-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathexample.py
281 lines (236 loc) · 10.3 KB
/
example.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
import mylif
import stdp
import spikeGen
from struct import unpack
import numpy as np
import matplotlib.pyplot as plt
def get_labeled_data(sampleFilename, labelFilename):
samples = open(sampleFilename, 'rb')
labels = open(labelFilename, 'rb')
# Get metadata for images
samples.read(4) # skip the magic_number
number_of_samples = unpack('>I', samples.read(4))[0]
rows = unpack('>I', samples.read(4))[0]
cols = unpack('>I', samples.read(4))[0]
# Get metadata for labels
labels.read(4) # skip the magic_number
N = unpack('>I', labels.read(4))[0]
if number_of_samples != N:
raise Exception('number of labels did not match the number of samples')
# Get the data
x = np.zeros((N, rows, cols))#, dtype=np.uint8) # Initialize numpy array
y = np.zeros((N, 1))#, dtype=np.uint8) # Initialize numpy array
for i in range(N):
# if i % 1000 == 0:
# print("i: %i" % i)
x[i] = [[unpack('>B', samples.read(1))[0] for unused_col in range(cols)] for unused_row in range(rows) ]
y[i] = unpack('>B', labels.read(1))[0]
#np.savetxt("data/trainData.txt", x)
#np.savetxt("data/trainLabel.txt", y)
data = {'x': x, 'y': y, 'rows': rows, 'cols': cols}
return data
def save_weights(filename, w):
print('save trained weights')
np.savetxt(filename, w)
#---------------------
# structure of network
#---------------------
numOfSpkieGenerate = 28*28
numOfHidden1 = 100
numOfHidden2 = 50
numOfHidden3 = 100
numOfOutput = 10
#-------------------
# parameters
#-------------------
#-------------------------
# random initialize weight with uniform distribution
#-------------------------
wi1 = np.random.rand(numOfSpkieGenerate, numOfHidden1)
w12 = np.random.rand(numOfHidden1, numOfHidden2)
w23 = np.random.rand(numOfHidden2, numOfHidden3)
w3o = np.random.rand(numOfHidden3, numOfOutput)
#---------------------------
# create networks
#---------------------------
myNeuralParam = mylif.neuralParam(2, 0, 0.3, -68, 1, 0.03, -50, -70)
# layer_in = mylif.layerModel(0, -68, numOfSpkieGenerate)
layer_h1 = mylif.layerModel(0, -68, numOfHidden1)
layer_h2 = mylif.layerModel(0, -68, numOfHidden2)
layer_h3 = mylif.layerModel(0, -68, numOfHidden3)
layer_out = mylif.layerModel(0, -68, numOfOutput)
myNeuralParam = mylif.neuralParam(2, 0, 0.3, -68, 1, 0.1, -50, -70)
myStdpParam = stdp.stdpParam(0.01, 100, 0.01, 1000, 15, 1)
#--------------------------
# stimulation
#--------------------------
t_total = 1
dt = 0.01
stepNum = int(t_total/dt)
mode = 'test'
if mode == 'train':
#------------------
# load MNIST
#------------------
trainDataName = 'data/train-images.idx3-ubyte'
trainLabelName = 'data/train-labels.idx1-ubyte'
trainData = get_labeled_data(trainDataName, trainLabelName)
sampleNum = trainData['y'].size
featureLength = trainData['x'].shape[1] * trainData['x'].shape[2]
