-
Notifications
You must be signed in to change notification settings - Fork 2
/
main_multiobj_pruning_demo_cifar_and_mnist.py
773 lines (664 loc) · 38 KB
/
main_multiobj_pruning_demo_cifar_and_mnist.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
'''
Author : Malena Reiners, M.Sc. Mathematics
Description : Demo application of the stochastic multi gradient descent algorithm and its extension, implemented in
MultiobjectiveOptimizers.py (SMGD, MAdam, MRMSProp version 0.1).
Combined for our purpose with a network pruning strategy based on l1 regularization on MNIST and CIFAR10
in different settings.
Import of MultiobjectiveClasses.py, CustomLosses.py, CustomModels.py is neccessary for usage.
Two objectives/loss functions are used to optimize the neural network in more than one goal
For the multi-gradient algorithm we refer to the following paper for more details:
S. Liu and L. N. Vicente, The stochastic multi-gradient algorithm for multi-objective optimization and its
application to supervised machine learning, ISE Technical Report 19T-011, Lehigh University.
and for the success using different extensions and combine the MOP training with pruning weights:
Reiners, M., Klamroth, K., Stiglmayr, M., 2020, Efficient and Sparse Neural Networks by Pruning
Weights in a Multiobjective Learning Approach, https://arxiv.org/abs/2008.13590
This is the main script for demonstration purpose. Most of the plots and experimental results can be traced here.
Input(s) : The SMGD Algorithm to solve the multiobjective optimization problem with two loss functions, in this script is used in two
SGD extensions as well to demonstrate the difference between the optimizers.
Output(s) : SMGD outputs a Pareto stationary point for the trade-off between both objective functions (loss functions).
Trained convolutional neural network architecture. More details on the used architectures can be found
in CustomModels.py.
Notes : The code is implemented using Python 3.7, Keras 2.3.1 and Tensorflow 1.14.0
Please not that it is mandatory to use these versions of Tensorflow and Keras, otherwise the program cannot be executed.
The reason for this are the changed and adapted Keras and Tensorflow functions of this particular versions.
To extend it to multiple objectives (more than two), one may need additional packages, e.g., Gurobi, to solve a quadratic
subproblem and compute a common descent direction of multi-objective at current point.
'''
import numpy as np
import os
import sys
import seaborn as sns
import matplotlib.pyplot as plt
import tensorflow as tf
import keras
from keras import backend as K
from keras import regularizers
from keras.callbacks import LearningRateScheduler, LambdaCallback, Callback
from keras.engine.training import Model
from keras.optimizers import Optimizer, SGD, Adam, RMSprop
from keras.backend.tensorflow_backend import set_session
### custom scripts
from MultiobjectiveOptimizers import SMGD, MAdam, MRMSprop
from MultiobjectiveClasses import Multi
from CustomLosses import L1loss, L2loss, L1lossDense, L2lossConv, L1L2lossDenseConv
from CustomModels import get_data, lenet5multimodel, lenet5regmodel, vggnetmultimodel, vggnetregmodel
### for comparison reasons train on the same init weights
from tensorflow import set_random_seed
from numpy.random import seed
### Uncomment if you want to use same initialization for comparison reasons
#seed(1)
#set_random_seed(2)
### Customized inputs
m=input("Please choose a dataset to train: mnist or cifar10? ")
print( "We will train on: " ,m)
if m == 'mnist':
mnist=True
cifar10=False
opt=input("Choose: SGD or Adam or both? ")
if opt=='SGD':
optimizers=['multi', 'sgd']
elif opt == 'Adam':
optimizers=['multiadam', 'adam']
elif opt == 'both':
optimizers=['multiadam', 'multi', 'adam', 'sgd']
else:
print("Specify an algorithm!")
elif m== 'cifar10':
mnist=False
cifar10=True
opt=input("Choose: SGD or RMSProp or both? ")
if opt=='SGD':
optimizers=['multi', 'sgd']
elif opt == 'RMSProp':
optimizers=['multirms', 'rms']
elif opt == 'both':
optimizers=['multirms', 'multi', 'rms', 'sgd']
else:
print("Specify an algorithm!")
