-
Notifications
You must be signed in to change notification settings - Fork 1
/
model_analyzer.py
914 lines (690 loc) · 24.3 KB
/
model_analyzer.py
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
# -*- coding: utf-8 -*-
"""
Created on Thu Jan 6 16:27:08 2022
!!!!!
Model analyzer specific for the cae model as defined in cae_32x32x32_zero_pad_bin_comp.py.
It can not be used for other models.
Not intended to be used outside of generate_results.py
Current per layers indicators
- number of input channels
- input shape
- number of output_channels
- output shape
- number of learnable parameters / sparsity (on the final line)
- number of weight vanishing coefficients
- number of weight vanishing kernel columns
- number of weight vanishing kernel rows
- bias number of vanishing coefficients
- weight entropy
- bias entropy
- MACC
- reduced MACC
@author: YFGI6212
"""
import pandas as pd
import torch
from fvcore.nn import FlopCountAnalysis
# model specific
from models.cae_32x32x32_zero_pad_bin_comp import CAE
class CAEModelAnalyzer:
def __init__(self):
self._model = CAE()
self._indicator_names = [
"type",
"in_channels",
"in_shape",
"out_channels",
"out_shape",
"params/sparsity",
"v_weight",
"v_r_weight",
"v_c_weight",
"v_bias",
"weight_ent",
"bias_ent",
"MACC",
"reduced MACC",
"KB32",
]
def load_model(self, model_path):
self._model.load_state_dict(
torch.load(model_path, map_location="cpu")["model_state_dict"]
)
self._model.eval()
self._input_h = 128 # horizontal size of input image
self._input_w = 128 # vertical size of input image
self._input_channels = self._model.e_conv_1[1].in_channels
self._fake_input = torch.rand(
1, self._input_channels, self._input_h, self._input_w, requires_grad=False
)
self._get_all_layers()
def _get_all_layers(self):
inputs = self._fake_input.clone()
encoder = [
self._model.e_conv_1,
self._model.e_conv_2,
self._model.e_block_1,
self._model.e_block_2,
self._model.e_block_3,
self._model.e_conv_3,
]
self.encoder = torch.nn.Sequential(*encoder)
self.encoder_tuples = [k[i] for k in encoder for i in range(k.__len__())]
for i in range(len(self.encoder_tuples)):
input_shape = inputs.shape
inputs = self.encoder_tuples[i](inputs)
self.encoder_tuples[i] = (
self.encoder_tuples[i],
list(input_shape),
list(inputs.shape),
)
decoder = [
self._model.d_up_conv_1,
self._model.d_block_1,
self._model.d_block_2,
self._model.d_block_3,
self._model.d_up_conv_2,
self._model.d_up_conv_3,
]
self.decoder = torch.nn.Sequential(*decoder)
self.decoder_tuples = [k[i] for k in decoder for i in range(k.__len__())]
for i in range(len(self.decoder_tuples)):
input_shape = inputs.shape
inputs = self.decoder_tuples[i](inputs)
self.decoder_tuples[i] = (
self.decoder_tuples[i],
list(input_shape),
list(inputs.shape),
)
def analyze_encoder(self, add_sum=True):
results = []
for i in range(len(self.encoder_tuples)):
layer_type = eval(f"Analyze{self.encoder_tuples[i][0]._get_name()}")
results.append(layer_type(self.encoder_tuples[i]))
if add_sum:
theSum = ["General"] # 0 type
theSum.append("-") # 1 in_channels
theSum.append("-") # 2 in_shape
theSum.append("-") # 3 out_channels
theSum.append("-") # 4 out_shape
# theSum.append(sum([results[i][5] for i in range(len(results))])) # 5 params
theSum.append(ModuleSparsity(self.encoder_tuples)) # 5 sparsity
theSum.append(
sum([results[i][6] for i in range(len(results))])
) # 6 v_weight
theSum.append(0.0) # 7 v_r_weight
theSum.append(0.0) # 8 v_c_weight
