-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
400 lines (344 loc) · 17.5 KB
/
main.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
import numpy as np
import os
import pandas as pd
from keras.preprocessing.image import ImageDataGenerator
from keras import callbacks
from keras.utils.vis_utils import plot_model
import matplotlib.pyplot as plt #for plotting
from sklearn.metrics import confusion_matrix
import itertools
import seaborn as sns
from utils import plot_log
'''
Code Credits
capsule net for keras :
https://github.com/XifengGuo/CapsNet-Keras/blob/master/capsulenet.py
'''
import keras.backend as K
import tensorflow as tf
from keras import initializers, layers
class Length(layers.Layer):
"""
Compute the length of vectors. This is used to compute a Tensor that has the same shape with y_true in margin_loss
inputs: shape=[dim_1, ..., dim_{n-1}, dim_n]
output: shape=[dim_1, ..., dim_{n-1}]
"""
def call(self, inputs, **kwargs):
return K.sqrt(K.sum(K.square(inputs), -1))
def compute_output_shape(self, input_shape):
return input_shape[:-1]
class Mask(layers.Layer):
"""
Mask a Tensor with shape=[None, d1, d2] by the max value in axis=1.
Output shape: [None, d2]
"""
def call(self, inputs, **kwargs):
# use true label to select target capsule, shape=[batch_size, num_capsule]
if type(inputs) is list: # true label is provided with shape = [batch_size, n_classes], i.e. one-hot code.
assert len(inputs) == 2
inputs, mask = inputs
else: # if no true label, mask by the max length of vectors of capsules
x = inputs
# Enlarge the range of values in x to make max(new_x)=1 and others < 0
x = (x - K.max(x, 1, True)) / K.epsilon() + 1
mask = K.clip(x, 0, 1) # the max value in x clipped to 1 and other to 0
# masked inputs, shape = [batch_size, dim_vector]
inputs_masked = K.batch_dot(inputs, mask, [1, 1])
return inputs_masked
def compute_output_shape(self, input_shape):
if type(input_shape[0]) is tuple: # true label provided
return tuple([None, input_shape[0][-1]])
else:
return tuple([None, input_shape[-1]])
def squash(vectors, axis=-1):
"""
The non-linear activation used in Capsule. It drives the length of a large vector to near 1 and small vector to 0
:param vectors: some vectors to be squashed, N-dim tensor
:param axis: the axis to squash
:return: a Tensor with same shape as input vectors
"""
s_squared_norm = K.sum(K.square(vectors), axis, keepdims=True)
scale = s_squared_norm / (1 + s_squared_norm) / K.sqrt(s_squared_norm)
return scale * vectors
class CapsuleLayer(layers.Layer):
"""
The capsule layer. It is similar to Dense layer. Dense layer has `in_num` inputs, each is a scalar, the output of the
neuron from the former layer, and it has `out_num` output neurons. CapsuleLayer just expand the output of the neuron
from scalar to vector. So its input shape = [None, input_num_capsule, input_dim_vector] and output shape = \
[None, num_capsule, dim_vector]. For Dense Layer, input_dim_vector = dim_vector = 1.
:param num_capsule: number of capsules in this layer
:param dim_vector: dimension of the output vectors of the capsules in this layer
:param num_routings: number of iterations for the routing algorithm
"""
def __init__(self, num_capsule, dim_vector, num_routing=3,
kernel_initializer='glorot_uniform',
bias_initializer='zeros',
**kwargs):
super(CapsuleLayer, self).__init__(**kwargs)
self.num_capsule = num_capsule
self.dim_vector = dim_vector
self.num_routing = num_routing
self.kernel_initializer = initializers.get(kernel_initializer)
self.bias_initializer = initializers.get(bias_initializer)
def build(self, input_shape):
assert len(input_shape) >= 3, "The input Tensor should have shape=[None, input_num_capsule, input_dim_vector]"
self.input_num_capsule = input_shape[1]
self.input_dim_vector = input_shape[2]
# Transform matrix
self.W = self.add_weight(
shape=[self.input_num_capsule, self.num_capsule, self.input_dim_vector, self.dim_vector],
initializer=self.kernel_initializer,
name='W')
