/
TestMiscOpValidation.java
2305 lines (1813 loc) · 86.9 KB
/
TestMiscOpValidation.java
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
/*
* ******************************************************************************
* *
* *
* * This program and the accompanying materials are made available under the
* * terms of the Apache License, Version 2.0 which is available at
* * https://www.apache.org/licenses/LICENSE-2.0.
* *
* * See the NOTICE file distributed with this work for additional
* * information regarding copyright ownership.
* * Unless required by applicable law or agreed to in writing, software
* * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
* * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
* * License for the specific language governing permissions and limitations
* * under the License.
* *
* * SPDX-License-Identifier: Apache-2.0
* *****************************************************************************
*/
package org.eclipse.deeplearning4j.nd4j.autodiff.opvalidation;
import lombok.extern.slf4j.Slf4j;
import lombok.val;
import org.datavec.api.records.reader.RecordReader;
import org.datavec.api.records.reader.impl.collection.CollectionRecordReader;
import org.datavec.api.writable.IntWritable;
import org.deeplearning4j.datasets.datavec.RecordReaderDataSetIterator;
import org.junit.jupiter.api.Disabled;
import org.junit.jupiter.api.Tag;
import org.junit.jupiter.params.ParameterizedTest;
import org.junit.jupiter.params.provider.MethodSource;
import org.mockito.internal.matchers.Same;
import org.nd4j.autodiff.samediff.SDIndex;
import org.nd4j.autodiff.samediff.SDVariable;
import org.nd4j.autodiff.samediff.SameDiff;
import org.nd4j.autodiff.samediff.TrainingConfig;
import org.nd4j.autodiff.validation.OpTestCase;
import org.nd4j.autodiff.validation.OpValidation;
import org.nd4j.autodiff.validation.TestCase;
import org.nd4j.common.base.Preconditions;
import org.nd4j.common.tests.tags.TagNames;
import org.nd4j.enums.PartitionMode;
import org.nd4j.linalg.api.blas.params.MMulTranspose;
import org.nd4j.linalg.api.buffer.DataType;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.api.ops.CustomOp;
import org.nd4j.linalg.api.ops.DynamicCustomOp;
import org.nd4j.linalg.api.ops.custom.Digamma;
import org.nd4j.linalg.api.ops.custom.DivideNoNan;
import org.nd4j.linalg.api.ops.custom.Flatten;
import org.nd4j.linalg.api.ops.custom.FusedBatchNorm;
import org.nd4j.linalg.api.ops.custom.Igamma;
import org.nd4j.linalg.api.ops.custom.Igammac;
import org.nd4j.linalg.api.ops.custom.Lgamma;
import org.nd4j.linalg.api.ops.custom.Lu;
import org.nd4j.linalg.api.ops.custom.MatrixBandPart;
import org.nd4j.linalg.api.ops.custom.Polygamma;
import org.nd4j.linalg.api.ops.custom.Roll;
import org.nd4j.linalg.api.ops.custom.TriangularSolve;
import org.nd4j.linalg.api.ops.impl.broadcast.BiasAdd;
import org.nd4j.linalg.api.ops.impl.broadcast.BiasAddGrad;
import org.nd4j.linalg.api.ops.impl.controlflow.compat.StopGradient;
import org.nd4j.linalg.api.ops.impl.reduce.Mmul;
import org.nd4j.linalg.api.ops.impl.shape.DiagPart;
import org.nd4j.linalg.api.ops.impl.shape.OneHot;
import org.nd4j.linalg.api.ops.impl.shape.ZerosLike;
import org.nd4j.linalg.api.ops.impl.transforms.CheckNumerics;
import org.nd4j.linalg.api.ops.impl.transforms.clip.ClipByNorm;
import org.nd4j.linalg.api.ops.impl.transforms.custom.CumProd;
import org.nd4j.linalg.api.ops.impl.transforms.custom.CumSum;
import org.nd4j.linalg.api.ops.impl.transforms.custom.Fill;
import org.nd4j.linalg.api.ops.impl.transforms.pairwise.arithmetic.FloorDivOp;
import org.nd4j.linalg.api.ops.impl.transforms.pairwise.arithmetic.FloorModOp;
import org.nd4j.linalg.api.shape.LongShapeDescriptor;
