forked from wavefrontshaping/complexPyTorch
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathcomplexLayers.py
More file actions
executable file
·480 lines (376 loc) · 18.6 KB
/
Copy pathcomplexLayers.py
File metadata and controls
executable file
·480 lines (376 loc) · 18.6 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Tue Mar 19 10:30:02 2019
@author: Sebastien M. Popoff
Based on https://openreview.net/forum?id=H1T2hmZAb
"""
import torch
from torch.nn import Module, Parameter, init
from torch.nn import Conv2d, Linear, BatchNorm1d, BatchNorm2d
from torch.nn import ConvTranspose2d
<<<<<<< HEAD:complexPyTorch/complexLayers.py
from .complexFunctions import complex_relu, complex_max_pool2d, complex_avg_pool2d
from .complexFunctions import complex_dropout, complex_dropout2d
=======
from complexFunctions import (
complex_relu,
complex_tanh,
complex_sigmoid,
complex_max_pool2d,
complex_avg_pool2d,
complex_dropout,
complex_dropout2d,
complex_opposite,
)
>>>>>>> d00e50c5d93386dd85b363a39c24cf783b7914aa:complexLayers.py
def apply_complex(fr, fi, input, dtype = torch.complex64):
return (fr(input.real)-fi(input.imag)).type(dtype) \
+ 1j*(fr(input.imag)+fi(input.real)).type(dtype)
class ComplexDropout(Module):
def __init__(self,p=0.5):
super(ComplexDropout,self).__init__()
self.p = p
def forward(self,input):
if self.training:
return complex_dropout(input,self.p)
else:
return input
class ComplexDropout2d(Module):
def __init__(self,p=0.5):
super(ComplexDropout2d,self).__init__()
self.p = p
def forward(self,input):
if self.training:
return complex_dropout2d(input,self.p)
else:
return input
class ComplexMaxPool2d(Module):
def __init__(self,kernel_size, stride= None, padding = 0,
dilation = 1, return_indices = False, ceil_mode = False):
super(ComplexMaxPool2d,self).__init__()
self.kernel_size = kernel_size
self.stride = stride
self.padding = padding
self.dilation = dilation
self.ceil_mode = ceil_mode
self.return_indices = return_indices
def forward(self,input):
return complex_max_pool2d(input,kernel_size = self.kernel_size,
stride = self.stride, padding = self.padding,
dilation = self.dilation, ceil_mode = self.ceil_mode,
return_indices = self.return_indices)
class ComplexAvgPool2d(Module):
def __init__(self,kernel_size, stride= None, padding = 0,
dilation = 1, return_indices = False, ceil_mode = False):
super(ComplexAvgPool2d,self).__init__()
self.kernel_size = kernel_size
self.stride = stride
self.padding = padding
self.dilation = dilation
self.ceil_mode = ceil_mode
self.return_indices = return_indices
def forward(self,input):
return complex_avg_pool2d(input,kernel_size = self.kernel_size,
stride = self.stride, padding = self.padding,
dilation = self.dilation, ceil_mode = self.ceil_mode,
return_indices = self.return_indices)
class ComplexReLU(Module):
def forward(self,input):
return complex_relu(input)
class ComplexSigmoid(Module):
def forward(self,input):
return complex_sigmoid(input)
class ComplexTanh(Module):
def forward(self,input):
return complex_tanh(input)
class ComplexConvTranspose2d(Module):
def __init__(self,in_channels, out_channels, kernel_size, stride=1, padding=0,
output_padding=0, groups=1, bias=True, dilation=1, padding_mode='zeros'):
super(ComplexConvTranspose2d, self).__init__()
self.conv_tran_r = ConvTranspose2d(in_channels, out_channels, kernel_size, stride, padding,
output_padding, groups, bias, dilation, padding_mode)
self.conv_tran_i = ConvTranspose2d(in_channels, out_channels, kernel_size, stride, padding,
output_padding, groups, bias, dilation, padding_mode)
def forward(self,input):
return apply_complex(self.conv_tran_r, self.conv_tran_i, input)
class ComplexConv2d(Module):
def __init__(self,in_channels, out_channels, kernel_size=3, stride=1, padding = 0,
dilation=1, groups=1, bias=True):
super(ComplexConv2d, self).__init__()
self.conv_r = Conv2d(in_channels, out_channels, kernel_size, stride, padding, dilation, groups, bias)
self.conv_i = Conv2d(in_channels, out_channels, kernel_size, stride, padding, dilation, groups, bias)
def forward(self,input):
return apply_complex(self.conv_r, self.conv_i, input)
class ComplexLinear(Module):
def __init__(self, in_features, out_features):
super(ComplexLinear, self).__init__()
self.fc_r = Linear(in_features, out_features)
self.fc_i = Linear(in_features, out_features)
def forward(self, input):
return apply_complex(self.fc_r, self.fc_i, input)
class NaiveComplexBatchNorm1d(Module):
'''
Naive approach to complex batch norm, perform batch norm independently on real and imaginary part.