theSum.append(0.0) # 9 v_bias
theSum.append(ModuleEntropy(self.encoder_tuples)) # 10 weight_ent
theSum.append(0.0) # 11 bias_ent
theSum.append(sum([results[i][12] for i in range(len(results))])) # 12 MACC
theSum.append(
sum([results[i][13] for i in range(len(results))])
) # 13 reduced MACC
theSum.append(sum([results[i][14] for i in range(len(results))])) # 14 KB32
# theSum.append(ModuleFlops(self.encoder, self._fake_input)) # 14 FLOPs
results.append(theSum)
return pd.DataFrame(results, columns=self._indicator_names)
def analyze_decoder(self, add_sum=True):
results = []
for i in range(len(self.decoder_tuples)):
layer_type = eval(f"Analyze{self.decoder_tuples[i][0]._get_name()}")
results.append(layer_type(self.decoder_tuples[i]))
if add_sum:
theSum = ["General"] # 0 type
theSum.append("-") # 1 in_channels
theSum.append("-") # 2 in_shape
theSum.append("-") # 3 out_channels
theSum.append("-") # 4 out_shape
# theSum.append(sum([results[i][5] for i in range(len(results))])) # 5 params
theSum.append(ModuleSparsity(self.decoder_tuples))
theSum.append(
sum([results[i][6] for i in range(len(results))])
) # 6 v_weight
theSum.append(0.0) # 7 v_r_weight
theSum.append(0.0) # 8 v_c_weight
theSum.append(0.0) # 9 v_bias
theSum.append(ModuleEntropy(self.decoder_tuples)) # 10 weight_ent
theSum.append(0.0) # 11 bias_ent
theSum.append(sum([results[i][12] for i in range(len(results))])) # 12 MACC
theSum.append(
sum([results[i][13] for i in range(len(results))])
) # 13 reduced MACC
theSum.append(sum([results[i][14] for i in range(len(results))])) # 14 KB32
# theSum.append(ModuleFlops(self.decoder, self.encoder(self._fake_input)))
results.append(theSum)
return pd.DataFrame(results, columns=self._indicator_names)
def show_model(self):
return self._model
def showencoder_tuples(self, add_sum=True):
encoder = []
names = ["layer", "input shape", "output shape"]
for k in self.encoder_tuples:
encoder.append(list(k))
return pd.DataFrame(encoder, columns=names)
def showdecoder_tuples(self, add_sum=True):
decoder = []
names = ["layer", "input shape", "output shape"]
for k in self.decoder_tuples:
decoder.append(list(k))
return pd.DataFrame(decoder, columns=names)
# Analysis for specific layers
# The layer_tuple must be of the form (layer_object, input_shape, output_shape)
# with input_shape (C_in,h,w) and similarily for output_shape
# Useful methods
def TensorEntropy(tensor, tol=2):
""" Determine the entropy of a tensor.
- tol: number of decimals
"""
value, count = torch.unique(tensor.clone().detach(), return_counts=True)
p = count / count.sum().item()
return round(-(p * torch.log2(p)).sum().item(), tol)
# Entropy of a module tuple (as used in CAEModelAnalyzer)
def ModuleEntropy(module):
"""Returns the entropy of a pytorch module"""
all_weights = GetAllParams(module)
return TensorEntropy(all_weights)
def ModuleSparsity(module):
"""Returns the sparsity (proportion of zero parameters) of a pytorch module"""
all_weights = GetAllParams(module)
N = all_weights.shape[0] # weights array is flat
sparsity = (N - all_weights.count_nonzero()) / N
return sparsity.item()
def GetAllParams(module):
"""Return a flattened array of all learnable parameters of a model"""
all_weights = []
for layer_i in range(len(module)):
for param in module[layer_i][0].parameters():
all_weights.append(param.view(-1))
return torch.cat(all_weights)
def ModuleFlops(module, input):
flops = FlopCountAnalysis(module, input)
return flops.total()