import org.nd4j.linalg.dataset.DataSet;
import org.nd4j.linalg.dataset.api.iterator.DataSetIterator;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.factory.Nd4jBackend;
import org.nd4j.linalg.indexing.NDArrayIndex;
import org.nd4j.linalg.learning.config.Adam;
import org.nd4j.linalg.ops.transforms.Transforms;
import org.nd4j.common.primitives.Triple;
import org.nd4j.common.util.ArrayUtil;
import org.nd4j.weightinit.impl.XavierInitScheme;
import org.nd4j.weightinit.impl.ZeroInitScheme;
import java.util.*;
import java.util.stream.Collectors;
import static org.junit.jupiter.api.Assertions.*;
import static org.nd4j.linalg.api.buffer.DataType.FLOAT;
import static org.nd4j.linalg.api.buffer.DataType.INT32;
@Slf4j
@Tag(TagNames.SAMEDIFF)
public class TestMiscOpValidation extends BaseOpValidation {
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testGradientAutoBroadcast1(Nd4jBackend backend) {
Nd4j.getRandom().setSeed(12345);
List<String> failed = new ArrayList<>();
for (int dim_sz1 : new int[]{0, 1, 2}) {
int[] in2Shape = {3, 4, 5};
in2Shape[dim_sz1] = 1;
for (int i = 0; i < 8; i++) {
SameDiff sd = SameDiff.create();
SDVariable in3 = sd.var("in3", Nd4j.rand(new int[]{3, 4, 5}));
SDVariable in2 = sd.var("in2", in2Shape);
SDVariable bcOp;
String name;
switch (i) {
case 0:
bcOp = in3.add(in2);
name = "add";
break;
case 1:
bcOp = in3.sub(in2);
name = "sub";
break;
case 2:
bcOp = in3.mul(in2);
name = "mul";
break;
case 3:
bcOp = in3.div(in2);
name = "div";
break;
case 4:
bcOp = in3.rsub(in2);
name = "rsub";
break;
case 5:
bcOp = in3.rdiv(in2);
name = "rdiv";
break;
case 6:
//bcOp = sd.scalarFloorDiv(in3, in2);
bcOp = new FloorDivOp(sd, in3, in2).outputVariable();
name = "floordiv";
break;
case 7:
//bcOp = sd.scalarFloorMod(in3, in2);
bcOp = new FloorModOp(sd, in3, in2).outputVariable();
name = "floormod";
break;
default:
throw new RuntimeException();
}
SDVariable outVar = sd.sum(bcOp);
String msg = "(test " + i + ": " + name + ", dimension=" + dim_sz1 + ")";
log.info("*** Starting test: " + msg);
INDArray in3Arr = Nd4j.randn(new int[]{3, 4, 5}).muli(100);
INDArray in2Arr = Nd4j.randn(in2Shape).muli(100);
sd.associateArrayWithVariable(in3Arr, in3);
sd.associateArrayWithVariable(in2Arr, in2);
TestCase tc = new TestCase(sd);
String error = OpValidation.validate(tc);
if(error != null){
failed.add(name);
}
}
}
assertEquals(0, failed.size(),"Failed: " + failed);
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testGradientAutoBroadcast2(Nd4jBackend backend) {
Nd4j.getRandom().setSeed(12345);
List<String> failed = new ArrayList<>();
for (int[] dim_sz1s : new int[][]{{0, 1}, {0, 2}, {1, 2}, {0, 1, 2}}) {
long[] otherShape = {3, 4, 5};
otherShape[dim_sz1s[0]] = 1;
otherShape[dim_sz1s[1]] = 1;
if (dim_sz1s.length == 3) {
otherShape[dim_sz1s[2]] = 1;
}
//note that we disable FloorModBP here
//gradient checks for this one fail despite grad being correct
for (int i = 0; i < 7; i++) {
SameDiff sd = SameDiff.create();
SDVariable in3 = sd.var("in3", DataType.DOUBLE, 3, 4, 5);
SDVariable in2 = sd.var("inToBc", DataType.DOUBLE, otherShape);
String name;
SDVariable bcOp;
switch (i) {
case 0:
bcOp = in3.add(in2);
name = "add";
break;
case 1:
bcOp = in3.sub(in2);
name = "sub";
break;
case 2:
bcOp = in3.mul(in2);
name = "mul";
break;
case 3:
bcOp = in3.div(in2);
name = "div";
break;
case 4:
bcOp = in3.rsub(in2);
name = "rsub";
break;
case 5:
bcOp = in3.rdiv(in2);
name = "rdiv";
break;
case 6:
//bcOp = sd.scalarFloorDiv(in3, in2);
bcOp = new FloorDivOp(sd, in3, in2).outputVariable();
name = "floordiv";
break;
case 7:
//bcOp = sd.scalarFloorMod(in3, in2);
bcOp = new FloorModOp(sd, in3, in2).outputVariable();
name = "floormod";
break;
default:
throw new RuntimeException();
