-
Notifications
You must be signed in to change notification settings - Fork 3
/
BASNN.py
288 lines (246 loc) · 11.8 KB
/
BASNN.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
# Copyright 2017 Shihui Yin Arizona State University
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED
# WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE,
# MERCHANTABLITY OR NON-INFRINGEMENT.
# See the Apache 2 License for the specific language governing permissions and
# limitations under the License.
# Description: BASNN.py includes a few modules for Binary-Activation SNN
import lasagne
import theano
import theano.tensor as T
from theano.sandbox.rng_mrg import MRG_RandomStreams as RandomStreams
import numpy as np
import time
class HardFire(theano.Op):
__props__ = ()
def make_node(self, x):
x = T.as_tensor_variable(x)
return theano.Apply(self, [x], [x.type()])
def perform(self, node, inputs, output_storage):
x = inputs[0]
z = output_storage[0]
z[0] = np.greater(x, 1.).astype(theano.config.floatX)
def grad(self, inputs, output_grads):
return [0.5*output_grads[0] * (T.and_(T.gt(inputs[0], 0.), T.lt(inputs[0], 2.)))]
hard_fire = HardFire()
def hard_fire_testonly(x):
return T.cast(T.gt(x, 1.,), theano.config.floatX)
class InputLayer(lasagne.layers.InputLayer):
def __init__(self, shape, input_var=None, name=None, binary=True, deterministic=False, threshold=0.5, batch_size=100, n_bits=-1, **kwargs):
self.rng_mrg = RandomStreams(lasagne.random.get_rng().randint(1, 2394349593))
if binary == False:
if n_bits == -1: # no quantization at all
super(InputLayer, self).__init__(shape=shape, input_var=input_var, name=name, **kwargs)
else:
# Normalize to [0 ~ 1 - 2^(-n_bits)]
input_var_normed = input_var * (1 - 2**(-n_bits))
if deterministic == False:
shape_rand = list(shape)
if shape_rand[0] is None:
shape_rand[0] = batch_size
shape_rand = tuple(shape_rand)
input_var_ceil = T.ceil(input_var_normed * 2**n_bits) / 2**n_bits
input_var_floor = T.floor(input_var_normed * 2**n_bits) / 2**n_bits
input_var_above_floor = input_var - input_var_floor
input_var_stochastic_quantized = T.cast(T.switch(T.ge(input_var_above_floor, self.rng_mrg.uniform(shape_rand, low=0.0, high=2**(-n_bits), dtype=theano.config.floatX)), input_var_ceil, input_var_floor), theano.config.floatX)
super(InputLayer, self).__init__(shape=shape, input_var=input_var_stochastic_quantized, name=name, **kwargs)
else:
input_var_deterministic_quantized = T.cast(T.round(input_var_normed * 2**n_bits) / 2**n_bits, theano.config.floatX)
super(InputLayer, self).__init__(shape=shape, input_var=input_var_deterministic_quantized, name=name, **kwargs)
else:
if deterministic == False:
shape_rand = list(shape)
if shape_rand[0] is None:
shape_rand[0] = batch_size
shape_rand = tuple(shape_rand)
# Bernoulli spikes
input_var_stochastic_binarized = T.cast(T.gt(input_var, self.rng_mrg.uniform(shape_rand, low=0.0, high=1.0, dtype=theano.config.floatX)), theano.config.floatX)
super(InputLayer, self).__init__(shape=shape, input_var=input_var_stochastic_binarized, name=name, **kwargs)
else:
input_var_deterministic_binarized = T.cast(T.switch(T.ge(input_var, threshold), 1.0, 0.), theano.config.floatX)
super(InputLayer, self).__init__(shape=shape, input_var=input_var_deterministic_binarized, name=name, **kwargs)
class NonlinearityLayer(lasagne.layers.Layer):
def __init__(self, incoming, nonlinearity=hard_fire, num_time_steps=2, reset=True, **kwargs):
super(NonlinearityLayer, self).__init__(incoming, **kwargs)
self.nonlinearity = (nonlinearities.identity if nonlinearity is None else nonlinearity)
self.num_time_steps = num_time_steps
self.reset = reset
def get_output_for(self, input, **kwargs):
output_list = []
output_list.append(self.nonlinearity(input[0]))
for i in range(1,self.num_time_steps):
temp = 0
temp += input[i]
for j in range(i):
if self.reset == True:
temp += input[j] - output_list[j]
else:
temp += input[j]
output_list.append(self.nonlinearity(temp))
output = T.stack(output_list)
return output
def train_sgd(train_fn,val_fn,
model,
batch_size,
LR_start,LR_decay,
num_epochs,
X_train,y_train,
X_val,y_val,
X_test,y_test,
save_path=None,
shuffle_parts=1,
batch_size_val=None,
batch_size_axis=0,
repeat_in_batch=1,
monitor_valid_err=True,
k_start=0,
k_decay=1,
save_last_epoch=False,
k_decay_mode="exponential"
):
if batch_size_val is None:
batch_size_val = batch_size
# A function which shuffles a dataset
def shuffle(X,y):