# Conv2d
def AnalyzeConv2d(layer_tuple):
"""
Parameters
----------
layer : torch.nn.modules.conv.Conv2d
"""
assert (
layer_tuple[0]._get_name() == "Conv2d"
), f"Layer must be a Conv2d but received {layer_tuple[0]}"
indicators = [layer_tuple[0]._get_name()]
indicators.append(layer_tuple[0].in_channels)
indicators.append((layer_tuple[1][1], layer_tuple[1][2], layer_tuple[1][3]))
indicators.append(layer_tuple[0].out_channels)
indicators.append((layer_tuple[2][1], layer_tuple[2][3], layer_tuple[2][3]))
indicators.append(Conv2dLearnableParameters(layer_tuple))
indicators.append(Conv2dWeightVanishingCoefficients(layer_tuple[0]))
indicators.append(Conv2dWeightVanishingRows(layer_tuple[0]))
indicators.append(Conv2dWeightVanishingColumns(layer_tuple[0]))
indicators.append(Conv2dBiasVanishingCoefficients(layer_tuple[0]))
indicators.append(Conv2dWeightEntropy(layer_tuple[0]))
indicators.append(Conv2dBiasEntropy(layer_tuple[0]))
indicators.append(Conv2dMacc(layer_tuple))
indicators.append(Conv2dReducedMacc(layer_tuple))
indicators.append(Conv2dKB32(layer_tuple))
# indicators.append(0)
return indicators
def Conv2dMacc(layer_tuple):
"""
Parameters
----------
layer_tuple : (layer object, input_shape, output_shape)
with
- input_shape = (batch_size,C_in,x_in,y_in)
- output_shape = (batch_size,C_out,x_out,y_out)
Returns: the number of MACC given with K_x * K_y * C_in * C_out * x_out * y_out
"""
return (
layer_tuple[0].kernel_size[0]
* layer_tuple[0].kernel_size[1]
* layer_tuple[0].out_channels
* layer_tuple[0].in_channels
* layer_tuple[2][2]
* layer_tuple[2][3]
)
def Conv2dReducedMacc(layer_tuple):
"""
Parameters
----------
layer_tuple : (layer object, input_shape, output_shape)
with
- input_shape = (batch_size,C_in,x_in,y_in)
- output_shape = (batch_size,C_out,x_out,y_out)
Returns: the reduced number of MACC given with K_x * K_y * C_in * C_out * x_out * y_out
"""
return (
layer_tuple[0].kernel_size[0]
* layer_tuple[0].kernel_size[1]
* (layer_tuple[0].out_channels - Conv2dWeightVanishingRows(layer_tuple[0]))
* (layer_tuple[0].in_channels - Conv2dWeightVanishingColumns(layer_tuple[0]))
* layer_tuple[2][2]
* layer_tuple[2][3]
)
def Conv2dLearnableParameters(layer_tuple):
"""
Parameters
----------
layer_tuple : (layer object, input_shape, output_shape)
with
- input_shape = (batch_size,C_in,x_in,y_in)
- output_shape = (batch_size,C_out,x_out,y_out)
Returns: the number of learnable parameters: #weight + #bias
"""
return layer_tuple[0].weight.nelement() + layer_tuple[0].bias.nelement()
def Conv2dReducedLearnableParameters(layer_tuple):
"""
Parameters
----------
layer_tuple : (layer object, input_shape, output_shape)
with
- input_shape = (batch_size,C_in,x_in,y_in)
- output_shape = (batch_size,C_out,x_out,y_out)
Returns: estimates the number of learnable parameters of the
projected layer
"""
return (
(layer_tuple[0].in_channels - Conv2dWeightVanishingColumns(layer_tuple[0]))
* (layer_tuple[0].out_channels - Conv2dWeightVanishingRows(layer_tuple[0]))
* layer_tuple[0].kernel_size[0]
* layer_tuple[0].kernel_size[1]
+ layer_tuple[0].out_channels # weight
- Conv2dWeightVanishingRows(layer_tuple[0])
) # bias
def Conv2dWeightVanishingCoefficients(layer):
"""
Parameters
----------
layer : Conv2d object
Returns: the number of vanishing coefficients in the weight
"""
return torch.where(layer.weight == 0, 1, 0).sum().item()
def Conv2dWeightVanishingRows(layer):
"""
Parameters
----------
layer : Conv2d object
