li><p>Implementation of Debiased Sinkhorn Barycenters (<a class="reference external" href="https://github.com/PythonOT/POT/pull/291">PR #291</a>)</p></li>
<li><p>Implementation of Sampled Gromov Wasserstein and Pointwise Gromov Wasserstein
(<a class="reference external" href="https://github.com/PythonOT/POT/pull/275">PR #275</a>)</p></li>
<li><p>Add <code class="docutils literal notranslate"><span class="pre">pyproject.toml</span></code> and build POT without installing cython first (<a class="reference external" href="https://github.com/PythonOT/POT/pull/293">PR #293</a>)</p></li>
<li><p>Lazy implementation in log space for sinkhorn on samples (<a class="reference external" href="https://github.com/PythonOT/POT/pull/259">PR #259</a>)</p></li>
<li><p>Documentation cleanup (<a class="reference external" href="https://github.com/PythonOT/POT/pull/298">PR #298</a>)</p></li>
<li><p>Two up-to-date documentations <a class="reference external" href="https://PythonOT.github.io/">for stable
release</a> and for <a class="reference external" href="https://pythonot.github.io/master/">master branch</a>.</p></li>
<li><p>Building wheels on ARM for Raspberry PI and smartphones (<a class="reference external" href="https://github.com/PythonOT/POT/pull/238">PR #238</a>)</p></li>
<li><p>Update build wheels to new version and new pythons (<a class="reference external" href="https://github.com/PythonOT/POT/pull/236">PR #236</a>, #253)</p></li>
<li><p>Implementation of sliced Wasserstein distance (<a class="reference external" href="https://github.com/PythonOT/POT/issues/202">Issue #202</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/203">PR #203</a>)</p></li>
<li><p>Add minimal build to CI and perform pep8 test separately (<a class="reference external" href="https://github.com/PythonOT/POT/pull/210">PR #210</a>)</p></li>
<li><p>Speedup of tests and return run time (<a class="reference external" href="https://github.com/PythonOT/POT/pull/262">PR #262</a>)</p></li>
<li><p>Add “Why OT” discussion to the documentation (<a class="reference external" href="https://github.com/PythonOT/POT/pull/220">PR #220</a>)</p></li>
<li><p>New introductory example to discrete OT in the documentation (<a class="reference external" href="https://github.com/PythonOT/POT/pull/191">PR #191</a>)</p></li>
<li><p>Add templates for Issues/PR on Github (PR#181)</p></li>
</ul>
</section>
<section id="id33">
<h3>Closed issues<a class="headerlink" href="#id33" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Debug Memory leak in GAN example (#254)</p></li>
<li><p>DEbug GPU bug (<a class="reference external" href="https://github.com/PythonOT/POT/issues/284">Issue #284</a>, #287, <a class="reference external" href="https://github.com/PythonOT/POT/pull/288">PR #288</a>)</p></li>
<li><p>set_gradients method for JAX backend (<a class="reference external" href="https://github.com/PythonOT/POT/pull/278">PR #278</a>)</p></li>
<li><p>Quicker GAN example for CircleCI build (<a class="reference external" href="https://github.com/PythonOT/POT/pull/258">PR #258</a>)</p></li>
<li><p>Better formatting in Readme (<a class="reference external" href="https://github.com/PythonOT/POT/pull/234">PR #234</a>)</p></li>
<li><p>Debug CI tests (<a class="reference external" href="https://github.com/PythonOT/POT/pull/240">PR #240</a>, #241, #242)</p></li>
<li><p>Bug in Partial OT solver dummy points (<a class="reference external" href="https://github.com/PythonOT/POT/pull/215">PR #215</a>)</p></li>
<li><p>Bug when Armijo linesearch (<a class="reference external" href="https://github.com/PythonOT/POT/issues/184">Issue #184</a>, #198, #281, <a class="reference external" href="https://github.com/PythonOT/POT/pull/189">PR #189</a>, #199, #286)</p></li>
<li><p>Bug Barycenter Sinkhorn (Issue 134, <a class="reference external" href="https://github.com/PythonOT/POT/pull/195">PR #195</a>)</p></li>
<li><p>Infeasible solution in exact OT (Issues #126,#93, <a class="reference external" href="https://github.com/PythonOT/POT/pull/217">PR #217</a>)</p></li>