}
SDVariable outVar = sd.sum(bcOp);
String msg = "(test " + i + ": " + name + ", dimensions=" + Arrays.toString(dim_sz1s) + ")";
log.info("*** Starting test: " + msg);
INDArray in3Arr = Nd4j.randn(DataType.DOUBLE, 3, 4, 5).muli(100);
INDArray in2Arr = Nd4j.randn(DataType.DOUBLE, otherShape).muli(100);
sd.associateArrayWithVariable(in3Arr, in3);
sd.associateArrayWithVariable(in2Arr, in2);
TestCase tc = new TestCase(sd);
String error = OpValidation.validate(tc);
if(error != null){
failed.add(name);
}
}
}
assertEquals(0, failed.size(),"Failed: " + failed);
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testGradientAutoBroadcast3(Nd4jBackend backend) {
//These tests: output size > input sizes
Nd4j.getRandom().setSeed(12345);
List<String> failed = new ArrayList<>();
//Test cases: in1Shape, in2Shape, shapeOf(op(in1,in2))
List<Triple<long[], long[], long[]>> testCases = new ArrayList<>();
testCases.add(new Triple<>(new long[]{3, 1}, new long[]{1, 4}, new long[]{3, 4}));
testCases.add(new Triple<>(new long[]{3, 1}, new long[]{3, 4}, new long[]{3, 4}));
testCases.add(new Triple<>(new long[]{3, 4}, new long[]{1, 4}, new long[]{3, 4}));
testCases.add(new Triple<>(new long[]{3, 4, 1}, new long[]{1, 1, 5}, new long[]{3, 4, 5}));
testCases.add(new Triple<>(new long[]{3, 4, 1}, new long[]{3, 1, 5}, new long[]{3, 4, 5}));
testCases.add(new Triple<>(new long[]{3, 1, 5}, new long[]{1, 4, 1}, new long[]{3, 4, 5}));
testCases.add(new Triple<>(new long[]{3, 1, 5}, new long[]{1, 4, 5}, new long[]{3, 4, 5}));
testCases.add(new Triple<>(new long[]{3, 1, 5}, new long[]{3, 4, 5}, new long[]{3, 4, 5}));
testCases.add(new Triple<>(new long[]{3, 1, 1, 1}, new long[]{1, 4, 5, 6}, new long[]{3, 4, 5, 6}));
testCases.add(new Triple<>(new long[]{1, 1, 1, 6}, new long[]{3, 4, 5, 6}, new long[]{3, 4, 5, 6}));
testCases.add(new Triple<>(new long[]{1, 4, 5, 1}, new long[]{3, 1, 1, 6}, new long[]{3, 4, 5, 6}));
testCases.add(new Triple<>(new long[]{1, 6}, new long[]{3, 4, 5, 1}, new long[]{3, 4, 5, 6}));
for (val p : testCases) {
for (int i = 0; i < 7; i++) {
SameDiff sd = SameDiff.create();
SDVariable in3 = sd.var("in1", DataType.DOUBLE, p.getFirst());
SDVariable in2 = sd.var("in2", DataType.DOUBLE, p.getSecond());
String name;
SDVariable bcOp;
switch (i) {
case 0:
bcOp = in3.add(in2);
name = "add";
break;
case 1:
bcOp = in3.sub(in2);
name = "sub";
break;
case 2:
bcOp = in3.mul(in2);
name = "mul";
break;
case 3:
bcOp = in3.div(in2);
name = "div";
break;
case 4:
bcOp = in3.rsub(in2);
name = "rsub";
break;
case 5:
bcOp = in3.rdiv(in2);
name = "rdiv";
break;
case 6:
bcOp = new FloorDivOp(sd, in3, in2).outputVariable();
name = "floordiv";
break;
case 7:
//Grad checks fail but grad is correct: https://github.com/eclipse/deeplearning4j/issues/5976
//bcOp = sd.scalarFloorMod(in3, in2);
bcOp = new FloorModOp(sd, in3, in2).outputVariable();
name = "floormod";
break;
default:
throw new RuntimeException();
}
SDVariable outVar = sd.sum(bcOp);
String msg = "(test " + i + ": " + name + ", array 1 size =" + Arrays.toString(p.getFirst())
+ ", array 2 size = " + Arrays.toString(p.getSecond()) + ")";
log.info("*** Starting test: " + msg);
INDArray in3Arr = Nd4j.rand(DataType.DOUBLE, p.getFirst()).muli(100);
INDArray in2Arr = Nd4j.rand(DataType.DOUBLE, p.getSecond()).muli(100);
sd.associateArrayWithVariable(in3Arr, in3);
sd.associateArrayWithVariable(in2Arr, in2);
TestCase tc = new TestCase(sd);
String error = OpValidation.validate(tc);
if(error != null){
failed.add(name + " " + i + " - " + error);
}
}
}
assertEquals(0, failed.size(),"Failed: " + failed);
}
@Override
public long getTimeoutMilliseconds() {
return Long.MAX_VALUE;
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testScatterOpGradients(Nd4jBackend backend) {
