-
Notifications
You must be signed in to change notification settings - Fork 4.6k
/
layers.py
733 lines (588 loc) · 26.9 KB
/
layers.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
from __future__ import print_function, division
import math
import numpy as np
import copy
from mlfromscratch.deep_learning.activation_functions import Sigmoid, ReLU, SoftPlus, LeakyReLU
from mlfromscratch.deep_learning.activation_functions import TanH, ELU, SELU, Softmax
class Layer(object):
def set_input_shape(self, shape):
""" Sets the shape that the layer expects of the input in the forward
pass method """
self.input_shape = shape
def layer_name(self):
""" The name of the layer. Used in model summary. """
return self.__class__.__name__
def parameters(self):
""" The number of trainable parameters used by the layer """
return 0
def forward_pass(self, X, training):
""" Propogates the signal forward in the network """
raise NotImplementedError()
def backward_pass(self, accum_grad):
""" Propogates the accumulated gradient backwards in the network.
If the has trainable weights then these weights are also tuned in this method.
As input (accum_grad) it receives the gradient with respect to the output of the layer and
returns the gradient with respect to the output of the previous layer. """
raise NotImplementedError()
def output_shape(self):
""" The shape of the output produced by forward_pass """
raise NotImplementedError()
class Dense(Layer):
"""A fully-connected NN layer.
Parameters:
-----------
n_units: int
The number of neurons in the layer.
input_shape: tuple
The expected input shape of the layer. For dense layers a single digit specifying
the number of features of the input. Must be specified if it is the first layer in
the network.
"""
def __init__(self, n_units, input_shape=None):
self.layer_input = None
self.input_shape = input_shape
self.n_units = n_units
self.trainable = True
self.W = None
self.w0 = None
def initialize(self, optimizer):
# Initialize the weights
limit = 1 / math.sqrt(self.input_shape[0])
self.W = np.random.uniform(-limit, limit, (self.input_shape[0], self.n_units))
self.w0 = np.zeros((1, self.n_units))
# Weight optimizers
self.W_opt = copy.copy(optimizer)
self.w0_opt = copy.copy(optimizer)
def parameters(self):
return np.prod(self.W.shape) + np.prod(self.w0.shape)
def forward_pass(self, X, training=True):
self.layer_input = X
return X.dot(self.W) + self.w0
def backward_pass(self, accum_grad):
# Save weights used during forwards pass
W = self.W
if self.trainable:
# Calculate gradient w.r.t layer weights
grad_w = self.layer_input.T.dot(accum_grad)
grad_w0 = np.sum(accum_grad, axis=0, keepdims=True)
# Update the layer weights
self.W = self.W_opt.update(self.W, grad_w)
self.w0 = self.w0_opt.update(self.w0, grad_w0)
# Return accumulated gradient for next layer
# Calculated based on the weights used during the forward pass
accum_grad = accum_grad.dot(W.T)
return accum_grad
def output_shape(self):
return (self.n_units, )
class RNN(Layer):
"""A Vanilla Fully-Connected Recurrent Neural Network layer.
Parameters:
-----------
n_units: int
The number of hidden states in the layer.
activation: string
The name of the activation function which will be applied to the output of each state.
bptt_trunc: int
Decides how many time steps the gradient should be propagated backwards through states
given the loss gradient for time step t.
input_shape: tuple
The expected input shape of the layer. For dense layers a single digit specifying
the number of features of the input. Must be specified if it is the first layer in
the network.