print("We will use the following optimizers: " + str(optimizers))
pruning=input("Typ True, if you want to use pruning as well: ")
if pruning:
print("We will use ITP")
lr=input( "Please choose a starting learning rate: ")
learning_rate=float(lr)
print("The learning rate is" + lr )
### Uncomment if you want to use the settings without a terminal input
# mnist= False
# cifar10= True
# learning_rate=float(sys.argv[1])
# pruning= False
# optimizers=['multiadam', 'multi', 'adam', 'sgd']
# Define Weight Update for prunign weights
def update_weights(weight_matrix, threshold):
sparsified_weights = []
for w in weight_matrix:
bool_mask = (abs(w) > threshold).astype(int)
sparsified_weights.append(w * bool_mask)
return sparsified_weights
class TestCallback_multi(Callback):
def __init__(self, train_data):
self.train_data = train_data
def on_epoch_end(self, epoch, logs={}):
x, y = self.train_data
loss1,loss2, acc = self.model.evaluate_multi(x, y, verbose=0)
loss1_values.append(loss1)
loss2_values.append(loss2)
accuracy_values.append(acc)
class TestCallback(Callback):
def __init__(self, train_data):
self.train_data = train_data
def on_epoch_end(self, epoch, logs={}):
x, y = self.train_data
loss, acc = self.model.evaluate(x, y, verbose=0)
loss_values.append(loss)
accuracy_values.append(acc)
# Define learning rate schedule (LRS) for MNIST
def lr_schedule_mnist(epoch):
lrate = learning_rate
lrate_e = 0.1*learning_rate
epochs = 30
middle = 0.75 * epochs
frame = middle / 10
if epoch > 0:
lrate = -(lrate - lrate_e) * (np.exp((epoch - middle) / frame) / (
np.exp((epoch - middle) / frame) + 1)) + lrate
return lrate
# Define learning rate schedule CIFAR
def lr_schedule_cifar(epoch):
lrate = learning_rate
lrate_e = 0.1*learning_rate
epochs = 125
middle = 0.75 * epochs
frame = middle / 10
if epoch > 0:
lrate = -(lrate - lrate_e) * (np.exp((epoch - middle) / frame) / (
np.exp((epoch - middle) / frame) + 1)) + lrate
return lrate
x_train, y_train, x_test, y_test, train_data,input_shape, epochs, num_classes = get_data(mnist, cifar10)
###################################### MNIST Training ###########################################################
if mnist: # test performance of MAdam and Adam optimizer on the same problem
for opt in optimizers:
# start with the same init weights
#seed(1)
#set_random_seed(2)
# collect values while training (in evaluation mode)
loss1_values = []
loss2_values = []
loss_values=[]
val_loss_values = []
accuracy_values = []
val_accuracy_values = []
nonzero_weights1 = []
nonzero_weights2 = []
nonzero_weights3 = []
if opt == 'multiadam':
lambdaconv = 1e-4
model = lenet5multimodel(input_shape=input_shape, weight_decay=lambdaconv)
model.mcompile(optimizer=MAdam(multi=True, split=False, learning_rate=learning_rate),
loss1='sparse_categorical_crossentropy', loss2=L1lossDense(model),
metrics=['accuracy'])
nonzero_weights1.append([np.count_nonzero(
model.get_layer('denselayer1').get_weights()[0])])
nonzero_weights2.append([np.count_nonzero(
model.get_layer('denselayer2').get_weights()[0])])
nonzero_weights3.append([np.count_nonzero(
model.get_layer('denselayer3').get_weights()[0])])
elif opt == 'multi':
lambdaconv = 1e-4
momentum = 0.9
decay_rate = learning_rate / epochs
model = lenet5multimodel(input_shape=input_shape, weight_decay=lambdaconv)
model.mcompile(optimizer=SMGD(multi=True, split=False, learning_rate=learning_rate, decay= decay_rate, momentum=momentum),
loss1='sparse_categorical_crossentropy', loss2=L1lossDense(model),
metrics=['accuracy'])
nonzero_weights1.append([np.count_nonzero(
model.get_layer('denselayer1').get_weights()[0])])
nonzero_weights2.append([np.count_nonzero(
model.get_layer('denselayer2').get_weights()[0])])
nonzero_weights3.append([np.count_nonzero(
model.get_layer('denselayer3').get_weights()[0])])
elif opt == 'adam':
lambdaconv=1e-4 # empirically determined