# Coupling coefficient. The redundant dimensions are just to facilitate subsequent matrix calculation.
self.bias = self.add_weight(shape=[1, self.input_num_capsule, self.num_capsule, 1, 1],
initializer=self.bias_initializer,
name='bias',
trainable=False)
self.built = True
def call(self, inputs, training=None):
# inputs.shape=[None, input_num_capsule, input_dim_vector]
# Expand dims to [None, input_num_capsule, 1, 1, input_dim_vector]
inputs_expand = K.expand_dims(K.expand_dims(inputs, 2), 2)
# Replicate num_capsule dimension to prepare being multiplied by W
# Now it has shape = [None, input_num_capsule, num_capsule, 1, input_dim_vector]
inputs_tiled = K.tile(inputs_expand, [1, 1, self.num_capsule, 1, 1])
"""
# Compute `inputs * W` by expanding the first dim of W. More time-consuming and need batch_size.
# Now W has shape = [batch_size, input_num_capsule, num_capsule, input_dim_vector, dim_vector]
w_tiled = K.tile(K.expand_dims(self.W, 0), [self.batch_size, 1, 1, 1, 1])
# Transformed vectors, inputs_hat.shape = [None, input_num_capsule, num_capsule, 1, dim_vector]
inputs_hat = K.batch_dot(inputs_tiled, w_tiled, [4, 3])
"""
# Compute `inputs * W` by scanning inputs_tiled on dimension 0. This is faster but requires Tensorflow.
# inputs_hat.shape = [None, input_num_capsule, num_capsule, 1, dim_vector]
inputs_hat = tf.scan(lambda ac, x: K.batch_dot(x, self.W, [3, 2]),
elems=inputs_tiled,
initializer=K.zeros([self.input_num_capsule, self.num_capsule, 1, self.dim_vector]))
"""
# Routing algorithm V1. Use tf.while_loop in a dynamic way.
def body(i, b, outputs):
c = tf.nn.softmax(self.bias, dim=2) # dim=2 is the num_capsule dimension
outputs = squash(K.sum(c * inputs_hat, 1, keepdims=True))
b = b + K.sum(inputs_hat * outputs, -1, keepdims=True)
return [i-1, b, outputs]
cond = lambda i, b, inputs_hat: i > 0
loop_vars = [K.constant(self.num_routing), self.bias, K.sum(inputs_hat, 1, keepdims=True)]
_, _, outputs = tf.while_loop(cond, body, loop_vars)
"""
# Routing algorithm V2. Use iteration. V2 and V1 both work without much difference on performance
assert self.num_routing > 0, 'The num_routing should be > 0.'
for i in range(self.num_routing):
c = tf.nn.softmax(self.bias, dim=2) # dim=2 is the num_capsule dimension
# outputs.shape=[None, 1, num_capsule, 1, dim_vector]
outputs = squash(K.sum(c * inputs_hat, 1, keepdims=True))
# last iteration needs not compute bias which will not be passed to the graph any more anyway.
if i != self.num_routing - 1:
# self.bias = K.update_add(self.bias, K.sum(inputs_hat * outputs, [0, -1], keepdims=True))
self.bias += K.sum(inputs_hat * outputs, -1, keepdims=True)
# tf.summary.histogram('BigBee', self.bias) # for debugging
return K.reshape(outputs, [-1, self.num_capsule, self.dim_vector])
def compute_output_shape(self, input_shape):
return tuple([None, self.num_capsule, self.dim_vector])
def PrimaryCap(inputs, dim_vector, n_channels, kernel_size, strides, padding):
"""
Apply Conv2D `n_channels` times and concatenate all capsules
:param inputs: 4D tensor, shape=[None, width, height, channels]
:param dim_vector: the dim of the output vector of capsule
:param n_channels: the number of types of capsules
:return: output tensor, shape=[None, num_capsule, dim_vector]
"""
output = layers.Conv2D(filters=dim_vector * n_channels, kernel_size=kernel_size, strides=strides, padding=padding)(
inputs)
outputs = layers.Reshape(target_shape=[-1, dim_vector])(output)
return layers.Lambda(squash)(outputs)
from keras import layers, models
from keras import backend as K
from keras.utils import to_categorical
def CapsNet(input_shape, n_class, num_routing):
"""
A Capsule Network on MNIST.