List<String> failed = new ArrayList<>();
for (int i = 0; i < 7; i++) {
Nd4j.getRandom().setSeed(12345);
SameDiff sd = SameDiff.create();
SDVariable in = sd.var("in", DataType.DOUBLE, 20, 10);
SDVariable indices = sd.var("indices", DataType.INT, new long[]{5});
SDVariable updates = sd.var("updates", DataType.DOUBLE, 5, 10);
in.setArray(Nd4j.rand(DataType.DOUBLE, 20, 10));
indices.setArray(Nd4j.create(new double[]{3, 4, 5, 10, 18}).castTo(DataType.INT));
updates.setArray(Nd4j.rand(DataType.DOUBLE, 5, 10).muli(2).subi(1));
SDVariable scatter;
String name;
switch (i) {
case 0:
scatter = sd.scatterAdd("s", in, indices, updates);
name = "scatterAdd";
break;
case 1:
scatter = sd.scatterSub("s", in, indices, updates);
name = "scatterSub";
break;
case 2:
scatter = sd.scatterMul("s", in, indices, updates);
name = "scatterMul";
break;
case 3:
scatter = sd.scatterDiv("s", in, indices, updates);
name = "scatterDiv";
break;
case 4:
/* scatter = sd.scatterUpdate("s", in, indices, updates);
name = "scatterUpdate";
break;*/
continue;
case 5:
scatter = sd.scatterMax("s", in, indices, updates);
name = "scatterMax";
break;
case 6:
scatter = sd.scatterMin("s", in, indices, updates);
name = "scatterMin";
break;
default:
throw new RuntimeException();
}
INDArray exp = in.getArr().dup();
int[] indicesInt = indices.getArr().dup().data().asInt();
for( int j=0; j<indicesInt.length; j++ ){
INDArray updateRow = updates.getArr().getRow(j);
INDArray destinationRow = exp.getRow(indicesInt[j]);
switch (i){
case 0:
destinationRow.addi(updateRow);
break;
case 1:
destinationRow.subi(updateRow);
break;
case 2:
destinationRow.muli(updateRow);
break;
case 3:
destinationRow.divi(updateRow);
break;
case 4:
destinationRow.assign(updateRow);
break;
case 5:
destinationRow.assign(Transforms.max(destinationRow, updateRow, true));
break;
case 6:
destinationRow.assign(Transforms.min(destinationRow, updateRow, true));
break;
default:
throw new RuntimeException();
}
}
SDVariable loss = sd.sum(scatter); //.standardDeviation(scatter, true); //.sum(scatter); //TODO stdev might be better here as gradients are non-symmetrical...
TestCase tc = new TestCase(sd)
.expected(scatter, exp)
.gradCheckSkipVariables(indices.name());
String error = OpValidation.validate(tc);
if(error != null){
failed.add(name);
}
}
assertEquals(0, failed.size(),failed.toString());
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testScatterUpdate(){
INDArray x = Nd4j.linspace(FLOAT, 1, 30, 1).reshape(10, 3);
INDArray updates = Nd4j.create(new float[][]{
{100, 101, 102},
{200, 201, 202}});
INDArray indices = Nd4j.createFromArray(2, 5);
INDArray exp = x.dup();
exp.putRow(2, updates.getRow(0));
exp.putRow(5, updates.getRow(1));
INDArray out = exp.ulike();
Nd4j.exec(DynamicCustomOp.builder("scatter_upd")
.addInputs(x, indices, updates)
.addOutputs(out)
.build());
assertEquals(exp, out);
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
@Tag(TagNames.LONG_TEST)
@Tag(TagNames.LARGE_RESOURCES)
public void testGatherGradient(Nd4jBackend backend) {
Nd4j.getRandom().setSeed(12345);
List<String> failed = new ArrayList<>();
for (int rank = 2; rank <= 3; rank++) {
for (int dim = 0; dim < rank; dim++) {
SameDiff sd = SameDiff.create();
int[] inShape;
if (rank == 2) {
inShape = new int[]{10, 10};
} else {
inShape = new int[]{10, 10, 10};
}
SDVariable in = sd.var("in", Nd4j.rand(DataType.DOUBLE, inShape));
SDVariable indices = sd.constant("indices", Nd4j.createFromArray(0, 3, 7));
INDArray gatherExp = null;
if(rank == 2){
int tadDim = dim == 0 ? 1 : 0; //Swap: pullRows dim is "tensor along dimension" vs. gather's "index is value for this dimension"
gatherExp = Nd4j.pullRows(in.getArr(), tadDim, new int[]{0,3,7});