lambdadense= 3e-4 # from first experiments (empirically determined)
model=lenet5regmodel(lambdas=lambdadense, weight_decay=lambdaconv, input_shape= input_shape)
model.compile(optimizer=Adam(learning_rate=learning_rate),loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
nonzero_weights1.append([np.count_nonzero(
model.get_layer('denselayer1').get_weights()[0])])
nonzero_weights2.append([np.count_nonzero(
model.get_layer('denselayer2').get_weights()[0])])
nonzero_weights3.append([np.count_nonzero(
model.get_layer('denselayer3').get_weights()[0])])
elif opt == 'sgd':
lambdaconv=1e-4 # empirically determined
lambdadense= 3e-4 # from first experiments (empirically determined)
momentum = 0.9
decay_rate = learning_rate / epochs
model=lenet5regmodel(lambdas=lambdadense, weight_decay=lambdaconv, input_shape= input_shape)
model.compile(optimizer=SGD(learning_rate=learning_rate, decay=decay_rate, momentum=momentum),loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
nonzero_weights1.append([np.count_nonzero(
model.get_layer('denselayer1').get_weights()[0])])
nonzero_weights2.append([np.count_nonzero(
model.get_layer('denselayer2').get_weights()[0])])
nonzero_weights3.append([np.count_nonzero(
model.get_layer('denselayer3').get_weights()[0])])
## Define Callbacks for pruning
threshold=0.001
weight_callback_batch=LambdaCallback(on_batch_end= lambda batch,
logs: [model.get_layer(f"{name}").set_weights(update_weights(
model.get_layer(f"{name}").get_weights(), threshold))
for name in ['denselayer1', 'denselayer2','denselayer3']]
)
safe_nonzeroweights1=LambdaCallback(on_epoch_end= lambda epoch,
logs: [nonzero_weights1.append([np.count_nonzero(
model.get_layer('denselayer1').get_weights()[0])])
]
)
safe_nonzeroweights2=LambdaCallback(on_epoch_end= lambda epoch,
logs: [nonzero_weights2.append([np.count_nonzero(
model.get_layer('denselayer2').get_weights()[0])])
]
)
safe_nonzeroweights3=LambdaCallback(on_epoch_end= lambda epoch,
logs: [nonzero_weights3.append([np.count_nonzero(
model.get_layer('denselayer3').get_weights()[0])])
]
)
if pruning:
## Get, Update and Set Weights before Training
weights1 = model.get_layer('denselayer1').get_weights()
weights2 = model.get_layer('denselayer2').get_weights()
weights3 = model.get_layer('denselayer3').get_weights()
sparsified_weights1 = update_weights(weights1, threshold)
sparsified_weights2 = update_weights(weights2, threshold)
sparsified_weights3 = update_weights(weights3, threshold)
model.get_layer('denselayer1').set_weights(sparsified_weights1)
model.get_layer('denselayer2').set_weights(sparsified_weights2)
model.get_layer('denselayer3').set_weights(sparsified_weights3)
nonzero_weights1.append([np.count_nonzero(
model.get_layer('denselayer1').get_weights()[0])])
nonzero_weights2.append([np.count_nonzero(
model.get_layer('denselayer2').get_weights()[0])])
nonzero_weights3.append([np.count_nonzero(
model.get_layer('denselayer3').get_weights()[0])])
## Start Training
if opt == 'multiadam':
if pruning:
history_multiadam=model.mfit(x_train, y_train, epochs=epochs, validation_data=[x_test,y_test], callbacks=[LearningRateScheduler(lr_schedule_mnist),weight_callback_batch,safe_nonzeroweights1, safe_nonzeroweights2, safe_nonzeroweights3,TestCallback_multi((x_train,y_train))])
nonzero_weights1.append([np.count_nonzero(
model.get_layer('denselayer1').get_weights()[0])])
nonzero_weights2.append([np.count_nonzero(
model.get_layer('denselayer2').get_weights()[0])])
nonzero_weights3.append([np.count_nonzero(
model.get_layer('denselayer3').get_weights()[0])])
nonzero_weights1_multiadam= nonzero_weights1
nonzero_weights2_multiadam= nonzero_weights2
nonzero_weights3_multiadam= nonzero_weights3
else:
history_multiadam=model.mfit(x_train, y_train, epochs=epochs, validation_data=[x_test,y_test], callbacks=[LearningRateScheduler(lr_schedule_mnist),TestCallback_multi((x_train,y_train))])
accuracy_values_multiadam= accuracy_values