Reference:
http://www.wildml.com/2015/09/recurrent-neural-networks-tutorial-part-2-implementing-a-language-model-rnn-with-python-numpy-and-theano/
"""
def __init__(self, n_units, activation='tanh', bptt_trunc=5, input_shape=None):
self.input_shape = input_shape
self.n_units = n_units
self.activation = activation_functions[activation]()
self.trainable = True
self.bptt_trunc = bptt_trunc
self.W = None # Weight of the previous state
self.V = None # Weight of the output
self.U = None # Weight of the input
def initialize(self, optimizer):
timesteps, input_dim = self.input_shape
# Initialize the weights
limit = 1 / math.sqrt(input_dim)
self.U = np.random.uniform(-limit, limit, (self.n_units, input_dim))
limit = 1 / math.sqrt(self.n_units)
self.V = np.random.uniform(-limit, limit, (input_dim, self.n_units))
self.W = np.random.uniform(-limit, limit, (self.n_units, self.n_units))
# Weight optimizers
self.U_opt = copy.copy(optimizer)
self.V_opt = copy.copy(optimizer)
self.W_opt = copy.copy(optimizer)
def parameters(self):
return np.prod(self.W.shape) + np.prod(self.U.shape) + np.prod(self.V.shape)
def forward_pass(self, X, training=True):
self.layer_input = X
batch_size, timesteps, input_dim = X.shape
# Save these values for use in backprop.
self.state_input = np.zeros((batch_size, timesteps, self.n_units))
self.states = np.zeros((batch_size, timesteps+1, self.n_units))
self.outputs = np.zeros((batch_size, timesteps, input_dim))
# Set last time step to zero for calculation of the state_input at time step zero
self.states[:, -1] = np.zeros((batch_size, self.n_units))
for t in range(timesteps):
# Input to state_t is the current input and output of previous states
self.state_input[:, t] = X[:, t].dot(self.U.T) + self.states[:, t-1].dot(self.W.T)
self.states[:, t] = self.activation(self.state_input[:, t])
self.outputs[:, t] = self.states[:, t].dot(self.V.T)
return self.outputs
def backward_pass(self, accum_grad):
_, timesteps, _ = accum_grad.shape
# Variables where we save the accumulated gradient w.r.t each parameter
grad_U = np.zeros_like(self.U)
grad_V = np.zeros_like(self.V)
grad_W = np.zeros_like(self.W)
# The gradient w.r.t the layer input.
# Will be passed on to the previous layer in the network
accum_grad_next = np.zeros_like(accum_grad)
# Back Propagation Through Time
for t in reversed(range(timesteps)):
# Update gradient w.r.t V at time step t
grad_V += accum_grad[:, t].T.dot(self.states[:, t])
# Calculate the gradient w.r.t the state input
grad_wrt_state = accum_grad[:, t].dot(self.V) * self.activation.gradient(self.state_input[:, t])
# Gradient w.r.t the layer input
accum_grad_next[:, t] = grad_wrt_state.dot(self.U)
# Update gradient w.r.t W and U by backprop. from time step t for at most
# self.bptt_trunc number of time steps
for t_ in reversed(np.arange(max(0, t - self.bptt_trunc), t+1)):
grad_U += grad_wrt_state.T.dot(self.layer_input[:, t_])
grad_W += grad_wrt_state.T.dot(self.states[:, t_-1])
# Calculate gradient w.r.t previous state
grad_wrt_state = grad_wrt_state.dot(self.W) * self.activation.gradient(self.state_input[:, t_-1])
# Update weights
self.U = self.U_opt.update(self.U, grad_U)
self.V = self.V_opt.update(self.V, grad_V)
self.W = self.W_opt.update(self.W, grad_W)
return accum_grad_next
def output_shape(self):
return self.input_shape
class Conv2D(Layer):
"""A 2D Convolution Layer.
Parameters:
-----------
n_filters: int
The number of filters that will convolve over the input matrix. The number of channels
of the output shape.
filter_shape: tuple
A tuple (filter_height, filter_width).
input_shape: tuple
The shape of the expected input of the layer. (batch_size, channels, height, width)
Only needs to be specified for first layer in the network.
padding: string
Either 'same' or 'valid'. 'same' results in padding being added so that the output height and width
matches the input height and width. For 'valid' no padding is added.
stride: int
The stride length of the filters during the convolution over the input.