#-------------------------
# extract subset
#-------------------------
subset = np.loadtxt('data/subset.txt', dtype=np.int, delimiter="\n")
for i in subset:
singleData = (trainData['x'][i-1, :, :].reshape((featureLength))).copy()
singleData[singleData < 1] = 4./(dt*stepNum)
spikeCycle = 4. / singleData.reshape((featureLength)) / dt
spikeSeq = np.zeros((featureLength, stepNum))
# generate input spike
for j in range(featureLength):
spikeSeq[j, :] = spikeGen.possionSpike(int(spikeCycle[j]), stepNum)
# training
mask_i1 = np.zeros((numOfSpkieGenerate, numOfHidden1))
pair_i1 = [[(x,y) for y in layer_h1.spikeTime] for x in np.ones(numOfSpkieGenerate) * (-1) ]
mask_12 = np.zeros((numOfHidden1, numOfHidden2))
pair_12 = [[(x,y) for y in layer_h2.spikeTime] for x in layer_h1.spikeTime ]
mask_23 = np.zeros((numOfHidden2, numOfHidden3))
pair_23 = [[(x,y) for y in layer_h3.spikeTime] for x in layer_h2.spikeTime ]
mask_3o = np.zeros((numOfHidden3, numOfOutput))
pair_3o = [[(x,y) for y in layer_out.spikeTime] for x in layer_h3.spikeTime ]
for k in range(stepNum):
# input layer and hidden layer 1
spike_in_time = np.ones(featureLength) * (-1)
spike_in_time[spikeSeq[:, k] > 0] = k
spike_h1 = layer_h1.update(myNeuralParam, wi1, spikeSeq[:, k], dt, k)
dwi1 = stdp.stdp(wi1, myStdpParam, spike_in_time, layer_h1.spikeTime, 'bi')
wi1 += dwi1
# record the spiking time pair between two layers
for m in range(numOfSpkieGenerate):
for n in range(numOfHidden1):
if dwi1[m,n] != 0:
if pair_i1[m][n][0] != spike_in_time[m] and pair_i1[m][n][1] != layer_h1.spikeTime[n]:
pair_i1[m][n] = (spike_in_time[m], layer_h1.spikeTime[n])
mask_i1[m][n] = 0
dwi1[mask_i1!=0] = 0
wi1 += dwi1
# update mask
mask_i1[dwi1!=0] = 1
# hidden layer 1 and hidden layer 2
spike_h2 = layer_h2.update(myNeuralParam, w12, spike_h1, dt, k)
dw12 = stdp.stdp(w12, myStdpParam, layer_h1.spikeTime, layer_h2.spikeTime, 'bi')
w12 += dw12
# record the spiking time pair between two layers
for m in range(numOfHidden1):
for n in range(numOfHidden2):
if dw12[m,n] != 0:
if pair_12[m][n][0] != layer_h1.spikeTime[m] and pair_12[m][n][1] != layer_h2.spikeTime[n]:
pair_12[m][n] = (layer_h1.spikeTime[m], layer_h2.spikeTime[n])
mask_12[m][n] = 0
dw12[mask_12!=0] = 0
w12 += dw12
# update mask
mask_12[dw12!=0] = 1
# hidden layer 2 and hidden layer 3
spike_h3 = layer_h3.update(myNeuralParam, w23, spike_h2, dt, k)
dw23 = stdp.stdp(w23, myStdpParam, layer_h2.spikeTime, layer_h3.spikeTime, 'bi')
w23 += dw23
# record the spiking time pair between two layers
for m in range(numOfHidden2):
for n in range(numOfHidden3):
if dw23[m,n] != 0:
if pair_23[m][n][0] != layer_h2.spikeTime[m] and pair_23[m][n][1] != layer_h3.spikeTime[n]:
pair_23[m][n] = (layer_h2.spikeTime[m], layer_h3.spikeTime[n])
mask_23[m][n] = 0
dw23[mask_23!=0] = 0
w23 += dw23
# update mask
mask_23[dw23!=0] = 1
# hidden layer 3 and output layer
spike_out = layer_out.update(myNeuralParam, w3o, spike_h3, dt, k)