:param input_shape: data shape, 4d, [None, width, height, channels]
:param n_class: number of classes
:param num_routing: number of routing iterations
:return: A Keras Model with 2 inputs and 2 outputs
"""
x = layers.Input(shape=input_shape)
# Layer 1: Just a conventional Conv2D layer
conv1 = layers.Conv2D(filters=128, kernel_size=9, strides=1, padding='valid', activation='relu', name='conv1')(x)
# Layer 2: Conv2D layer with `squash` activation, then reshape to [None, num_capsule, dim_vector]
primarycaps = PrimaryCap(conv1, dim_vector=8, n_channels=16, kernel_size=9, strides=2, padding='valid')
# Layer 3: Capsule layer. Routing algorithm works here.
digitcaps = CapsuleLayer(num_capsule=n_class, dim_vector=16, num_routing=num_routing, name='digitcaps')(primarycaps)
# Layer 4: This is an auxiliary layer to replace each capsule with its length. Just to match the true label's shape.
# If using tensorflow, this will not be necessary. :)
out_caps = Length(name='out_caps')(digitcaps)
# Decoder network.
y = layers.Input(shape=(n_class,))
masked = Mask()([digitcaps, y]) # The true label is used to mask the output of capsule layer.
x_recon = layers.Dense(512, activation='relu')(masked)
x_recon = layers.Dense(1024, activation='relu')(x_recon)
x_recon = layers.Dense(784, activation='sigmoid')(x_recon)
x_recon = layers.Reshape(target_shape=[28, 28, 1], name='out_recon')(x_recon)
# two-input-two-output keras Model
return models.Model([x, y], [out_caps, x_recon])
def margin_loss(y_true, y_pred):
"""
Margin loss for Eq.(4). When y_true[i, :] contains not just one `1`, this loss should work too. Not test it.
:param y_true: [None, n_classes]
:param y_pred: [None, num_capsule]
:return: a scalar loss value.
"""
L = y_true * K.square(K.maximum(0., 0.9 - y_pred)) + \
0.5 * (1 - y_true) * K.square(K.maximum(0., y_pred - 0.1))
return K.mean(K.sum(L, 1))
def train(model, data, epoch_size_frac=1.0,epochs=1):
"""
Training a CapsuleNet
:param model: the CapsuleNet model
:param data: a tuple containing training and testing data, like `((x_train, y_train), (x_test, y_test))`
:param args: arguments
:return: The trained model
"""
# unpacking the data
(x_train, y_train), (x_test, y_test) = data
# callbacks
log = callbacks.CSVLogger('log.csv')
checkpoint = callbacks.ModelCheckpoint('weights-{epoch:02d}.h5',
save_best_only=True, save_weights_only=True, verbose=1)
lr_decay = callbacks.LearningRateScheduler(schedule=lambda epoch: 0.001 * np.exp(-epoch / 10.))
# compile the model
model.compile(optimizer='adam',
loss=[margin_loss, 'mse'],
loss_weights=[1., 0.0005],
metrics={'out_caps': 'accuracy'})
"""
# Training without data augmentation:
model.fit([x_train, y_train], [y_train, x_train], batch_size=args.batch_size, epochs=args.epochs,
validation_data=[[x_test, y_test], [y_test, x_test]], callbacks=[log, tb, checkpoint])
"""
# -----------------------------------Begin: Training with data augmentation -----------------------------------#
def train_generator(x, y, batch_size, shift_fraction=0.):
train_datagen = ImageDataGenerator(width_shift_range=shift_fraction,
height_shift_range=shift_fraction) # shift up to 2 pixel for MNIST
generator = train_datagen.flow(x, y, batch_size=batch_size)
while 1:
x_batch, y_batch = generator.next()
yield ([x_batch, y_batch], [y_batch, x_batch])
# Training with data augmentation. If shift_fraction=0., also no augmentation.
model.fit_generator(generator=train_generator(x_train, y_train, 64, 0.1),
steps_per_epoch=int(epoch_size_frac*y_train.shape[0] / 64),
epochs=epochs,
validation_data=[[x_test, y_test], [y_test, x_test]],
callbacks=[log, checkpoint, lr_decay])
# -----------------------------------End: Training with data augmentation -----------------------------------#
model.save_weights('trained_model.h5')
print('Trained model saved to \'trained_model.h5\'')
return model
def combine_images(generated_images):
num = generated_images.shape[0]
width = int(np.sqrt(num))
height = int(np.ceil(float(num)/width))
shape = generated_images.shape[1:3]
image = np.zeros((height*shape[0], width*shape[1]),
dtype=generated_images.dtype)
for index, img in enumerate(generated_images):
i = int(index/width)
j = index % width
image[i*shape[0]:(i+1)*shape[0], j*shape[1]:(j+1)*shape[1]] = \
img[:, :, 0]
return image
def test(model, data):
x_test, y_test = data
y_pred, x_recon = model.predict([x_test, y_test], batch_size=100)
return y_pred
def plot_confusion_matrix(cm, classes,
normalize=False,
title='Confusion matrix',
cmap=plt.cm.Blues):
"""
This function prints and plots the confusion matrix.