"""
def __init__(self, n_filters, filter_shape, input_shape=None, padding='same', stride=1):
self.n_filters = n_filters
self.filter_shape = filter_shape
self.padding = padding
self.stride = stride
self.input_shape = input_shape
self.trainable = True
def initialize(self, optimizer):
# Initialize the weights
filter_height, filter_width = self.filter_shape
channels = self.input_shape[0]
limit = 1 / math.sqrt(np.prod(self.filter_shape))
self.W = np.random.uniform(-limit, limit, size=(self.n_filters, channels, filter_height, filter_width))
self.w0 = np.zeros((self.n_filters, 1))
# Weight optimizers
self.W_opt = copy.copy(optimizer)
self.w0_opt = copy.copy(optimizer)
def parameters(self):
return np.prod(self.W.shape) + np.prod(self.w0.shape)
def forward_pass(self, X, training=True):
batch_size, channels, height, width = X.shape
self.layer_input = X
# Turn image shape into column shape
# (enables dot product between input and weights)
self.X_col = image_to_column(X, self.filter_shape, stride=self.stride, output_shape=self.padding)
# Turn weights into column shape
self.W_col = self.W.reshape((self.n_filters, -1))
# Calculate output
output = self.W_col.dot(self.X_col) + self.w0
# Reshape into (n_filters, out_height, out_width, batch_size)
output = output.reshape(self.output_shape() + (batch_size, ))
# Redistribute axises so that batch size comes first
return output.transpose(3,0,1,2)
def backward_pass(self, accum_grad):
# Reshape accumulated gradient into column shape
accum_grad = accum_grad.transpose(1, 2, 3, 0).reshape(self.n_filters, -1)
if self.trainable:
# Take dot product between column shaped accum. gradient and column shape
# layer input to determine the gradient at the layer with respect to layer weights
grad_w = accum_grad.dot(self.X_col.T).reshape(self.W.shape)
# The gradient with respect to bias terms is the sum similarly to in Dense layer
grad_w0 = np.sum(accum_grad, axis=1, keepdims=True)
# Update the layers weights
self.W = self.W_opt.update(self.W, grad_w)
self.w0 = self.w0_opt.update(self.w0, grad_w0)
# Recalculate the gradient which will be propogated back to prev. layer
accum_grad = self.W_col.T.dot(accum_grad)
# Reshape from column shape to image shape
accum_grad = column_to_image(accum_grad,
self.layer_input.shape,
self.filter_shape,
stride=self.stride,
output_shape=self.padding)
return accum_grad
def output_shape(self):
channels, height, width = self.input_shape
pad_h, pad_w = determine_padding(self.filter_shape, output_shape=self.padding)
output_height = (height + np.sum(pad_h) - self.filter_shape[0]) / self.stride + 1
output_width = (width + np.sum(pad_w) - self.filter_shape[1]) / self.stride + 1
return self.n_filters, int(output_height), int(output_width)
class BatchNormalization(Layer):
"""Batch normalization.
"""
def __init__(self, momentum=0.99):
self.momentum = momentum
self.trainable = True
self.eps = 0.01
self.running_mean = None
self.running_var = None
def initialize(self, optimizer):
# Initialize the parameters
self.gamma = np.ones(self.input_shape)
self.beta = np.zeros(self.input_shape)
# parameter optimizers
self.gamma_opt = copy.copy(optimizer)
self.beta_opt = copy.copy(optimizer)
def parameters(self):
return np.prod(self.gamma.shape) + np.prod(self.beta.shape)
def forward_pass(self, X, training=True):
# Initialize running mean and variance if first run
if self.running_mean is None:
self.running_mean = np.mean(X, axis=0)
self.running_var = np.var(X, axis=0)
if training and self.trainable:
mean = np.mean(X, axis=0)
var = np.var(X, axis=0)
self.running_mean = self.momentum * self.running_mean + (1 - self.momentum) * mean
self.running_var = self.momentum * self.running_var + (1 - self.momentum) * var
else:
mean = self.running_mean
var = self.running_var