}
SDVariable gather = sd.gather(in, indices, dim);
SDVariable loss = sd.standardDeviation("loss", gather, true, Integer.MAX_VALUE);
String msg = "rank=" + rank + " dim=" + dim;
TestCase tc = new TestCase(sd)
.testName(msg)
.gradCheckSkipVariables(indices.name());
if (gatherExp != null) {
tc.expected(gather, gatherExp);
}
String error = OpValidation.validate(tc);
if(error != null){
failed.add(msg + " - " + error);
}
}
}
assertEquals(0, failed.size(),failed.toString());
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testTrace(){
Nd4j.getRandom().setSeed(12345);
for( int[] inShape : new int[][]{{3,3}}){
INDArray in = Nd4j.rand(inShape);
SameDiff sd = SameDiff.create();
SDVariable i = sd.var("in", in);
SDVariable trace = sd.math().trace(i);
double exp = Nd4j.diag(in).sumNumber().doubleValue();
TestCase tc = new TestCase(sd)
.expected(trace, Nd4j.scalar(exp))
.testName(Arrays.toString(inShape));
String err = OpValidation.validate(tc);
assertNull(err);
}
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void gatherTest(Nd4jBackend backend) {
/**
* TODO: add tests for new indexing engine based on SDVariable
* TODO: add tests for pure point based indices derived from an SDVariable
*/
SameDiff sd = SameDiff.create();
long sequenceLength = 256;
int hiddenSize = 768;
var sequenceLengthSd = sd.constant(sequenceLength);
var hiddenSizeSd = sd.constant(hiddenSize);
var classesAmount = 10;
var tokenEmbeddingsMatrix = sd.var("tokenEmbeddingsMatrix", new XavierInitScheme('c', 30522, hiddenSize), FLOAT, 30522, hiddenSize);
var classificationWeights =
sd.var("classificationWeights", new XavierInitScheme('c', hiddenSize, classesAmount), FLOAT, hiddenSize, classesAmount);
var inputTokenVocabIndices = sd.placeHolder("inputTokenVocabIndices", INT32, -1, sequenceLength);
var batchTokenEmbeddings = sd.gather("batchTokenEmbeddings", tokenEmbeddingsMatrix, inputTokenVocabIndices, 0);
var batchSize = batchTokenEmbeddings.shape().get(SDIndex.point(0)).castTo("batchSize", INT32);
var flatBatchSizeSd = batchSize.mul(sequenceLengthSd);
var hiddenLayerInputShape = sd.stack("hiddenLayerInputShape", 0, flatBatchSizeSd, hiddenSizeSd);
var layerInput = batchTokenEmbeddings.reshape(hiddenLayerInputShape);
var labels = sd.placeHolder("labels", INT32, sequenceLength);
var logits = sd.nn().linear(layerInput, classificationWeights, sd.zero("bias", classesAmount).convertToVariable().castTo(FLOAT));
sd.loss().softmaxCrossEntropy(sd.oneHot(labels, classesAmount), logits, null);
TrainingConfig trainingConfig = TrainingConfig.builder()
.dataSetFeatureMapping("inputTokenVocabIndices")
.dataSetLabelMapping("labels")
.updater(new Adam())
.build();
sd.setTrainingConfig(trainingConfig);
TestCase tc = new TestCase(sd)
.testName("gather_grad")
.gradientCheck(true)
.gradCheckPrint(true)
.testFlatBufferSerialization(TestCase.TestSerialization.BOTH)
.placeholderValue("inputTokenVocabIndices",Nd4j.zeros(1,sequenceLength))
.placeholderValue("labels",Nd4j.ones(sequenceLength));
String err = OpValidation.validate(tc);
assertNull(err);
/* DataSet d = new DataSet();
d.setFeatures(Nd4j.zeros(1, sequenceLength));
d.setLabels(Nd4j.ones(sequenceLength));
sd.fit(d);*/
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testTensorGradTensorMmul(Nd4jBackend backend) {
Nd4j.getExecutioner().enableVerboseMode(true);
Nd4j.getExecutioner().enableDebugMode(true);
Nd4j.getRandom().setSeed(12345);
SameDiff sameDiff = SameDiff.create();
INDArray arr = Nd4j.rand(new long[]{2, 2, 2}).castTo(DataType.DOUBLE);
INDArray arr2 = Nd4j.rand(new long[]{2, 2, 2}).castTo(DataType.DOUBLE);
SDVariable x = sameDiff.var("x", arr);
SDVariable y = sameDiff.var("y", arr2);