<li><p>Doc for SUpport Barycenters (<a class="reference external" href="https://github.com/PythonOT/POT/issues/200">Issue #200</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/201">PR #201</a>)</p></li>
<li><p>Fix labels transport in BaseTransport (<a class="reference external" href="https://github.com/PythonOT/POT/issues/207">Issue #207</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/208">PR #208</a>)</p></li>
<li><p>Bug in <code class="docutils literal notranslate"><span class="pre">emd_1d</span></code>, non respected bounds (<a class="reference external" href="https://github.com/PythonOT/POT/issues/169">Issue #169</a>, <a class="reference external" href="https://github.com/PythonOT/POT/pull/170">PR #170</a>)</p></li>
<li><p>Removed Python 2.7 support and update codecov file (<a class="reference external" href="https://github.com/PythonOT/POT/pull/178">PR #178</a>)</p></li>
<li><p>Add normalization for WDA and test it (<a class="reference external" href="https://github.com/PythonOT/POT/pull/172">PR #172</a>, #296)</p></li>
<li><p>Cleanup code for new version of <code class="docutils literal notranslate"><span class="pre">flake8</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/176">PR #176</a>)</p></li>
<li><p>Fixed requirements in <code class="docutils literal notranslate"><span class="pre">setup.py</span></code> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/174">PR #174</a>)</p></li>
<li><p>Removed specific MacOS flags (<a class="reference external" href="https://github.com/PythonOT/POT/pull/175">PR #175</a>)</p></li>
</ul>
</section>
</section>
<section id="id34">
<h2>0.7.0<a class="headerlink" href="#id34" title="Link to this heading"></a></h2>
<p><em>May 2020</em></p>
<p>This is the new stable release for POT. We made a lot of changes in the
documentation and added several new features such as Partial OT, Unbalanced and
Multi Sources OT Domain Adaptation and several bug fixes. One important change
is that we have created the GitHub organization
<a class="reference external" href="https://github.com/PythonOT">PythonOT</a> that now owns the main POT repository
<a class="reference external" href="https://github.com/PythonOT/POT">https://github.com/PythonOT/POT</a> and the
repository for the new documentation is now hosted at
<a class="reference external" href="https://PythonOT.github.io/">https://PythonOT.github.io/</a>.</p>
<p>This is the first release where the Python 2.7 tests have been removed. Most of
the toolbox should still work but we do not offer support for Python 2.7 and
will close related Issues.</p>
<p>A lot of changes have been done to the documentation that is now hosted on
<a class="reference external" href="https://PythonOT.github.io/">https://PythonOT.github.io/</a> instead of
readthedocs. It was a hard choice but readthedocs did not allow us to run
sphinx-gallery to update our beautiful examples and it was a huge amount of work
to maintain. The documentation is now automatically compiled and updated on
merge. We also removed the notebooks from the repository for space reason and
also because they are all available in the <a class="reference external" href="https://pythonot.github.io/auto_examples/index.html">example
gallery</a>. Note that now the
output of the documentation build for each commit in the PR is available to
check that the doc builds correctly before merging which was not possible with
readthedocs.</p>
<p>The CI framework has also been changed with a move from Travis to Github Action
which allows to get faster tests on Windows, MacOS and Linux. We also now report
our coverage on <a class="reference external" href="https://codecov.io/gh/PythonOT/POT">Codecov.io</a> and we have a
reasonable 92% coverage. We also now generate wheels for a number of OS and
Python versions at each merge in the master branch. They are available as
outputs of this
<a class="reference external" href="https://github.com/PythonOT/POT/actions?query=workflow%3A%22Build+dist+and+wheels%22">action</a>.
