def pre_train(self, train_round, size, rate, dr, scale):
#for each pair of neighbor layers, training RBM
for k in range(self.depth - 1):
rbm = RBM(self.layers[k], self.layers[k+1], self.weightLayers[k], rate)
# mini-batch training times
for tr in range(train_round):
#mini-batch size
for i in range(size):
index = random.randint(0, scale)
im = dr.readOne(index)
for p in range(len(im)):
if im[p] >= 128:
self.layers[0].outs[p] = 1.0
else:
self.layers[0].outs[p] = 0.0
for j in range(k):
toHiddenLayer(self.layers[j], self.weightLayers[j], self.layers[j+1])
#self.layers[j+1].outs = sample(self.layers[j+1].outs)
sample(self.layers[j+1].outs)
rbm.training(1)
rbm.update_Parameters(size)
#if tr%100 == 0:
#print tr
print "Layer: ", k+1, " Completed!"
return
def result(self, im, label):
self.layers[0].outs = im.copy()
self.layers[0].nets = im.copy()
for i in range(self.depth-2):
toHiddenLayer(self.layers[i], self.weightLayers[i], self.layers[i+1])
toOutputLayer(self.layers[self.depth-2], self.weightLayers[self.depth-2], self.layers[self.depth-1])
return self.layers[self.depth-1].outs[label]
def result_print(self, im, label):
self.layers[0].outs = im.copy()
self.layers[0].nets = im.copy()
for i in range(self.depth-2):
toHiddenLayer(self.layers[i], self.weightLayers[i], self.layers[i+1])
toOutputLayer(self.layers[self.depth-2], self.weightLayers[self.depth-2], self.layers[self.depth-1])
r = findMax(self.layers[self.depth-1])
if label == r :
return 1
else:
'''
for i in range(10):
print i, " ", self.layers[self.depth-1].outs[i]
print label
print "------------------------------------------------------------------------------"
'''
return 0