SDVariable result = sameDiff.tensorMmul(x, y, new int[]{0}, new int[]{1},false,false,false);
assertArrayEquals(ArrayUtil.getTensorMmulShape(new long[]{2, 2, 2}, new long[]{2, 2, 2}, new int[][]{{0}, {1}}),
result.eval().shape());
assertEquals(16, sameDiff.numElements());
String err = OpValidation.validate(new TestCase(sameDiff));
assertNull(err);
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testMulGradient(Nd4jBackend backend) {
INDArray arr1 = Nd4j.linspace(1, 4, 4, DataType.DOUBLE).reshape(2, 2);
INDArray arr2 = Nd4j.linspace(1, 4, 4, DataType.DOUBLE).reshape(2, 2);
INDArray gradAssertion = Nd4j.ones(arr1.shape());
INDArray scalar = Nd4j.scalar(1.0);
INDArray aGradAssertion = Nd4j.create(new double[][]{
{1, 4},
{9, 16}
});
INDArray cGradAssertion = Nd4j.create(new double[][]{
{1, 2},
{3, 4}
});
INDArray wGradAssertion = Nd4j.create(new double[][]{
{2, 8},
{18, 32}
});
INDArray dGradAssertion = Nd4j.ones(2, 2);
SameDiff sameDiff = SameDiff.create();
SDVariable sdVariable = sameDiff.var("a", arr1);
SDVariable sdVariable1 = sameDiff.var("w", arr2);
SDVariable varMulPre = sdVariable.mul("c", sdVariable1);
SDVariable varMul = varMulPre.mul("d", sdVariable1);
SDVariable sum = sameDiff.sum("ret", varMul, Integer.MAX_VALUE);
Map<String,INDArray> m = sameDiff.outputAll(null);
Map<String,INDArray> gm = sameDiff.calculateGradients(null, m.keySet());
SDVariable finalResult = sameDiff.grad(sum.name());
SDVariable cGrad = sameDiff.grad(varMulPre.name());
SDVariable mulGradResult = sameDiff.grad(varMul.name());
SDVariable aGrad = sameDiff.grad(sdVariable.name());
SDVariable wGrad = sameDiff.grad(sdVariable1.name());
SDVariable dGrad = sameDiff.grad(varMul.name());
INDArray scalarGradTest = gm.get(sum.name());
assertEquals(scalar, scalarGradTest);
INDArray gradTest = mulGradResult.getArr();
assertEquals(gradAssertion, gradTest);
INDArray aGradTest = aGrad.getArr();
assertEquals(aGradAssertion, aGradTest);
INDArray cGradTest = cGrad.getArr();
assertEquals(cGradAssertion, cGradTest);
INDArray wGradTest = wGrad.getArr();
assertEquals(wGradAssertion, wGradTest);
INDArray dGradTest = dGrad.getArr();
assertEquals(dGradAssertion, dGradTest);
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testMmulGradients(Nd4jBackend backend){
int[] aShape = new int[]{2,3};
int[] bShape = new int[]{3,4};
List<String> failed = new ArrayList<>();
for( char aOrder : new char[]{'c', 'f'}) {
for (char bOrder : new char[]{'c', 'f'}) {
for (boolean transposeA : new boolean[]{false, true}) {
for (boolean transposeB : new boolean[]{false, true}) {
for (boolean transposeResult : new boolean[]{false, true}) { //https://github.com/eclipse/deeplearning4j/issues/5648
Nd4j.getRandom().setSeed(12345);
INDArray aArr = Nd4j.rand(DataType.DOUBLE, t(transposeA, aShape)).dup(aOrder);
INDArray bArr = Nd4j.rand(DataType.DOUBLE, t(transposeB, bShape)).dup(bOrder);
SameDiff sd = SameDiff.create();
SDVariable a = sd.var("a", aArr);
SDVariable b = sd.var("b", bArr);
SDVariable mmul = sd.mmul(a, b, transposeA, transposeB, transposeResult);
INDArray exp = (transposeA ? aArr.transpose() : aArr);
exp = exp.mmul(transposeB ? bArr.transpose() : bArr);
exp = (transposeResult ? exp.transpose() : exp);
SDVariable loss = mmul.std(true);
String name = aOrder + "," + bOrder + ",tA=" + transposeA + ",tB=" + transposeB +
",tRes=" + transposeResult;
TestCase tc = new TestCase(sd).testName(name)
.expected(mmul, exp);
String err = OpValidation.validate(tc, true);
if(err != null)
failed.add(err);
}
}
}
}
}
assertEquals(0, failed.size(),failed.toString());
}
private static int[] t(boolean transpose, int[] orig){
if(!transpose)
return orig;
return new int[]{orig[1], orig[0]};
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testBatchMmulBasic(Nd4jBackend backend) {