# Statistics saved for backward pass
self.X_centered = X - mean
self.stddev_inv = 1 / np.sqrt(var + self.eps)
X_norm = self.X_centered * self.stddev_inv
output = self.gamma * X_norm + self.beta
return output
def backward_pass(self, accum_grad):
# Save parameters used during the forward pass
gamma = self.gamma
# If the layer is trainable the parameters are updated
if self.trainable:
X_norm = self.X_centered * self.stddev_inv
grad_gamma = np.sum(accum_grad * X_norm, axis=0)
grad_beta = np.sum(accum_grad, axis=0)
self.gamma = self.gamma_opt.update(self.gamma, grad_gamma)
self.beta = self.beta_opt.update(self.beta, grad_beta)
batch_size = accum_grad.shape[0]
# The gradient of the loss with respect to the layer inputs (use weights and statistics from forward pass)
accum_grad = (1 / batch_size) * gamma * self.stddev_inv * (
batch_size * accum_grad
- np.sum(accum_grad, axis=0)
- self.X_centered * self.stddev_inv**2 * np.sum(accum_grad * self.X_centered, axis=0)
)
return accum_grad
def output_shape(self):
return self.input_shape
class PoolingLayer(Layer):
"""A parent class of MaxPooling2D and AveragePooling2D
"""
def __init__(self, pool_shape=(2, 2), stride=1, padding=0):
self.pool_shape = pool_shape
self.stride = stride
self.padding = padding
self.trainable = True
def forward_pass(self, X, training=True):
self.layer_input = X
batch_size, channels, height, width = X.shape
_, out_height, out_width = self.output_shape()
X = X.reshape(batch_size*channels, 1, height, width)
X_col = image_to_column(X, self.pool_shape, self.stride, self.padding)
# MaxPool or AveragePool specific method
output = self._pool_forward(X_col)
output = output.reshape(out_height, out_width, batch_size, channels)
output = output.transpose(2, 3, 0, 1)
return output
def backward_pass(self, accum_grad):
batch_size, _, _, _ = accum_grad.shape
channels, height, width = self.input_shape
accum_grad = accum_grad.transpose(2, 3, 0, 1).ravel()
# MaxPool or AveragePool specific method
accum_grad_col = self._pool_backward(accum_grad)
accum_grad = column_to_image(accum_grad_col, (batch_size * channels, 1, height, width), self.pool_shape, self.stride, 0)
accum_grad = accum_grad.reshape((batch_size,) + self.input_shape)
return accum_grad
def output_shape(self):
channels, height, width = self.input_shape
out_height = (height - self.pool_shape[0]) / self.stride + 1
out_width = (width - self.pool_shape[1]) / self.stride + 1
assert out_height % 1 == 0
assert out_width % 1 == 0
return channels, int(out_height), int(out_width)
class MaxPooling2D(PoolingLayer):
def _pool_forward(self, X_col):
arg_max = np.argmax(X_col, axis=0).flatten()
output = X_col[arg_max, range(arg_max.size)]
self.cache = arg_max
return output
def _pool_backward(self, accum_grad):
accum_grad_col = np.zeros((np.prod(self.pool_shape), accum_grad.size))
arg_max = self.cache
accum_grad_col[arg_max, range(accum_grad.size)] = accum_grad
return accum_grad_col
class AveragePooling2D(PoolingLayer):
def _pool_forward(self, X_col):
output = np.mean(X_col, axis=0)
return output
def _pool_backward(self, accum_grad):
accum_grad_col = np.zeros((np.prod(self.pool_shape), accum_grad.size))
accum_grad_col[:, range(accum_grad.size)] = 1. / accum_grad_col.shape[0] * accum_grad
return accum_grad_col
class ConstantPadding2D(Layer):
"""Adds rows and columns of constant values to the input.
Expects the input to be of shape (batch_size, channels, height, width)
Parameters:
-----------
padding: tuple
The amount of padding along the height and width dimension of the input.
If (pad_h, pad_w) the same symmetric padding is applied along height and width dimension.
If ((pad_h0, pad_h1), (pad_w0, pad_w1)) the specified padding is added to beginning and end of
the height and width dimension.
padding_value: int or tuple
The value the is added as padding.