This will allow simpler multi-platform releases from now on.</p>
<p>In terms of new features we now have <a class="reference external" href="https://pythonot.github.io/gen_modules/ot.da.html#ot.da.UnbalancedSinkhornTransport">OTDA Classes for unbalanced
OT</a>,
a new Domain adaptation class form <a class="reference external" href="https://pythonot.github.io/auto_examples/domain-adaptation/plot_otda_jcpot.html#sphx-glr-auto-examples-domain-adaptation-plot-otda-jcpot-py">multi domain problems
(JCPOT)</a>,
and several solvers to solve the <a class="reference external" href="https://pythonot.github.io/auto_examples/unbalanced-partial/plot_partial_wass_and_gromov.html#sphx-glr-auto-examples-unbalanced-partial-plot-partial-wass-and-gromov-py">Partial Optimal
Transport</a>
problems.</p>
<p>This release is also the moment to thank all the POT contributors (old and new)
for helping making POT such a nice toolbox. A lot of changes (also in the API)
are coming for the next versions.</p>
<section id="features">
<h3>Features<a class="headerlink" href="#features" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>New documentation on <a class="reference external" href="https://PythonOT.github.io/">https://PythonOT.github.io/</a> (<a class="reference external" href="https://github.com/PythonOT/POT/pull/160">PR #160</a>, PR #143, PR #144)</p></li>
<li><p>Documentation build on CircleCI with sphinx-gallery (<a class="reference external" href="https://github.com/PythonOT/POT/pull/145">PR #145</a>,PR #146, #155)</p></li>
<li><p>Run sphinx gallery in CI (<a class="reference external" href="https://github.com/PythonOT/POT/pull/146">PR #146</a>)</p></li>
<li><p>Remove notebooks from repo because available in doc (<a class="reference external" href="https://github.com/PythonOT/POT/pull/156">PR #156</a>)</p></li>
<li><p>Build wheels in CI (#157)</p></li>
<li><p>Move from travis to GitHub Action for Windows, MacOS and Linux (<a class="reference external" href="https://github.com/PythonOT/POT/pull/148">PR #148</a>, PR #150)</p></li>
<li><p>Partial Optimal Transport (PR#141 and <a class="reference external" href="https://github.com/PythonOT/POT/pull/142">PR #142</a>)</p></li>
<li><p>Laplace regularized OTDA (<a class="reference external" href="https://github.com/PythonOT/POT/pull/140">PR #140</a>)</p></li>
<li><p>Multi source DA with target shift (<a class="reference external" href="https://github.com/PythonOT/POT/pull/137">PR #137</a>)</p></li>
<li><p>Screenkhorn algorithm (<a class="reference external" href="https://github.com/PythonOT/POT/pull/121">PR #121</a>)</p></li>
</ul>
</section>
<section id="id35">
<h3>Closed issues<a class="headerlink" href="#id35" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Add JMLR paper to the readme and Mathieu Blondel to the Acknoledgments (PR
#231, #232)</p></li>
<li><p>Bug in Unbalanced OT example (<a class="reference external" href="https://github.com/PythonOT/POT/issues/127">Issue #127</a>)</p></li>
<li><p>Clean Cython output when calling setup.py clean (<a class="reference external" href="https://github.com/PythonOT/POT/issues/122">Issue #122</a>)</p></li>
<li><p>Various Macosx compilation problems (<a class="reference external" href="https://github.com/PythonOT/POT/issues/113">Issue #113</a>, Issue #118, PR#130)</p></li>
<li><p>EMD dimension mismatch (<a class="reference external" href="https://github.com/PythonOT/POT/issues/114">Issue #114</a>, Fixed in <a class="reference external" href="https://github.com/PythonOT/POT/pull/116">PR #116</a>)</p></li>
<li><p>2D barycenter bug for non square images (<a class="reference external" href="https://github.com/PythonOT/POT/issues/124">Issue #124</a>, fixed in <a class="reference external" href="https://github.com/PythonOT/POT/pull/132">PR #132</a>)</p></li>
<li><p>Bad value in EMD 1D (<a class="reference external" href="https://github.com/PythonOT/POT/issues/138">Issue #138</a>, fixed in <a class="reference external" href="https://github.com/PythonOT/POT/pull/139">PR #139</a>)</p></li>
<li><p>Log bugs for Gromov-Wassertein solver (<a class="reference external" href="https://github.com/PythonOT/POT/issues/107">Issue #107</a>, fixed in <a class="reference external" href="https://github.com/PythonOT/POT/pull/108">PR #108</a>)</p></li>
<li><p>Weight issues in barycenter function (<a class="reference external" href="https://github.com/PythonOT/POT/pull/106">PR #106</a>)</p></li>
</ul>
</section>
</section>
<section id="id36">
<h2>0.6.0<a class="headerlink" href="#id36" title="Link to this heading"></a></h2>
<p><em>July 2019</em></p>
<p>This is the first official stable release of POT and this means a jump to 0.6!
The library has been used in
the wild for a while now and we have reached a state where a lot of fundamental
OT solvers are available and tested. It has been quite stable in the last months
but kept the beta flag in its Pypi classifiers until now.</p>
<p>Note that this release will be the last one supporting officially Python 2.7 (See
https://python3statement.org/ for more reasons). For next release we will keep
the travis tests for Python 2 but will make them non necessary for merge in 2020.</p>
<p>The features are never complete in a toolbox designed for solving mathematical
problems and research but with the new contributions we now implement algorithms
and solvers from 24 scientific papers (listed in the README.md file). New
features include a direct implementation of the <a class="reference external" href="https://pot.readthedocs.io/en/latest/all.html#ot.bregman.empirical_sinkhorn_divergence">empirical Sinkhorn
divergence</a>,
a new efficient (Cython implementation) solver for <a class="reference external" href="https://pot.readthedocs.io/en/latest/all.html#ot.lp.emd_1d">EMD in
1D</a> and
corresponding <a class="reference external" href="https://pot.readthedocs.io/en/latest/all.html#ot.lp.wasserstein_1d">Wasserstein
1D</a>. We now
also have implementations for <a class="reference external" href="https://github.com/rflamary/POT/blob/master/notebooks/plot_UOT_1D.ipynb">Unbalanced
OT</a> and
a solver for <a class="reference external" href="https://github.com/rflamary/POT/blob/master/notebooks/plot_UOT_barycenter_1D.ipynb">Unbalanced OT
barycenters</a>.