accuracy_values=[]
loss1_values_multiadam = loss1_values
loss1_values = []
loss2_values_multiadam = loss2_values
loss2_values = []
weights1 = model.get_layer("denselayer1").get_weights() #weights and biases of the layer
L1w1=sum(sum(sum(np.abs(weights1))))
L0w1=np.count_nonzero(weights1[0]) + np.count_nonzero(weights1[1])
weights2 = model.get_layer("denselayer2").get_weights()
L1w2=sum(sum(sum(np.abs(weights2))))
L0w2= np.count_nonzero(weights2[0]) + np.count_nonzero(weights2[1])
weights3 = model.get_layer("denselayer3").get_weights()
L1w3=sum(sum(sum(np.abs(weights3))))
L0w3=np.count_nonzero(weights3[0]) + np.count_nonzero(weights3[1])
L0ges_multiadam=L0w1+L0w2+L0w3
L1ges_multiadam= L1w1+L1w2+L1w3
L0multiadam= [L0w1, L0w2, L0w3, L0ges_multiadam]
L1multiadam= [L1w1, L1w2, L1w3, L1ges_multiadam]
# EVALUATE
[train_loss1_multiadam, train_loss2_multiadam, train_accuracy_multiadam] = model.evaluate_multi(x_train, y_train)
[test_loss1_multiadam, test_loss2_multiadam, test_accuracy_multiadam] = model.evaluate_multi(x_test, y_test)
elif opt == 'multi':
if pruning:
history_multi=model.mfit(x_train, y_train, epochs=epochs, validation_data=[x_test,y_test], callbacks=[LearningRateScheduler(lr_schedule_mnist), weight_callback_batch,safe_nonzeroweights1, safe_nonzeroweights2, safe_nonzeroweights3,TestCallback_multi((x_train,y_train))])
nonzero_weights1.append([np.count_nonzero(
model.get_layer('denselayer1').get_weights()[0])])
nonzero_weights2.append([np.count_nonzero(
model.get_layer('denselayer2').get_weights()[0])])
nonzero_weights3.append([np.count_nonzero(
model.get_layer('denselayer3').get_weights()[0])])
nonzero_weights1_multi= nonzero_weights1
nonzero_weights2_multi= nonzero_weights2
nonzero_weights3_multi= nonzero_weights3
else:
history_multi=model.mfit(x_train, y_train, epochs=epochs, validation_data=[x_test,y_test], callbacks=[LearningRateScheduler(lr_schedule_mnist),TestCallback_multi((x_train,y_train))])
accuracy_values_multi= accuracy_values
accuracy_values=[]
loss1_values_multi = loss1_values
loss1_values = []
loss2_values_multi = loss2_values
loss2_values = []
weights1 = model.get_layer("denselayer1").get_weights() #weights and biases of the layer
L1w1=sum(sum(sum(np.abs(weights1))))
L0w1=np.count_nonzero(weights1[0]) + np.count_nonzero(weights1[1])
weights2 = model.get_layer("denselayer2").get_weights()
L1w2=sum(sum(sum(np.abs(weights2))))
L0w2= np.count_nonzero(weights2[0]) + np.count_nonzero(weights2[1])
weights3 = model.get_layer("denselayer3").get_weights()
L1w3=sum(sum(sum(np.abs(weights3))))
L0w3=np.count_nonzero(weights3[0]) + np.count_nonzero(weights3[1])
L0ges_multi=L0w1+L0w2+L0w3
L1ges_multi= L1w1+L1w2+L1w3
L0multi= [L0w1, L0w2, L0w3, L0ges_multi]
L1multi= [L1w1, L1w2, L1w3, L1ges_multi]
# EVALUATE
[train_loss1_multi, train_loss2_multi, train_accuracy_multi] = model.evaluate_multi(x_train, y_train)
[test_loss1_multi, test_loss2_multi, test_accuracy_multi] = model.evaluate_multi(x_test, y_test)
elif opt == 'adam':
if pruning:
history_adam= model.fit(x_train, y_train, epochs=epochs, validation_data=[x_test,y_test], callbacks=[LearningRateScheduler(lr_schedule_mnist),weight_callback_batch,safe_nonzeroweights1, safe_nonzeroweights2, safe_nonzeroweights3,TestCallback((x_train,y_train))])
nonzero_weights1.append([np.count_nonzero(
model.get_layer('denselayer1').get_weights()[0])])
nonzero_weights2.append([np.count_nonzero(
model.get_layer('denselayer2').get_weights()[0])])
nonzero_weights3.append([np.count_nonzero(
model.get_layer('denselayer3').get_weights()[0])])
nonzero_weights1_adam= nonzero_weights1
nonzero_weights2_adam= nonzero_weights2
nonzero_weights3_adam= nonzero_weights3
else:
history_adam= model.fit(x_train, y_train, epochs=epochs, validation_data=[x_test,y_test], callbacks=[LearningRateScheduler(lr_schedule_mnist),TestCallback((x_train,y_train))])
accuracy_values_adam= accuracy_values