Returns: the number of vanishing kernel rows in the weight
"""
return sum(
[int(torch.norm(layer.weight[i], p=1) == 0) for i in range(layer.out_channels)]
)
def Conv2dWeightVanishingColumns(layer):
"""
Parameters
----------
layer : Conv2d object
Returns: the number of vanishing kernel columns in the weight
"""
return sum(
[
int(torch.norm(layer.weight[:, i], p=1) == 0)
for i in range(layer.in_channels)
]
)
def Conv2dBiasVanishingCoefficients(layer):
"""
Parameters
----------
layer : Conv2d object
Returns: the number of vanishing coefficients in the bias
"""
return torch.where(layer.bias == 0, 1, 0).sum().item()
def Conv2dWeightEntropy(layer):
"""
Parameters
----------
layer : Conv2d object
Returns: entropy of the weight
"""
return TensorEntropy(layer.weight)
def Conv2dBiasEntropy(layer):
"""
Parameters
----------
layer : Conv2d object
Returns: entropy of the bias
"""
return TensorEntropy(layer.bias)
def Conv2dKB32(layer_tuple):
"""
Parameters
----------
layer : Conv2d object
Returns: evaluation the storage requierments in KB
"""
return (
32.0
* Conv2dLearnableParameters(layer_tuple)
* Conv2dWeightEntropy(layer_tuple[0])
/ 8000.0
)
# ZeroPad2d
def AnalyzeZeroPad2d(layer_tuple):
"""
Parameters
----------
layer : torch.nn.modules.conv.Conv2d
"""
assert (
layer_tuple[0]._get_name() == "ZeroPad2d"
), f"Layer must be a ZeroPad2d by received {layer_tuple[0]}"
indicators = [layer_tuple[0]._get_name()]
# number of input channels
indicators.append(layer_tuple[1][0])
# input shape
indicators.append((layer_tuple[1][1], layer_tuple[1][2], layer_tuple[1][3]))
# number of output channels
indicators.append(layer_tuple[2][0])
# output shape
indicators.append((layer_tuple[2][1], layer_tuple[2][3], layer_tuple[2][3]))
# number of learnable parameters
indicators.append(0)
# number of weight vanishing coefficients
indicators.append(0)
# number of weight vanishing kernel columns
indicators.append(0)
# number of weight vanishing kernel rows
indicators.append(0)
# bias number of vanishing coefficients
indicators.append(0)
# weight entropy
indicators.append(0)
# bias entropy
indicators.append(0)
# MACC
indicators.append(0)
# Reduced MACC
indicators.append(0)
# KB32
indicators.append(0)
return indicators
# LeakyReLU
def AnalyzeLeakyReLU(layer_tuple):
"""
Parameters
----------
layer : torch.nn.modules.conv.Conv2d
"""
assert (
layer_tuple[0]._get_name() == "LeakyReLU"
), f"Layer must be a LeakyReLU by received {layer_tuple[0]}"
indicators = [layer_tuple[0]._get_name()]
# number of input channels
indicators.append(layer_tuple[1][0])
# input shape
indicators.append((layer_tuple[1][1], layer_tuple[1][2], layer_tuple[1][3]))
# number of output channels
indicators.append(layer_tuple[2][0])
# output shape
indicators.append((layer_tuple[2][1], layer_tuple[2][3], layer_tuple[2][3]))
# number of learnable parameters
indicators.append(0)
# number of weight vanishing coefficients
indicators.append(0)
# number of weight vanishing kernel columns
indicators.append(0)
# number of weight vanishing kernel rows
indicators.append(0)
# bias number of vanishing coefficients
indicators.append(0)
# weight entropy
indicators.append(0)
# bias entropy
indicators.append(0)
# MACC
indicators.append(0)
# Reduced MACC
indicators.append(0)
# KB32
indicators.append(0)
return indicators
# Tanh
def AnalyzeTanh(layer_tuple):
"""
Parameters
----------
layer : torch.nn.modules.conv.Conv2d
"""
assert (
layer_tuple[0]._get_name() == "Tanh"