'''
def __init__(self, num_features, eps=1e-5, momentum=0.1, affine=True, \
track_running_stats=True):
super(NaiveComplexBatchNorm1d, self).__init__()
self.bn_r = BatchNorm1d(num_features, eps, momentum, affine, track_running_stats)
self.bn_i = BatchNorm1d(num_features, eps, momentum, affine, track_running_stats)
def forward(self,input):
return self.bn_r(input.real).type(torch.complex64) +1j*self.bn_i(input.imag).type(torch.complex64)
class NaiveComplexBatchNorm2d(Module):
'''
Naive approach to complex batch norm, perform batch norm independently on real and imaginary part.
'''
def __init__(self, num_features, eps=1e-5, momentum=0.1, affine=True, \
track_running_stats=True):
super(NaiveComplexBatchNorm2d, self).__init__()
self.bn_r = BatchNorm2d(num_features, eps, momentum, affine, track_running_stats)
self.bn_i = BatchNorm2d(num_features, eps, momentum, affine, track_running_stats)
def forward(self,input):
return self.bn_r(input.real).type(torch.complex64) +1j*self.bn_i(input.imag).type(torch.complex64)
class _ComplexBatchNorm(Module):
def __init__(self, num_features, eps=1e-5, momentum=0.1, affine=True,
track_running_stats=True):
super(_ComplexBatchNorm, self).__init__()
self.num_features = num_features
self.eps = eps
self.momentum = momentum
self.affine = affine
self.track_running_stats = track_running_stats
if self.affine:
self.weight = Parameter(torch.Tensor(num_features,3))
self.bias = Parameter(torch.Tensor(num_features,2))
else:
self.register_parameter('weight', None)
self.register_parameter('bias', None)
if self.track_running_stats:
self.register_buffer('running_mean', torch.zeros(num_features, dtype = torch.complex64))
self.register_buffer('running_covar', torch.zeros(num_features,3))
self.running_covar[:,0] = 1.4142135623730951
self.running_covar[:,1] = 1.4142135623730951
self.register_buffer('num_batches_tracked', torch.tensor(0, dtype=torch.long))
else:
self.register_parameter('running_mean', None)
self.register_parameter('running_covar', None)
self.register_parameter('num_batches_tracked', None)
self.reset_parameters()
def reset_running_stats(self):
if self.track_running_stats:
self.running_mean.zero_()
self.running_covar.zero_()
self.running_covar[:,0] = 1.4142135623730951
self.running_covar[:,1] = 1.4142135623730951
self.num_batches_tracked.zero_()
def reset_parameters(self):
self.reset_running_stats()
if self.affine:
init.constant_(self.weight[:,:2],1.4142135623730951)
init.zeros_(self.weight[:,2])
init.zeros_(self.bias)
class ComplexBatchNorm2d(_ComplexBatchNorm):
def forward(self, input):
exponential_average_factor = 0.0
if self.training and self.track_running_stats:
if self.num_batches_tracked is not None:
self.num_batches_tracked += 1
if self.momentum is None: # use cumulative moving average
exponential_average_factor = 1.0 / float(self.num_batches_tracked)
else: # use exponential moving average
exponential_average_factor = self.momentum
if self.training or (not self.training and not self.track_running_stats):