dw3o = stdp.stdp(w3o, myStdpParam, layer_h3.spikeTime, layer_out.spikeTime, 'bi')
w3o += dw3o
# record the spiking time pair between two layers
for m in range(numOfHidden3):
for n in range(numOfOutput):
if dw3o[m,n] != 0:
if pair_3o[m][n][0] != layer_h3.spikeTime[m] and pair_3o[m][n][1] != layer_out.spikeTime[n]:
pair_3o[m][n] = (layer_h3.spikeTime[m], layer_out.spikeTime[n])
mask_3o[m][n] = 0
dw3o[mask_3o!=0] = 0
w3o += dw3o
# update mask
mask_3o[dw3o!=0] = 1
# reset
layer_h1.reset(0, -68)
layer_h2.reset(0, -68)
layer_h3.reset(0, -68)
layer_out.reset(0, -68)
# save trained weights
save_weights('wi1.txt', wi1)
save_weights('w12.txt', w12)
save_weights('w23.txt', w23)
save_weights('w3o.txt', w3o)
else:
# load MNIST test dataset
testDataName = 'data/t10k-images.idx3-ubyte'
testLabelName = 'data/t10k-labels.idx1-ubyte'
testData = get_labeled_data(testDataName, testLabelName)
featureLength = testData['x'].shape[1] * testData['x'].shape[2]
# load weights from txt
wi1 = np.loadtxt('weights/wi1.txt')/5.
w12 = np.loadtxt('weights/w12.txt')/5.
w23 = np.loadtxt('weights/w23.txt')/5.
w3o = np.loadtxt('weights/w3o.txt')/5.
# extract subset
subset = np.loadtxt('weights/subset.txt', dtype=np.int, delimiter="\n")
count = 0
# start test
for i in subset:
count += 1
singleData = (testData['x'][i-1, :, :].reshape((featureLength))).copy()
singleData[singleData < 1] = 4./(dt*stepNum)
spikeCycle = 4. / singleData.reshape((featureLength)) / dt
spikeSeq = np.zeros((featureLength, stepNum))
# generate input spike
for j in range(featureLength):
spikeSeq[j, :] = spikeGen.possionSpike(int(spikeCycle[j]), stepNum)
outSeq = np.zeros((stepNum, numOfOutput))
for k in range(stepNum):
# input layer and hidden layer 1
spike_in_time = np.ones(featureLength) * (-1)
spike_in_time[spikeSeq[:, k] > 0] = k
spike_h1 = layer_h1.update(myNeuralParam, wi1, spikeSeq[:, k], dt, k)
# hidden layer 1 and hidden layer 2
spike_h2 = layer_h2.update(myNeuralParam, w12, spike_h1, dt, k)
# hidden layer 2 and hidden layer 3
spike_h3 = layer_h3.update(myNeuralParam, w23, spike_h2, dt, k)
# hidden layer 3 and output layer
spike_out = layer_out.update(myNeuralParam, w3o, spike_h3, dt, k)
outSeq[k, :] = spike_out.copy()
# draw results
#plt.subplots(nrows=2, ncols=5)
#for j in range(10):
# plt.subplot(2, 5, j+1)
# plt.title('the '+str(count)+ ' iteration ' + str(sum(outSeq[:,j])))
#plt.show()
#plt.waitforbuttonpress()
#plt.clf()
# select the neuron that spikes the most
maxSpike = 0
maxInd = 1
for j in range(10):
if sum(outSeq[:,j]) > maxSpike:
maxSpike = sum(outSeq[:,j])
maxInd = j
print('spikes:%d, class:%d, label:%d\n' % (maxSpike, maxInd+1, count))
layer_h1.reset(0, -68)
layer_h2.reset(0, -68)
layer_h3.reset(0, -68)
layer_out.reset(0, -68)
#plt.title('the '+str(count)+ ' iteration, ' + 'spikes: ' + str(maxSpike) + ', label: ' + str(maxInd))
#plt.plot(outSeq[:,maxInd], 'b')
#
#plt.show()
#plt.waitforbuttonpress()
#plt.clf()