), f"Layer must be a Tanh by received {layer_tuple[0]}"
indicators = [layer_tuple[0]._get_name()]
# number of input channels
indicators.append(layer_tuple[1][0])
# input shape
indicators.append((layer_tuple[1][1], layer_tuple[1][2], layer_tuple[1][3]))
# number of output channels
indicators.append(layer_tuple[2][0])
# output shape
indicators.append((layer_tuple[2][1], layer_tuple[2][3], layer_tuple[2][3]))
# number of learnable parameters
indicators.append(0)
# number of weight vanishing coefficients
indicators.append(0)
# number of weight vanishing kernel columns
indicators.append(0)
# number of weight vanishing kernel rows
indicators.append(0)
# bias number of vanishing coefficients
indicators.append(0)
# weight entropy
indicators.append(0)
# bias entropy
indicators.append(0)
# MACC
indicators.append(0)
# Reduced MACC
indicators.append(0)
# KB32
indicators.append(0)
return indicators
# ReflectionPad2d
def AnalyzeReflectionPad2d(layer_tuple):
"""
Parameters
----------
layer : torch.nn.modules.conv.Conv2d
"""
assert (
layer_tuple[0]._get_name() == "ReflectionPad2d"
), f"Layer must be a ReflectionPad2d by received {layer_tuple[0]}"
indicators = [layer_tuple[0]._get_name()]
# number of input channels
indicators.append(layer_tuple[1][0])
# input shape
indicators.append((layer_tuple[1][1], layer_tuple[1][2], layer_tuple[1][3]))
# number of output channels
indicators.append(layer_tuple[2][0])
# output shape
indicators.append((layer_tuple[2][1], layer_tuple[2][3], layer_tuple[2][3]))
# number of learnable parameters
indicators.append(0)
# number of weight vanishing coefficients
indicators.append(0)
# number of weight vanishing kernel columns
indicators.append(0)
# number of weight vanishing kernel rows
indicators.append(0)
# bias number of vanishing coefficients
indicators.append(0)
# weight entropy
indicators.append(0)
# bias entropy
indicators.append(0)
# MACC
indicators.append(0)
# Reduced MACC
indicators.append(0)
# KB32
indicators.append(0)
return indicators
# Analysis of TransposeConv2d layers
def AnalyzeConvTranspose2d(layer_tuple):
assert (
layer_tuple[0]._get_name() == "ConvTranspose2d"
), f"Layer must be a ConvTranspose2d by received {layer_tuple[0]}"
indicators = [layer_tuple[0]._get_name()]
# number of input channels
indicators.append(layer_tuple[1][0])
# input shape
indicators.append((layer_tuple[1][1], layer_tuple[1][2], layer_tuple[1][3]))
# number of output channels
indicators.append(layer_tuple[2][0])
# output shape
indicators.append((layer_tuple[2][1], layer_tuple[2][3], layer_tuple[2][3]))
# number of learnable parameters
indicators.append(ConvTranspose2dLearnableParameters(layer_tuple[0]))
# number of weight vanishing coefficients
indicators.append(ConvTranspose2dWeightVanishingCoefficients(layer_tuple[0]))
# number of weight vanishing kernel columns
indicators.append(ConvTranspose2dWeightVanishingColumns(layer_tuple[0]))
# number of weight vanishing kernel rows
indicators.append(ConvTranspose2dWeightVanishingRows(layer_tuple[0]))
# bias number of vanishing coefficients
indicators.append(ConvTranspose2dBiasVanishingCoefficients(layer_tuple[0]))
# weight entropy
indicators.append(ConvTranspose2dWeightEntropy(layer_tuple[0]))
# bias entropy
indicators.append(ConvTranspose2dBiasEntropy(layer_tuple[0]))
# MACC
indicators.append(ConvTranspose2dMacc(layer_tuple))
# Reduced MACC
indicators.append(ConvTranspose2dReducedMacc(layer_tuple))