"""
def __init__(self, padding, padding_value=0):
self.padding = padding
self.trainable = True
if not isinstance(padding[0], tuple):
self.padding = ((padding[0], padding[0]), padding[1])
if not isinstance(padding[1], tuple):
self.padding = (self.padding[0], (padding[1], padding[1]))
self.padding_value = padding_value
def forward_pass(self, X, training=True):
output = np.pad(X,
pad_width=((0,0), (0,0), self.padding[0], self.padding[1]),
mode="constant",
constant_values=self.padding_value)
return output
def backward_pass(self, accum_grad):
pad_top, pad_left = self.padding[0][0], self.padding[1][0]
height, width = self.input_shape[1], self.input_shape[2]
accum_grad = accum_grad[:, :, pad_top:pad_top+height, pad_left:pad_left+width]
return accum_grad
def output_shape(self):
new_height = self.input_shape[1] + np.sum(self.padding[0])
new_width = self.input_shape[2] + np.sum(self.padding[1])
return (self.input_shape[0], new_height, new_width)
class ZeroPadding2D(ConstantPadding2D):
"""Adds rows and columns of zero values to the input.
Expects the input to be of shape (batch_size, channels, height, width)
Parameters:
-----------
padding: tuple
The amount of padding along the height and width dimension of the input.
If (pad_h, pad_w) the same symmetric padding is applied along height and width dimension.
If ((pad_h0, pad_h1), (pad_w0, pad_w1)) the specified padding is added to beginning and end of
the height and width dimension.
"""
def __init__(self, padding):
self.padding = padding
if isinstance(padding[0], int):
self.padding = ((padding[0], padding[0]), padding[1])
if isinstance(padding[1], int):
self.padding = (self.padding[0], (padding[1], padding[1]))
self.padding_value = 0
class Flatten(Layer):
""" Turns a multidimensional matrix into two-dimensional """
def __init__(self, input_shape=None):
self.prev_shape = None
self.trainable = True
self.input_shape = input_shape
def forward_pass(self, X, training=True):
self.prev_shape = X.shape
return X.reshape((X.shape[0], -1))
def backward_pass(self, accum_grad):
return accum_grad.reshape(self.prev_shape)
def output_shape(self):
return (np.prod(self.input_shape),)
class UpSampling2D(Layer):
""" Nearest neighbor up sampling of the input. Repeats the rows and
columns of the data by size[0] and size[1] respectively.
Parameters:
-----------
size: tuple
(size_y, size_x) - The number of times each axis will be repeated.
"""
def __init__(self, size=(2,2), input_shape=None):
self.prev_shape = None
self.trainable = True
self.size = size
self.input_shape = input_shape
def forward_pass(self, X, training=True):
self.prev_shape = X.shape
# Repeat each axis as specified by size
X_new = X.repeat(self.size[0], axis=2).repeat(self.size[1], axis=3)
return X_new
def backward_pass(self, accum_grad):
# Down sample input to previous shape
accum_grad = accum_grad[:, :, ::self.size[0], ::self.size[1]]
return accum_grad
def output_shape(self):
channels, height, width = self.input_shape
return channels, self.size[0] * height, self.size[1] * width
class Reshape(Layer):
""" Reshapes the input tensor into specified shape
Parameters:
-----------
shape: tuple
The shape which the input shall be reshaped to.
"""
def __init__(self, shape, input_shape=None):
self.prev_shape = None
self.trainable = True
self.shape = shape
self.input_shape = input_shape
def forward_pass(self, X, training=True):
self.prev_shape = X.shape
return X.reshape((X.shape[0], ) + self.shape)
def backward_pass(self, accum_grad):
return accum_grad.reshape(self.prev_shape)
def output_shape(self):
return self.shape
class Dropout(Layer):
"""A layer that randomly sets a fraction p of the output units of the previous layer
to zero.
Parameters:
-----------
p: float
The probability that unit x is set to zero.
"""
def __init__(self, p=0.2):
self.p = p
self._mask = None
self.input_shape = None
self.n_units = None
self.pass_through = True
self.trainable = True
def forward_pass(self, X, training=True):
c = (1 - self.p)
if training:
self._mask = np.random.uniform(size=X.shape) > self.p
c = self._mask
return X * c
def backward_pass(self, accum_grad):
return accum_grad * self._mask
def output_shape(self):
return self.input_shape
activation_functions = {
'relu': ReLU,
'sigmoid': Sigmoid,
'selu': SELU,
'elu': ELU,
'softmax': Softmax,
'leaky_relu': LeakyReLU,
'tanh': TanH,
'softplus': SoftPlus
}
class Activation(Layer):
"""A layer that applies an activation operation to the input.