chunk_size = int(len(X)/shuffle_parts)
shuffled_range = list(range(chunk_size))
X_buffer = np.copy(X[0:chunk_size])
y_buffer = np.copy(y[0:chunk_size])
for k in range(shuffle_parts):
np.random.shuffle(shuffled_range)
for i in range(chunk_size):
X_buffer[i] = X[k*chunk_size+shuffled_range[i]]
y_buffer[i] = y[k*chunk_size+shuffled_range[i]]
X[k*chunk_size:(k+1)*chunk_size] = X_buffer
y[k*chunk_size:(k+1)*chunk_size] = y_buffer
return X,y
# This function trains the model a full epoch (on the whole dataset)
def train_epoch(X,y,LR,k=0):
loss = 0
if batch_size_axis == 0:
batches = int(len(X)/batch_size)
for i in range(batches):
sample_indices = list(range(i*batch_size,(i+1)*batch_size))*repeat_in_batch
if k > 0:
loss += train_fn(X[sample_indices],y[sample_indices],LR, k)
else:
loss += train_fn(X[sample_indices],y[sample_indices],LR)
elif batch_size_axis == 1:
batches = int(X.shape[1]/batch_size)
for i in range(batches):
sample_indices = list(range(i*batch_size,(i+1)*batch_size))*repeat_in_batch
if k > 0:
loss += train_fn(X[:,sample_indices],y[:,sample_indices],LR, k)
else:
loss += train_fn(X[:,sample_indices],y[:,sample_indices],LR)
else:
print("Batch size axis = %d is not supported" % batch_size_axis)
loss/=batches
return loss
# This function tests the model a full epoch (on the whole dataset)
def val_epoch(X,y,k=0):
err = 0
loss = 0
if batch_size_axis == 0:
batches = int(len(X)/batch_size_val)
for i in range(batches):
sample_indices = list(range(i*batch_size,(i+1)*batch_size))*repeat_in_batch
if k > 0:
new_loss, new_err = val_fn(X[sample_indices], y[sample_indices], k)
else:
new_loss, new_err = val_fn(X[sample_indices], y[sample_indices])
err += new_err
loss += new_loss
elif batch_size_axis == 1:
batches = int(X.shape[1]/batch_size)
for i in range(batches):
sample_indices = list(range(i*batch_size,(i+1)*batch_size))*repeat_in_batch
if k > 0:
new_loss, new_err = val_fn(X[:,sample_indices], y[:,sample_indices], k)
else:
new_loss, new_err = val_fn(X[:,sample_indices], y[:,sample_indices])
err += new_err
loss += new_loss
else:
print("Batch size axis = %d is not supported" % batch_size_axis)
err = err / batches * 100
loss /= batches
return err, loss
# shuffle the train set
X_train,y_train = shuffle(X_train,y_train)
best_val_err = 100
best_val_loss = 100000
best_epoch = 1
LR = LR_start
k = k_start
k_finish = k_start * k_decay ** (num_epochs-1)
k_increm = (k_finish - k_start) / (num_epochs-1)
if k_start != 0:
print("k_start = %f" % k_start)
for epoch in range(num_epochs):
start_time = time.time()
if k > 0:
train_loss = train_epoch(X_train,y_train,LR,k)
else:
train_loss = train_epoch(X_train,y_train,LR)
X_train,y_train = shuffle(X_train,y_train)
print('finish one epoch training')
print(time.time() - start_time)
if k > 0:
val_err, val_loss = val_epoch(X_val,y_val,k=1)
else:
val_err, val_loss = val_epoch(X_val,y_val)
print('finish one epoch validation')
print(time.time() - start_time)
if save_last_epoch == False:
if monitor_valid_err == True:
if val_err < best_val_err:
best_val_err = val_err
best_epoch = epoch+1
if k > 0:
test_err, test_loss = val_epoch(X_test,y_test,k)
else:
test_err, test_loss = val_epoch(X_test,y_test)
if save_path is not None:
np.savez(save_path, *lasagne.layers.get_all_param_values(model))
else: # monitor validation loss
if val_loss < best_val_loss:
best_val_loss = val_loss
best_epoch = epoch+1
test_err, test_loss = val_epoch(X_test,y_test,k)
if save_path is not None:
np.savez(save_path, *lasagne.layers.get_all_param_values(model))
else:
test_err, test_loss = val_epoch(X_test,y_test,k)
np.savez(save_path, *lasagne.layers.get_all_param_values(model))
epoch_duration = time.time() - start_time
# Then we print the results for this epoch:
print("Epoch "+str(epoch + 1)+" of "+str(num_epochs)+" took "+str(epoch_duration)+"s")
print(" LR: "+str(LR))
if k > 0:
print(" k: "+str(k))
print(" training loss: "+str(train_loss))
print(" validation loss: "+str(val_loss))
print(" validation error rate: "+str(val_err)+"%")
print(" best epoch: "+str(best_epoch))
if monitor_valid_err:
print(" best validation loss: "+str(val_loss))
print(" best validation error rate: "+str(best_val_err)+"%")
else:
print(" best validation loss: "+str(best_val_loss))
print(" best validation error rate: "+str(val_err)+"%")
print(" test loss: "+str(test_loss))
print(" test error rate: "+str(test_err)+"%")
# decay the LR
LR *= LR_decay
if k_decay_mode == "exponential":
k *= k_decay
else:
k += k_increm