int M = 5;
int N = 3;
int K = 4;
INDArray A = Nd4j.create(new float[]{1,2,3,4,5,6,7,8,9,10,11,12,13,14,15}).reshape(M, N).castTo(DataType.DOUBLE);
INDArray B = Nd4j.create(new float[]{1,2,3,4,5,6,7,8,9,10,11,12}).reshape(N, K).castTo(DataType.DOUBLE);
SameDiff sd = SameDiff.create();
SDVariable A1 = sd.var("A1", A);
SDVariable A2 = sd.var("A2", A);
SDVariable B1 = sd.var("B1", B);
SDVariable B2 = sd.var("B2", B);
SDVariable[] batchMul = sd.batchMmul(sd.zero(null,M,N),sd.one(null,N,K),
new SDVariable[] {A1, A2}, new SDVariable[] {B1, B2});
Map<String,INDArray> m = sd.output(Collections.emptyMap(),Arrays.stream(batchMul).map(input -> input.name()).collect(Collectors.toList()));
INDArray resultingMatrix = m.get(batchMul[0].name());
assertArrayEquals(new long[]{5,4},resultingMatrix.shape());
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testBatchMMul(Nd4jBackend backend) {
Nd4j.getExecutioner().enableVerboseMode(true);
Nd4j.getExecutioner().enableDebugMode(true);
INDArray a = Nd4j.ones(2, 6);
INDArray b = Nd4j.ones(6, 4);
// 1. batchMmul in Nd4j causes an error in LibND4J
INDArray assertion = a.mmul(b);
INDArray[] res1 = Nd4j.base.batchMmul(a,b,new INDArray[]{a,a}, new INDArray[]{b,b}); // throws exception
assertEquals(assertion,res1[0]);
assertEquals(2,res1.length);
// 2. batchMmul in SameDiff produces a wrong result and sometimes even crashes (harder to reproduce; crash log linked below)
SameDiff sd = SameDiff.create();
SDVariable[] res2 = sd.batchMmul(sd.constant(a),
sd.constant(b),
new SDVariable[]{sd.constant(a),sd.constant(a)},new SDVariable[]{sd.constant(b),sd.constant(b)});
assertEquals(assertion,res2[0].eval()); // wrong result (or crash)
// 3. batchMmul in SameDiff on SDVariables of type ARRAY causes a NullPointerException
sd = SameDiff.create();
SDVariable[] res3 = sd.batchMmul(sd.constant(a).add(0),sd.constant(b).add(0),
new SDVariable[]{sd.constant(a).add(0),sd.constant(a).add(0)}, new SDVariable[]{sd.constant(b).add(0),sd.constant(b).add(0)}); // throws exception
assertEquals(assertion,res3[0].eval());
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testTensorMMulBp(Nd4jBackend backend) {
int batchSize = 4;
int seqLength = 8;
SameDiff sd = SameDiff.create();
SDVariable features = sd.placeHolder("features", DataType.DOUBLE, batchSize, seqLength);
SDVariable labels = sd.placeHolder("labels", DataType.DOUBLE, batchSize, batchSize);
SDVariable var = sd.var("variable", seqLength, batchSize);
SDVariable predictions = sd.tensorMmul("predictions", features, var, new int[]{1}, 0);
sd.loss.meanSquaredError("loss", labels, predictions, null);
TrainingConfig config = new TrainingConfig.Builder()
.updater(new Adam(0.1))
.dataSetFeatureMapping("features")
.dataSetLabelMapping("labels")
.build();
sd.setTrainingConfig(config);
RecordReader reader = new CollectionRecordReader(
Collections.nCopies(batchSize, Collections.nCopies(seqLength + batchSize, new IntWritable(1))));
DataSetIterator iterator = new RecordReaderDataSetIterator(
reader, batchSize, seqLength, seqLength + batchSize - 1, true);
System.out.println(sd.output(iterator, "predictions").get("predictions")); // forward pass works
sd.fit(iterator, 1); // backward pass throws exception
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testEmbedding() {
SameDiff sd = SameDiff.create();
SDVariable input = sd.placeHolder("input", DataType.INT32, -1, 2);
SDVariable input2 = sd.placeHolder("input2", INT32,2,2);
SDVariable lookUpDict = sd.var("lookUpDict",new XavierInitScheme('c', 4, 8), DataType.FLOAT, 4, 8);
SDVariable embeddingResult = sd.math.embeddingLookup("embeddingResult", lookUpDict, new SDVariable[]{input}, PartitionMode.MOD);
//