A new variant of Gromov-Wasserstein divergence called <a class="reference external" href="https://pot.readthedocs.io/en/latest/all.html?highlight=fused_#ot.gromov.fused_gromov_wasserstein">Fused
Gromov-Wasserstein</a>
has been also contributed with exemples of use on <a class="reference external" href="https://github.com/rflamary/POT/blob/master/notebooks/plot_fgw.ipynb">structured
data</a> and
computing <a class="reference external" href="https://github.com/rflamary/POT/blob/master/notebooks/plot_barycenter_fgw.ipynb">barycenters of labeld
graphs</a>.</p>
<p>A lot of work has been done on the documentation with several new
examples corresponding to the new features and a lot of corrections for the
docstrings. But the most visible change is a new
<a class="reference external" href="https://pot.readthedocs.io/en/latest/quickstart.html">quick start guide</a> for
POT that gives several pointers about which function or classes allow to solve which
specific OT problem. When possible a link is provided to relevant examples.</p>
<p>We will also provide with this release some pre-compiled Python wheels for Linux
64bit on
github and pip. This will simplify the install process that before required a C
compiler and numpy/cython already installed.</p>
<p>Finally we would like to acknowledge and thank the numerous contributors of POT
that has helped in the past build the foundation and are still contributing to
bring new features and solvers to the library.</p>
<section id="id37">
<h3>Features<a class="headerlink" href="#id37" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p>Add compiled manylinux 64bits wheels to pip releases (<a class="reference external" href="https://github.com/PythonOT/POT/pull/91">PR #91</a>)</p></li>
<li><p>Add quick start guide (<a class="reference external" href="https://github.com/PythonOT/POT/pull/88">PR #88</a>)</p></li>
<li><p>Make doctest work on travis (<a class="reference external" href="https://github.com/PythonOT/POT/pull/90">PR #90</a>)</p></li>
<li><p>Update documentation (<a class="reference external" href="https://github.com/PythonOT/POT/pull/79">PR #79</a>, PR #84)</p></li>
<li><p>Solver for EMD in 1D (<a class="reference external" href="https://github.com/PythonOT/POT/pull/89">PR #89</a>)</p></li>
<li><p>Solvers for regularized unbalanced OT (<a class="reference external" href="https://github.com/PythonOT/POT/pull/87">PR #87</a>, PR#99)</p></li>
<li><p>Solver for Fused Gromov-Wasserstein (<a class="reference external" href="https://github.com/PythonOT/POT/pull/86">PR #86</a>)</p></li>
<li><p>Add empirical Sinkhorn and empirical Sinkhorn divergences (<a class="reference external" href="https://github.com/PythonOT/POT/pull/80">PR #80</a>)</p></li>
</ul>
</section>
<section id="id38">
<h3>Closed issues<a class="headerlink" href="#id38" title="Link to this heading"></a></h3>
<ul class="simple">
<li><p><a class="reference external" href="https://github.com/PythonOT/POT/issues/59">Issue #59</a> fail when using “pip install POT” (new details in doc+ hopefully
wheels)</p></li>
<li><p><a class="reference external" href="https://github.com/PythonOT/POT/issues/85">Issue #85</a> Cannot run gpu modules</p></li>
<li><p><a class="reference external" href="https://github.com/PythonOT/POT/issues/75">Issue #75</a> Greenkhorn do not return log (solved in <a class="reference external" href="https://github.com/PythonOT/POT/pull/76">PR #76</a>)</p></li>
<li><p><a class="reference external" href="https://github.com/PythonOT/POT/issues/82">Issue #82</a> Gromov-Wasserstein fails when the cost matrices are slightly different</p></li>
<li><p><a class="reference external" href="https://github.com/PythonOT/POT/issues/72">Issue #72</a> Macosx build problem</p></li>
</ul>
</section>
</section>
<section id="id39">
<h2>0.5.0<a class="headerlink" href="#id39" title="Link to this heading"></a></h2>
<p><em>Sep 2018</em></p>