accuracy_values=[]
loss_values_adam = loss_values
loss_values = []
weights1 = model.get_layer("denselayer1").get_weights() #weights and biases of the layer
L1w1=sum(sum(sum(np.abs(weights1))))
L0w1=np.count_nonzero(weights1[0]) + np.count_nonzero(weights1[1])
weights2 = model.get_layer("denselayer2").get_weights()
L1w2=sum(sum(sum(np.abs(weights2))))
L0w2= np.count_nonzero(weights2[0]) + np.count_nonzero(weights2[1])
weights3 = model.get_layer("denselayer3").get_weights()
L1w3=sum(sum(sum(np.abs(weights3))))
L0w3=np.count_nonzero(weights3[0]) + np.count_nonzero(weights3[1])
L0ges_adam= L0w1+L0w2+L0w3
L1ges_adam= L1w1+L1w2+L1w3
L0_adam= [L0w1, L0w2, L0w3,L0ges_adam]
L1_adam= [L1w1, L1w2, L1w3, L1ges_adam]
# EVALUATE
[train_loss_adam, train_accuracy_adam]= model.evaluate(x_train, y_train)
[test_loss_adam, test_accuracy_adam] = model.evaluate(x_test, y_test)
elif opt == 'sgd':
if pruning:
history_sgd= model.fit(x_train, y_train, epochs=epochs, validation_data=[x_test,y_test], callbacks=[LearningRateScheduler(lr_schedule_mnist),weight_callback_batch,safe_nonzeroweights1, safe_nonzeroweights2, safe_nonzeroweights3,TestCallback((x_train,y_train))])
nonzero_weights1.append([np.count_nonzero(
model.get_layer('denselayer1').get_weights()[0])])
nonzero_weights2.append([np.count_nonzero(
model.get_layer('denselayer2').get_weights()[0])])
nonzero_weights3.append([np.count_nonzero(
model.get_layer('denselayer3').get_weights()[0])])
nonzero_weights1_sgd= nonzero_weights1
nonzero_weights2_sgd= nonzero_weights2
nonzero_weights3_sgd= nonzero_weights3
else:
history_sgd= model.fit(x_train, y_train, epochs=epochs, validation_data=[x_test,y_test], callbacks=[LearningRateScheduler(lr_schedule_mnist),TestCallback((x_train,y_train))])
accuracy_values_sgd= accuracy_values
accuracy_values=[]
loss_values_sgd = loss_values
loss_values = []
weights1 = model.get_layer("denselayer1").get_weights() #weights and biases of the layer
L1w1=sum(sum(sum(np.abs(weights1))))
L0w1=np.count_nonzero(weights1[0]) + np.count_nonzero(weights1[1])
weights2 = model.get_layer("denselayer2").get_weights()
L1w2=sum(sum(sum(np.abs(weights2))))
L0w2= np.count_nonzero(weights2[0]) + np.count_nonzero(weights2[1])
weights3 = model.get_layer("denselayer3").get_weights()
L1w3=sum(sum(sum(np.abs(weights3))))
L0w3=np.count_nonzero(weights3[0]) + np.count_nonzero(weights3[1])
L0ges_sgd= L0w1+L0w2+L0w3
L1ges_sgd= L1w1+L1w2+L1w3
L0_sgd= [L0w1, L0w2, L0w3,L0ges_sgd]
L1_sgd= [L1w1, L1w2, L1w3, L1ges_sgd]
# EVALUATE
[train_loss_sgd, train_accuracy_sgd]= model.evaluate(x_train, y_train)
[test_loss_sgd, test_accuracy_sgd] = model.evaluate(x_test, y_test)
## Plot ACCURACY
plt.plot(accuracy_values_multi, 'b')
plt.plot(accuracy_values_multiadam, 'm')
plt.plot(accuracy_values_sgd, 'c')
plt.plot(accuracy_values_adam, 'g')
xs = np.linspace(1, 21, epochs)
plt.hlines(y=0.989, xmin=0, xmax=len(xs), colors='0.5', linestyles='--', lw=2)
plt.plot(history_multi.history['val_accuracy'], 'b', linestyle= 'dotted')
plt.plot(history_multiadam.history['val_accuracy'], 'm', linestyle= 'dotted')
plt.plot(history_sgd.history['val_accuracy'], 'c', linestyle= 'dotted')
plt.plot(history_adam.history['val_accuracy'], 'g', linestyle= 'dotted')
plt.title(f'Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train SMGD', 'Train MAdam','Train SGD', 'Train Adam', 'Validate SMGD', 'Validate MAdam', 'Validate SGD', 'Validate Adam'], loc='lower right')
plt.savefig(f'Acc_mnist-LR-{learning_rate:.4}-{pruning}.png')
plt.close()
## Plot Loss/Loss1
plt.plot(loss1_values_multi, 'b')
plt.plot(loss1_values_multiadam, 'm')
plt.plot(loss_values_sgd, 'c')
plt.plot(loss_values_adam, 'g')
plt.plot(history_multi.history['val_loss1'], 'b',linestyle= 'dotted')
plt.plot(history_multiadam.history['val_loss1'], 'm',linestyle= 'dotted')
plt.plot(history_sgd.history['val_loss'], 'c',linestyle= 'dotted')