# KB32
indicators.append(ConvTranspose2dKB32(layer_tuple))
return indicators
def ConvTranspose2dLearnableParameters(layer):
"""
Parameters
----------
layer_tuple : (layer object, input_shape, output_shape)
with
- input_shape = (batch_size,C_in,x_in,y_in)
- output_shape = (batch_size,C_out,x_out,y_out)
Returns: the number of learnable parameters: #weight + #bias
"""
return layer.weight.nelement() + layer.bias.nelement()
def ConvTranspose2dWeightVanishingCoefficients(layer):
"""
Parameters
----------
layer : ConvTranspose2d object
Returns: the number of vanishing coefficients in the weight
"""
return torch.where(layer.weight == 0, 1, 0).sum().item()
def ConvTranspose2dBiasVanishingCoefficients(layer):
"""
Parameters
----------
layer : ConvTranspose2d object
Returns: the number of vanishing coefficients in the bias
"""
return torch.where(layer.bias == 0, 1, 0).sum().item()
def ConvTranspose2dWeightVanishingRows(layer):
"""
Parameters
----------
layer : ConvTranspose2d object
Returns: the number of vanishing kernel rows in the weight
"""
return sum(
[int(torch.norm(layer.weight[i], p=1) == 0) for i in range(layer.in_channels)]
)
def ConvTranspose2dWeightVanishingColumns(layer):
"""
Parameters
----------
layer : ConvTranspose2d object
Returns: the number of vanishing kernel columns in the weight
"""
return sum(
[
int(torch.norm(layer.weight[:, i], p=1) == 0)
for i in range(layer.out_channels)
]
)
def ConvTranspose2dBiasVanishingCoefficients(layer):
"""
Parameters
----------
layer : ConvTranspose2d object
Returns: the number of vanishing coefficients in the bias
"""
return torch.where(layer.bias == 0, 1, 0).sum().item()
def ConvTranspose2dWeightEntropy(layer):
"""
Parameters
----------
layer : TransposeConv2d object
Returns: entropy of the weight
"""
return TensorEntropy(layer.weight)
def ConvTranspose2dBiasEntropy(layer):
"""
Parameters
----------
layer : TransposeConv2d object
Returns: entropy of the bias
"""
return TensorEntropy(layer.bias)
def ConvTranspose2dMacc(layer_tuple):
"""
Parameters
----------
layer_tuple : (layer object, input_shape, output_shape)
with
- input_shape = (batch_size,C_in,x_in,y_in)
- output_shape = (batch_size,C_out,x_out,y_out)
Returns: the number of MACC given with K_x * K_y * C_in * C_out * x_in * y_in
"""
return (
layer_tuple[0].kernel_size[0]
* layer_tuple[0].kernel_size[1]
* layer_tuple[0].out_channels
* layer_tuple[0].in_channels
* layer_tuple[1][2]
* layer_tuple[1][3]
)
def ConvTranspose2dReducedMacc(layer_tuple):
"""
Parameters
----------
layer_tuple : (layer object, input_shape, output_shape)
with
- input_shape = (batch_size,C_in,x_in,y_in)
- output_shape = (batch_size,C_out,x_out,y_out)
Returns: the reduced number of MACC given with K_x * K_y * C_in * C_out * x_out * y_out
"""
return (
layer_tuple[0].kernel_size[0]
* layer_tuple[0].kernel_size[1]
* (
layer_tuple[0].out_channels
- ConvTranspose2dWeightVanishingRows(layer_tuple[0])
)
* (
layer_tuple[0].in_channels
- ConvTranspose2dWeightVanishingColumns(layer_tuple[0])
)
* layer_tuple[1][2]
* layer_tuple[1][3]
)
def ConvTranspose2dKB32(layer_tuple):
"""
Parameters
----------
layer : Conv2d object
Returns: evaluation the storage requirements in KB
"""
return (
32.0
* ConvTranspose2dLearnableParameters(layer_tuple[0])
* ConvTranspose2dWeightEntropy(layer_tuple[0])
/ 8000.0
)
### to be suppressed
# MA = CAEModelAnalyzer()
# MA.load_model("../experiments/trainComp_40_200/checkpoint/best_model_Mask.pth")