Normalization can be applied by setting `normalize=True`.
"""
plt.imshow(cm, interpolation='nearest', cmap=cmap)
plt.title(title)
plt.colorbar()
tick_marks = np.arange(len(classes))
plt.xticks(tick_marks, classes, rotation=45)
plt.yticks(tick_marks, classes)
if normalize:
cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]
thresh = cm.max() / 2.
for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):
plt.text(j, i, cm[i, j],
horizontalalignment="center",
color="white" if cm[i, j] > thresh else "black")
plt.tight_layout()
plt.ylabel('True label')
plt.xlabel('Predicted label')
def Main():
# define model
model = CapsNet(input_shape=[28, 28, 1],
n_class=10,
num_routing=3)
model.summary()
load_weights=True
try:
plot_model(model, to_file='model.png', show_shapes=True)
except Exception as e:
print('No fancy plot {}'.format(e))
from sklearn.model_selection import train_test_split
data_train = pd.read_csv('train.csv')
#Data counter
#sns.countplot(data_train['label'])
X_full = data_train.iloc[:, 1:]
y_full = data_train.iloc[:, :1]
x_train, x_test, y_train, y_test = train_test_split(X_full, y_full, test_size=0.3)
x_train = x_train.values.reshape(-1, 28, 28, 1).astype('float32') / 255.
x_test = x_test.values.reshape(-1, 28, 28, 1).astype('float32') / 255.
y_train = to_categorical(y_train.astype('float32'))
y_test = to_categorical(y_test.astype('float32'))
if load_weights==False:
train(model=model, data=((x_train, y_train), (x_test, y_test)),epochs=10,
epoch_size_frac=0.5)
else:
model.load_weights('trained_model.h5')
y_pred= test(model=model, data=(x_test, y_test))
print('Test acc:', np.sum(np.argmax(y_pred, 1) == np.argmax(y_test, 1)) / y_test.shape[0])
#Visualization for errors:
# Look at confusion matrix
# Note, this code is taken straight from the SKLEARN website, an nice way of viewing confusion matrix.
# Convert predictions classes to one hot vectors
Y_pred_classes = np.argmax(y_pred, axis=1)
# Convert validation observations to one hot vectors
# compute the confusion matrix
confusion_mtx = confusion_matrix(np.argmax(y_test, axis=1), Y_pred_classes)
# plot the confusion matrix
plot_confusion_matrix(confusion_mtx, classes=range(10))
#Train results
y_pred_train= test(model=model, data=(x_train, y_train))
#Visualization for errors:
# Look at confusion matrix
# Note, this code is taken straight from the SKLEARN website, an nice way of viewing confusion matrix.
# Convert predictions classes to one hot vectors
Y_pred_classes = np.argmax(y_pred_train, axis=1)
# Convert validation observations to one hot vectors
# compute the confusion matrix
y_train_cl=np.argmax(y_train, axis=1)
confusion_mtx = confusion_matrix(y_train_cl, Y_pred_classes)
# plot the confusion matrix
plot_confusion_matrix(confusion_mtx, classes=range(10))
print('Train acc:', np.sum(Y_pred_classes == y_train_cl) / y_train_cl.shape[0])
'''predict the test.csv
data_test = pd.read_csv('../input/test.csv')
data_test = data_test.values.reshape(-1, 28, 28, 1).astype('float32') / 255.
y_pred_sub, _ = model.predict([data_test,
np.zeros((data_test.shape[0],10))], # empty values for the second vector
batch_size = 32, verbose = True)
with open('submission.csv', 'w') as out_file:
out_file.write('ImageId,Label\n')
for img_id, guess_label in enumerate(np.argmax(y_pred, 1), 1):
out_file.write('%d,%d\n' % (img_id, guess_label))
pass
'''
if __name__=='__main__':
Main()