plt.plot(history_adam.history['val_loss'], 'g',linestyle= 'dotted')
plt.title(f'Model Loss ')
plt.ylabel('Loss')
plt.xlabel('Epoch')
plt.legend(['Train SMGD', 'Train MAdam','Train SGD', 'Train Adam', 'Validate SMGD', 'Validate MAdam', 'Validate SGD', 'Validate Adam'], loc='lower right')
plt.savefig(f'Loss_mnist-LR-{learning_rate:.4}-{pruning}.png')
plt.close()
plt.plot(history_multi.history['c2'])
plt.plot(history_multiadam.history['c2'])
xs = np.linspace(1, 21, epochs)
plt.hlines(y=3e-4, xmin=0, xmax=len(xs), colors='0.5', linestyles='--', lw=2)
plt.title('Loss2 Weights Chosen')
plt.legend(['SMGD', 'MAdam'])
plt.ylabel('Weighting')
plt.xlabel('Epoch')
plt.savefig(f'C2_mnist-LR-{learning_rate:.4}-multi-{pruning}.png')
plt.close()
if pruning:
## Plot L0 Values
plt.plot(nonzero_weights1_multi, 'b')
plt.plot(nonzero_weights1_multiadam, 'm')
plt.plot(nonzero_weights1_sgd, 'c')
plt.plot(nonzero_weights1_adam, 'g')
xs = np.linspace(1, 21, 35)
plt.hlines(y=2107, xmin=0, xmax=len(xs), colors='0.5', linestyles='--', lw=2)
plt.title(f'Nonzero Weights Layer1')
plt.ylabel('Amount of Nonzeros Weights')
plt.xlabel('Epoch')
plt.legend(['SMGD', 'MAdam','SGD', 'Adam'], loc='upper right')
plt.savefig(f'Nonzeros1_mnist-LR-{learning_rate:.4}-pruning.png')
plt.close()
plt.plot(nonzero_weights2_multi, 'b')
plt.plot(nonzero_weights2_multiadam, 'm')
plt.plot(nonzero_weights2_sgd, 'c')
plt.plot(nonzero_weights2_adam, 'g')
xs = np.linspace(1, 21, 35)
plt.hlines(y=300, xmin=0, xmax=len(xs), colors='0.5', linestyles='--', lw=2)
plt.axis([0, 35, 0, 11000])
plt.title(f'Nonzero Weights Layer2')
plt.ylabel('Amount of Nonzeros Weights')
plt.xlabel('Epoch')
plt.legend(['SMGD', 'MAdam','SGD', 'Adam'], loc='upper right')
plt.savefig(f'Nonzeros2_mnist-LR-{learning_rate:.4}-pruning.png')
plt.close()
plt.plot(nonzero_weights3_multi, 'b')
plt.plot(nonzero_weights3_multiadam, 'm')
plt.plot(nonzero_weights3_sgd, 'c')
plt.plot(nonzero_weights3_adam, 'g')
xs = np.linspace(1, 21, 35)
plt.hlines(y=170, xmin=0, xmax=len(xs), colors='0.5', linestyles='--', lw=2)
plt.title(f'Nonzero Weights Layer3')
plt.ylabel('Amount of Nonzeros Weights')
plt.xlabel('Epoch')
plt.legend(['SMGD', 'MAdam','SGD', 'Adam'], loc='upper right')
plt.savefig(f'Nonzeros3_mnist-LR-{learning_rate:.4}-pruning.png')
plt.close()
del model
###################################### CIFAR Training ###########################################################
elif cifar10: # test performance of MRMSProp and RMSProp optimizer on the same problem (optional SMGD and SGD as well)
for opt in optimizers:
# start with the same init weights
seed(1)
set_random_seed(2)
#collect values while training (in evaluation mode)
loss1_values = []
loss2_values = []
loss_values=[]
val_loss_values = []
accuracy_values = []
val_accuracy_values = []
nonzero_weights = []
if opt == 'multi':
weight_decay=1e-6
decay_rate = learning_rate / epochs
momentum = 0.9
model=vggnetmultimodel(input_shape=input_shape, weight_decay=weight_decay)
model.mcompile(optimizer=SMGD(multi=True, split=False, learning_rate=learning_rate, descent_weight1=1, descent_weight2=3e-2, momentum=momentum, decay=decay_rate), loss1='sparse_categorical_crossentropy', loss2=L1lossDense(model),metrics=['accuracy'])
elif opt == 'multirms':
weight_decay=1e-6
learning_rate_rmsprop = learning_rate * 0.1
model=vggnetmultimodel(input_shape=input_shape, weight_decay=weight_decay)
model.mcompile(optimizer=MRMSprop(multi=True, split=False, learning_rate=learning_rate_rmsprop, descent_weight1=1, descent_weight2=3e-2), loss1='sparse_categorical_crossentropy', loss2=L1lossDense(model),metrics=['accuracy'])
elif opt == 'sgd':
weight_decay=1e-6
decay_rate = learning_rate / epochs
momentum = 0.9
lambdas=0.03
model=vggnetregmodel(input_shape= input_shape, weight_decay=weight_decay,lambdas=lambdas)