# calculate mean of real and imaginary part
# mean does not support automatic differentiation for outputs with complex dtype.
mean_r = input.real.mean([0, 2, 3]).type(torch.complex64)
mean_i = input.imag.mean([0, 2, 3]).type(torch.complex64)
mean = mean_r + 1j*mean_i
else:
mean = self.running_mean
if self.training and self.track_running_stats:
# update running mean
with torch.no_grad():
self.running_mean = exponential_average_factor * mean\
+ (1 - exponential_average_factor) * self.running_mean
input = input - mean[None, :, None, None]
if self.training or (not self.training and not self.track_running_stats):
# Elements of the covariance matrix (biased for train)
n = input.numel() / input.size(1)
Crr = 1./n*input.real.pow(2).sum(dim=[0,2,3])+self.eps
Cii = 1./n*input.imag.pow(2).sum(dim=[0,2,3])+self.eps
Cri = (input.real.mul(input.imag)).mean(dim=[0,2,3])
else:
Crr = self.running_covar[:,0]+self.eps
Cii = self.running_covar[:,1]+self.eps
Cri = self.running_covar[:,2]#+self.eps
if self.training and self.track_running_stats:
with torch.no_grad():
self.running_covar[:,0] = exponential_average_factor * Crr * n / (n - 1)\
+ (1 - exponential_average_factor) * self.running_covar[:,0]
self.running_covar[:,1] = exponential_average_factor * Cii * n / (n - 1)\
+ (1 - exponential_average_factor) * self.running_covar[:,1]
self.running_covar[:,2] = exponential_average_factor * Cri * n / (n - 1)\
+ (1 - exponential_average_factor) * self.running_covar[:,2]
# calculate the inverse square root the covariance matrix
det = Crr*Cii-Cri.pow(2)
s = torch.sqrt(det)
t = torch.sqrt(Cii+Crr + 2 * s)
inverse_st = 1.0 / (s * t)
Rrr = (Cii + s) * inverse_st
Rii = (Crr + s) * inverse_st
Rri = -Cri * inverse_st
input = (Rrr[None,:,None,None]*input.real+Rri[None,:,None,None]*input.imag).type(torch.complex64) \
+ 1j*(Rii[None,:,None,None]*input.imag+Rri[None,:,None,None]*input.real).type(torch.complex64)
if self.affine:
input = (self.weight[None,:,0,None,None]*input.real+self.weight[None,:,2,None,None]*input.imag+\
self.bias[None,:,0,None,None]).type(torch.complex64) \
+1j*(self.weight[None,:,2,None,None]*input.real+self.weight[None,:,1,None,None]*input.imag+\
self.bias[None,:,1,None,None]).type(torch.complex64)
return input
class ComplexBatchNorm1d(_ComplexBatchNorm):
def forward(self, input):
exponential_average_factor = 0.0
if self.training and self.track_running_stats:
if self.num_batches_tracked is not None:
self.num_batches_tracked += 1
if self.momentum is None: # use cumulative moving average
exponential_average_factor = 1.0 / float(self.num_batches_tracked)
else: # use exponential moving average
exponential_average_factor = self.momentum
if self.training or (not self.training and not self.track_running_stats):
# calculate mean of real and imaginary part
mean_r = input.real.mean(dim=0).type(torch.complex64)
mean_i = input.imag.mean(dim=0).type(torch.complex64)
mean = mean_r + 1j*mean_i
else:
mean = self.running_mean
if self.training and self.track_running_stats:
# update running mean
with torch.no_grad():
self.running_mean = exponential_average_factor * mean\
+ (1 - exponential_average_factor) * self.running_mean
input = input - mean[None, ...]
if self.training or (not self.training and not self.track_running_stats):
# Elements of the covariance matrix (biased for train)
n = input.numel() / input.size(1)
Crr = input.real.var(dim=0,unbiased=False)+self.eps
Cii = input.imag.var(dim=0,unbiased=False)+self.eps
Cri = (input.real.mul(input.imag)).mean(dim=0)
else:
Crr = self.running_covar[:,0]+self.eps
Cii = self.running_covar[:,1]+self.eps
Cri = self.running_covar[:,2]
if self.training and self.track_running_stats:
self.running_covar[:,0] = exponential_average_factor * Crr * n / (n - 1)\
+ (1 - exponential_average_factor) * self.running_covar[:,0]
self.running_covar[:,1] = exponential_average_factor * Cii * n / (n - 1)\
+ (1 - exponential_average_factor) * self.running_covar[:,1]
self.running_covar[:,2] = exponential_average_factor * Cri * n / (n - 1)\
+ (1 - exponential_average_factor) * self.running_covar[:,2]
# calculate the inverse square root the covariance matrix
det = Crr*Cii-Cri.pow(2)