Map<String,INDArray> map = new HashMap<>();
INDArray inputArr = Nd4j.createFromArray(0, 3);
INDArray inputArr2 = Nd4j.createFromArray(2,3);
System.out.println(inputArr.shapeInfoToString());
map.put("input", inputArr);
map.put("input2",inputArr2);
//forward
Map<String,INDArray> result = sd.output(map, "embeddingResult");
System.out.println(result);
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testMmulWithTranspose(Nd4jBackend backend) {
//Here: [x,3]^T * [x,4] = [3,4]
for (int i : new int[]{2, 1}) {
System.out.println("i = " + i);
INDArray first = Nd4j.linspace(1, 3 * i, 3 * i, DataType.DOUBLE).reshape('c', i, 3); //To [1,3] or [2,3]
INDArray second = Nd4j.linspace(4, 4 + 4 * i, 4 * i, DataType.DOUBLE).reshape('c', i, 4); //To [1,4] or [2,4]
System.out.println("Shapes: " + Arrays.toString(first.shape()) + "\t" + Arrays.toString(second.shape()));
SameDiff sd = SameDiff.create();
SDVariable f = sd.var("in1", first);
SDVariable s = sd.var("in2", second);
MMulTranspose mt = MMulTranspose.builder()
.transposeA(true)
.transposeB(false)
.transposeResult(false)
.build();
SDVariable mmul = sd.mmul(f, s, true, false, false);
sd.updateVariableNameAndReference(mmul, "mmul");
INDArray out = mmul.eval();
INDArray exp = first.transpose().mmul(second);
assertEquals(exp, out);
SDVariable loss = sd.standardDeviation(mmul, true);
String err = OpValidation.validate(new TestCase(sd)
.expected(mmul.name(), exp));
assertNull(err);
}
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testMmulOutputSizeCalculation(){
//[3,2] x [2,4] with result transpose: output shape [4,3]
INDArray a = Nd4j.create(3,2);
INDArray b = Nd4j.create(2,4);
INDArray z = Nd4j.create(4,3);
Mmul m = new Mmul(a,b,z,MMulTranspose.builder()
.transposeA(false)
.transposeB(false)
.transposeResult(true)
.build());
val outShapes = Nd4j.getExecutioner().calculateOutputShape(m);
assertArrayEquals(new long[]{4,3}, outShapes.get(0).getShape());
Nd4j.getExecutioner().exec(m);
//Another case: ([3,4]*[2,4]T)T = [2,3] - tA=false, tB=true, tR=true
a = Nd4j.create(3,4);
b = Nd4j.create(2,4);
z = Nd4j.create(2,3);
m = new Mmul(a,b,z,MMulTranspose.builder()
.transposeA(false)
.transposeB(true)
.transposeResult(true)
.build());
val outShapes2 = Nd4j.getExecutioner().calculateOutputShape(m);
assertArrayEquals(new long[]{2,3}, outShapes2.get(0).getShape());
Nd4j.getExecutioner().exec(m);
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testFillOp(){
INDArray ia = Nd4j.createFromArray(new double[]{2,2}).castTo(DataType.INT);
double value = 42;
INDArray out = Nd4j.create(FLOAT, 2,2);
OpTestCase op = new OpTestCase(new Fill(ia, out, value));
INDArray expOut = Nd4j.valueArrayOf(new long[]{2,2}, 42.0f);
op.expectedOutput(0, expOut);
String err = OpValidation.validate(op);
assertNull(err);
}
@ParameterizedTest
@MethodSource("org.nd4j.linalg.BaseNd4jTestWithBackends#configs")
public void testClipByNorm(){
//Expected: if array.norm2(1) is less than 1.0, not modified
//Otherwise: array.tad(x,1) = array.tad(x,1) * 1.0 / array.tad(x,1).norm2()
Nd4j.getRandom().setSeed(12345);
INDArray arr = Nd4j.rand(3,5);
INDArray norm2_1 = arr.norm2(1);
arr.diviColumnVector(norm2_1);
norm2_1 = arr.norm2(1);
assertEquals(Nd4j.ones(3), norm2_1);
INDArray scale = Nd4j.create(new double[]{1.1, 1.0, 0.9}, new int[]{3});
arr.muliColumnVector(scale);
norm2_1 = arr.norm2(1);
INDArray out = Nd4j.create(arr.shape());
Nd4j.getExecutioner().exec(DynamicCustomOp.builder("clipbynorm")
.addInputs(arr)
.addOutputs(out)
.addIntegerArguments(1)
.addFloatingPointArguments(1.0)
.build());
INDArray norm2_1b = out.norm2(1);
INDArray exp = Nd4j.create(new double[]{1.0, 1.0, norm2_1.getDouble(2)}, new int[]{3});
assertEquals(exp, norm2_1b);
}
@ParameterizedTest