model.compile(optimizer=SGD(learning_rate=learning_rate, momentum=momentum, decay=decay_rate),loss='sparse_categorical_crossentropy',metrics=['accuracy'])
elif opt == 'rms':
weight_decay=1e-6 ##conv
lambdas=0.03 ##dense
learning_rate_rmsprop=learning_rate*0.1
model=vggnetregmodel(input_shape= input_shape, weight_decay=weight_decay,lambdas=lambdas)
model.compile(optimizer=RMSprop(learning_rate=learning_rate_rmsprop),loss='sparse_categorical_crossentropy',metrics=['accuracy'])
## Define Callbacks (all thresholds same)
threshold=0.001
safe_nonzeroweights=LambdaCallback(on_epoch_end= lambda epoch,
logs: [nonzero_weights.append([np.count_nonzero(
model.get_layer('denselayer').get_weights()[0])])
]
)
weight_callback_batch=LambdaCallback(on_batch_end= lambda batch,
logs: [model.get_layer(f"{name}").set_weights(update_weights(
model.get_layer(f"{name}").get_weights(),threshold))
for name in ['denselayer']],
on_epoch_end= lambda epoch,
logs: [print(np.count_nonzero(model.get_layer(f"{name}").get_weights()[0]))
for name in ['denselayer']]
)
if pruning:
## Get, Update and Set Weights before Training
weights = model.get_layer('denselayer').get_weights()
sparsified_weights = update_weights(weights, threshold)
model.get_layer('denselayer').set_weights(sparsified_weights)
nonzero_weights.append([np.count_nonzero(
model.get_layer('denselayer').get_weights()[0])])
if opt == 'multi':
if pruning:
history_multi=model.mfit(x_train, y_train, epochs=epochs,batch_size=64, validation_data=[x_test,y_test], callbacks=[weight_callback_batch,safe_nonzeroweights,TestCallback_multi((x_train,y_train))])
nonzero_weights_multi= nonzero_weights
nonzero_weights=[]
else:
history_multi=model.mfit(x_train, y_train, epochs=epochs,batch_size=64, validation_data=[x_test,y_test], callbacks=[TestCallback_multi((x_train,y_train))])
accuracy_values_multi= accuracy_values
accuracy_values=[]
loss2_values_multi = loss2_values
loss2_values = []
loss1_values_multi=loss1_values
loss1_values=[]
weights = model.get_layer("denselayer").get_weights() #weights and biases of the layer
L1w1=sum(sum(sum(np.abs(weights))))
L0w1=np.count_nonzero(weights[0]) + np.count_nonzero(weights[1])
L0ges_multi=L0w1
L1ges_multi= L1w1
L0L1_multi=[L0ges_multi,L1ges_multi]
# EVALUATE
[train_loss1_multi, train_loss2_multi, train_accuracy_multi] = model.evaluate_multi(x_train, y_train)
[test_loss1_multi, test_loss2_multi, test_accuracy_multi] = model.evaluate_multi(x_test, y_test)
elif opt == 'multirms':
if pruning:
history_multirms= model.mfit(x_train, y_train, epochs=epochs, batch_size=64, validation_data=[x_test,y_test], callbacks=[LearningRateScheduler(lr_schedule_cifar),weight_callback_batch,safe_nonzeroweights,TestCallback_multi((x_train,y_train))])
nonzero_weights_multirms= nonzero_weights
nonzero_weights=[]
else:
history_multirms=model.mfit(x_train, y_train, epochs=epochs,batch_size=64, validation_data=[x_test,y_test], callbacks=[LearningRateScheduler(lr_schedule_cifar),TestCallback_multi((x_train,y_train))])
accuracy_values_multirms= accuracy_values
accuracy_values=[]
loss2_values_multirms = loss2_values
loss2_values = []
loss1_values_multirms=loss1_values
loss1_values=[]
weights = model.get_layer("denselayer").get_weights()
L1w1=sum(sum(sum(np.abs(weights))))
L0w1=np.count_nonzero(weights[0]) + np.count_nonzero(weights[1])
L0ges_multirms=L0w1
L1ges_multirms= L1w1
L0L1_multirms=[L0ges_multirms,L1ges_multirms]
# EVALUATE
[train_loss1_multirms, train_loss2_multirms, train_accuracy_multirms] = model.evaluate_multi(x_train, y_train)
[test_loss1_multirms, test_loss2_multirms, test_accuracy_multirms] = model.evaluate_multi(x_test, y_test)
elif opt == 'sgd':
if pruning:
history= model.fit(x_train, y_train,batch_size=64,epochs=epochs, validation_data=[x_test,y_test], callbacks=[LearningRateScheduler(lr_schedule_cifar),weight_callback_batch,safe_nonzeroweights,TestCallbackSGD((x_train,y_train))])
nonzero_weights_SGD= nonzero_weights
nonzero_weights=[]