s = torch.sqrt(det)
t = torch.sqrt(Cii+Crr + 2 * s)
inverse_st = 1.0 / (s * t)
Rrr = (Cii + s) * inverse_st
Rii = (Crr + s) * inverse_st
Rri = -Cri * inverse_st
input = (Rrr[None,:]*input.real+Rri[None,:]*input.imag).type(torch.complex64) \
+ 1j*(Rii[None,:]*input.imag+Rri[None,:]*input.real).type(torch.complex64)
if self.affine:
input = (self.weight[None,:,0]*input.real+self.weight[None,:,2]*input.imag+\
self.bias[None,:,0]).type(torch.complex64) \
+1j*(self.weight[None,:,2]*input.real+self.weight[None,:,1]*input.imag+\
self.bias[None,:,1]).type(torch.complex64)
del Crr, Cri, Cii, Rrr, Rii, Rri, det, s, t
return input
class ComplexGRUCell(Module):
"""
A GRU cell for complex-valued inputs
"""
def __init__(self, input_length=10, hidden_length=20):
super(ComplexGRUCell, self).__init__()
self.input_length = input_length
self.hidden_length = hidden_length
# reset gate components
self.linear_reset_w1 = ComplexLinear(self.input_length, self.hidden_length)
self.linear_reset_r1 = ComplexLinear(self.hidden_length, self.hidden_length)
self.linear_reset_w2 = ComplexLinear(self.input_length, self.hidden_length)
self.linear_reset_r2 = ComplexLinear(self.hidden_length, self.hidden_length)
# update gate components
self.linear_gate_w3 = ComplexLinear(self.input_length, self.hidden_length)
self.linear_gate_r3 = ComplexLinear(self.hidden_length, self.hidden_length)
self.activation_gate = ComplexSigmoid()
self.activation_candidate = ComplexTanh()
def reset_gate(self, x, h):
x_1 = self.linear_reset_w1(x)
h_1 = self.linear_reset_r1(h)
# gate update
reset = self.activation_gate(x_1 + h_1)
return reset
def update_gate(self, x, h):
x_2 = self.linear_reset_w2(x)
h_2 = self.linear_reset_r2(h)
z = self.activation_gate(h_2 + x_2)
return z
def update_component(self, x, h, r):
x_3 = self.linear_gate_w3(x)
h_3 = r * self.linear_gate_r3(h) # element-wise multiplication
gate_update = self.activation_candidate(x_3 + h_3)
return gate_update
def forward(self, x, h):
# Equation 1. reset gate vector
r = self.reset_gate(x, h)
# Equation 2: the update gate - the shared update gate vector z
z = self.update_gate(x, h)
# Equation 3: The almost output component
n = self.update_component(x, h, r)
# Equation 4: the new hidden state
h_new = (1 + complex_opposite(z)) * n + z * h # element-wise multiplication
return h_new
class ComplexBNGRUCell(Module):
"""
A BN-GRU cell for complex-valued inputs
"""
def __init__(self, input_length=10, hidden_length=20):
super(ComplexBNGRUCell, self).__init__()
self.input_length = input_length
self.hidden_length = hidden_length
# reset gate components
self.linear_reset_w1 = ComplexLinear(self.input_length, self.hidden_length)
self.linear_reset_r1 = ComplexLinear(self.hidden_length, self.hidden_length)
self.linear_reset_w2 = ComplexLinear(self.input_length, self.hidden_length)
self.linear_reset_r2 = ComplexLinear(self.hidden_length, self.hidden_length)
# update gate components
self.linear_gate_w3 = ComplexLinear(self.input_length, self.hidden_length)
self.linear_gate_r3 = ComplexLinear(self.hidden_length, self.hidden_length)
self.activation_gate = ComplexSigmoid()
self.activation_candidate = ComplexTanh()
self.bn = ComplexBatchNorm2d(1)
def reset_gate(self, x, h):
x_1 = self.linear_reset_w1(x)
h_1 = self.linear_reset_r1(h)
# gate update
reset = self.activation_gate(self.bn(x_1) + self.bn(h_1))
return reset
def update_gate(self, x, h):
x_2 = self.linear_reset_w2(x)
h_2 = self.linear_reset_r2(h)
z = self.activation_gate(self.bn(h_2) + self.bn(x_2))
return z
def update_component(self, x, h, r):
x_3 = self.linear_gate_w3(x)
h_3 = r * self.bn(self.linear_gate_r3(h)) # element-wise multiplication
gate_update = self.activation_candidate(self.bn(self.bn(x_3) + h_3))
return gate_update
def forward(self, x, h):
# Equation 1. reset gate vector
r = self.reset_gate(x, h)
# Equation 2: the update gate - the shared update gate vector z
z = self.update_gate(x, h)
# Equation 3: The almost output component
n = self.update_component(x, h, r)
# Equation 4: the new hidden state
h_new = (1 + complex_opposite(z)) * n + z * h # element-wise multiplication
return h_new