Parameters:
-----------
name: string
The name of the activation function that will be used.
"""
def __init__(self, name):
self.activation_name = name
self.activation_func = activation_functions[name]()
self.trainable = True
def layer_name(self):
return "Activation (%s)" % (self.activation_func.__class__.__name__)
def forward_pass(self, X, training=True):
self.layer_input = X
return self.activation_func(X)
def backward_pass(self, accum_grad):
return accum_grad * self.activation_func.gradient(self.layer_input)
def output_shape(self):
return self.input_shape
# Method which calculates the padding based on the specified output shape and the
# shape of the filters
def determine_padding(filter_shape, output_shape="same"):
# No padding
if output_shape == "valid":
return (0, 0), (0, 0)
# Pad so that the output shape is the same as input shape (given that stride=1)
elif output_shape == "same":
filter_height, filter_width = filter_shape
# Derived from:
# output_height = (height + pad_h - filter_height) / stride + 1
# In this case output_height = height and stride = 1. This gives the
# expression for the padding below.
pad_h1 = int(math.floor((filter_height - 1)/2))
pad_h2 = int(math.ceil((filter_height - 1)/2))
pad_w1 = int(math.floor((filter_width - 1)/2))
pad_w2 = int(math.ceil((filter_width - 1)/2))
return (pad_h1, pad_h2), (pad_w1, pad_w2)
# Reference: CS231n Stanford
def get_im2col_indices(images_shape, filter_shape, padding, stride=1):
# First figure out what the size of the output should be
batch_size, channels, height, width = images_shape
filter_height, filter_width = filter_shape
pad_h, pad_w = padding
out_height = int((height + np.sum(pad_h) - filter_height) / stride + 1)
out_width = int((width + np.sum(pad_w) - filter_width) / stride + 1)
i0 = np.repeat(np.arange(filter_height), filter_width)
i0 = np.tile(i0, channels)
i1 = stride * np.repeat(np.arange(out_height), out_width)
j0 = np.tile(np.arange(filter_width), filter_height * channels)
j1 = stride * np.tile(np.arange(out_width), out_height)
i = i0.reshape(-1, 1) + i1.reshape(1, -1)
j = j0.reshape(-1, 1) + j1.reshape(1, -1)
k = np.repeat(np.arange(channels), filter_height * filter_width).reshape(-1, 1)
return (k, i, j)
# Method which turns the image shaped input to column shape.
# Used during the forward pass.
# Reference: CS231n Stanford
def image_to_column(images, filter_shape, stride, output_shape='same'):
filter_height, filter_width = filter_shape
pad_h, pad_w = determine_padding(filter_shape, output_shape)
# Add padding to the image
images_padded = np.pad(images, ((0, 0), (0, 0), pad_h, pad_w), mode='constant')
# Calculate the indices where the dot products are to be applied between weights
# and the image
k, i, j = get_im2col_indices(images.shape, filter_shape, (pad_h, pad_w), stride)
# Get content from image at those indices
cols = images_padded[:, k, i, j]
channels = images.shape[1]
# Reshape content into column shape
cols = cols.transpose(1, 2, 0).reshape(filter_height * filter_width * channels, -1)
return cols
# Method which turns the column shaped input to image shape.
# Used during the backward pass.
# Reference: CS231n Stanford
def column_to_image(cols, images_shape, filter_shape, stride, output_shape='same'):
batch_size, channels, height, width = images_shape
pad_h, pad_w = determine_padding(filter_shape, output_shape)
height_padded = height + np.sum(pad_h)
width_padded = width + np.sum(pad_w)
images_padded = np.zeros((batch_size, channels, height_padded, width_padded))
# Calculate the indices where the dot products are applied between weights
# and the image
k, i, j = get_im2col_indices(images_shape, filter_shape, (pad_h, pad_w), stride)
cols = cols.reshape(channels * np.prod(filter_shape), -1, batch_size)
cols = cols.transpose(2, 0, 1)
# Add column content to the images at the indices
np.add.at(images_padded, (slice(None), k, i, j), cols)
# Return image without padding
return images_padded[:, :, pad_h[0]:height+pad_h[0], pad_w[0]:width+pad_w[0]]