else:
history= model.fit(x_train, y_train, epochs=epochs, validation_data=[x_test,y_test], callbacks=[LearningRateScheduler(lr_schedule_cifar),TestCallback((x_train,y_train))])
accuracy_values_SGD= accuracy_values
accuracy_values=[]
loss_values_SGD= loss_values
loss_values=[]
weights = model.get_layer("denselayer").get_weights()
L1w1=sum(sum(sum(np.abs(weights))))
L0w1=np.count_nonzero(weights[0]) + np.count_nonzero(weights[1])
L0ges= L0w1
L1ges= L1w1
L0L1=[L0ges,L1ges]
# EVALUATE
[train_loss, train_accuracy]= model.evaluate(x_train, y_train)
[test_loss, test_accuracy] = model.evaluate(x_test, y_test)
elif opt == 'rms':
if pruning:
history_rms= model.fit(x_train, y_train,batch_size=64,epochs=epochs, validation_data=[x_test,y_test], callbacks=[LearningRateScheduler(lr_schedule_cifar),weight_callback_batch,safe_nonzeroweights,TestCallback((x_train,y_train))])
nonzero_weights_rms= nonzero_weights
nonzero_weights=[]
else:
history_rms= model.fit(x_train, y_train, epochs=epochs, validation_data=[x_test,y_test], callbacks=[LearningRateScheduler(lr_schedule_cifar),TestCallback((x_train,y_train))])
accuracy_values_rms= accuracy_values
accuracy_values=[]
loss_values_rms= loss_values
loss_values=[]
weights = model.get_layer("denselayer").get_weights()
L1w1=sum(sum(sum(np.abs(weights))))
L0w1=np.count_nonzero(weights[0]) + np.count_nonzero(weights[1])
L0ges_rms= L0w1
L1ges_rms= L1w1
L0L1_rms=[L0ges_rms,L1ges_rms]
# EVALUATE
[train_loss_rms, train_accuracy_rms]= model.evaluate(x_train, y_train)
[test_loss_rms, test_accuracy_rms] = model.evaluate(x_test, y_test)
## Plot Accuracy
plt.plot(accuracy_values_multi, 'g')
plt.plot(accuracy_values_multirms, 'm')
plt.plot(accuracy_values_SGD, 'b')
plt.plot(accuracy_values_rms, 'c')
plt.plot(history_multi.history['val_accuracy'], 'g', linestyle= 'dotted')
plt.plot(history_multirms.history['val_accuracy'], 'm', linestyle= 'dotted')
plt.plot(history.history['val_accuracy'], color='b', linestyle= 'dotted')
plt.plot(history_rms.history['val_accuracy'], color='c', linestyle= 'dotted')
plt.title(f'Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train SMGD','Train MRMSProp','Train SGD','Train RMSProp','Validate SMGD','Validate MRMSProp','Validate SGD','Validate RMSProp'], loc='lower right')
plt.savefig(f'Acc_multi+pruning_cifar-LR-{learning_rate:.4}-{pruning}.png')
plt.close()
## Plot Loss/Loss1
plt.plot(loss1_values_multi, 'g')
plt.plot(loss1_values_multirms, 'm')
plt.plot(loss_values_SGD,color='b')
plt.plot(loss_values_rms, color='c')
plt.plot(history_multi.history['val_loss1'], 'g', linestyle='dotted')
plt.plot(history_multirms.history['val_loss1'], 'm', linestyle= 'dotted')
plt.plot(history.history['val_loss'], color= 'b', linestyle='dotted')
plt.plot(history_rms.history['val_loss'], linestyle= 'dotted',color='c')
plt.title(f'Model Loss')
plt.ylabel('Loss')
plt.xlabel('Epoch')
plt.legend(['Train SMGD','Train MRMSProp','Train SGD','Train RMSProp','Validate SMGD','Validate MRMSProp','Validate SGD','Validate RMSProp'], loc='lower right')
plt.savefig(f'Loss_cifar-LR-{learning_rate:.4}-{pruning}.png')
plt.close()
if pruning:
## Plot L0 Values
plt.plot(nonzero_weights_multi, 'g')
plt.plot(nonzero_weights_multirms, 'm')
plt.plot(nonzero_weights_SGD, 'b')
plt.plot(nonzero_weights_rms, color='c')
plt.title(f'Nonzero Weights Layer1 t={learning_rate:.4}')
plt.ylabel('Amount of Nonzeros Weights')
plt.xlabel('Epoch')
plt.legend(['SMGD', 'MRMSProp', 'SGD', 'RMSProp'], loc='upper right')
plt.savefig(f'Nonzeros_cifar-LR-{learning_rate:.4}-{pruning}.png')
plt.close()
plt.plot(history_multi.history['c2'], 'g')
plt.plot(history_multirms.history['c2'], 'm')
plt.legend(['SMGD','MRMSProp'])
plt.title('Loss Weight C2 Chosen')
plt.ylabel('Weighting')
plt.xlabel('Epoch')
plt.savefig(f'C2_cifar-LR-{learning_rate:.4}-multi-{pruning}.png')
plt.close()
del model
else:
print("Please specify a data set!")