def _gradient(self, weights, lambda_value):
thetas = list(reshape_vector(weights, self.theta_shapes))
activations = self._activations(thetas)
l = lambda_value
# sigmas = [sigma3, sigma2, ...]
sigmas = [activations[-1]-self.Y]
for ind,layer in enumerate(self.z[-2::-1]):
if self.add_bias:
layer = add_bias(layer)
sigma = np.dot(sigmas[-1], thetas[-1-ind])*sigmoid_gradient(layer)
sigmas.append(del_bias(sigma))
# deltas = [delta1, delta2, ...]
deltas = []
for activation, sigma in zip(activations, sigmas[::-1]):
deltas.append(np.dot(sigma.T, activation))
# gradients = [theta1_grad, theta2_grad, ...]
gradients = []
for theta, delta in zip(thetas, deltas):
theta = del_bias(theta)
theta = add_bias(theta, values_function=np.zeros)
gradient = delta/self.m + np.dot((l/self.m), theta)
gradients.append(gradient.T.ravel())
return np.concatenate(gradients)
def _gradient(self, weights, lambda_value):
thetas = list(reshape_vector(weights, self.theta_shapes))
activations = self._activations(thetas)
l = lambda_value
# sigmas = [sigma3, sigma2, ...]
sigmas = [activations[-1] - self.Y]
for ind, layer in enumerate(self.z[-2::-1]):
if self.add_bias:
layer = add_bias(layer)
sigma = np.dot(sigmas[-1],
thetas[-1 - ind]) * sigmoid_gradient(layer)
sigmas.append(del_bias(sigma))
# deltas = [delta1, delta2, ...]
deltas = []
for activation, sigma in zip(activations, sigmas[::-1]):
deltas.append(np.dot(sigma.T, activation))
# gradients = [theta1_grad, theta2_grad, ...]
gradients = []
for theta, delta in zip(thetas, deltas):
theta = del_bias(theta)
theta = add_bias(theta, values_function=np.zeros)
gradient = delta / self.m + np.dot((l / self.m), theta)
gradients.append(gradient.T.ravel())
return np.concatenate(gradients)
def _activations(self, thetas):
if self.add_bias:
input_layer = add_bias(self.X)
else:
input_layer = self.X
# activations = [a1, a2, ...]
activations = [input_layer]
self.z = []
# Process hidden layers
for i in range(len(thetas)):
self.z.append(np.dot(activations[-1], thetas[i].T))
activations.append(sigmoid(self.z[-1]))
# Don't add bias terms on the last layer
if self.add_bias and i < len(thetas)-1:
activations[-1] = add_bias(activations[-1])
return activations
def _activations(self, thetas):
if self.add_bias:
input_layer = add_bias(self.X)
else:
input_layer = self.X
# activations = [a1, a2, ...]
activations = [input_layer]
self.z = []
# Process hidden layers
for i in range(len(thetas)):
self.z.append(np.dot(activations[-1], thetas[i].T))
activations.append(sigmoid(self.z[-1]))
# Don't add bias terms on the last layer
if self.add_bias and i < len(thetas) - 1:
activations[-1] = add_bias(activations[-1])
return activations
bc1 = np.load('../../Predictor/server_net/models/model_16/biases/bc1.npy')
bc2 = np.load('../../Predictor/server_net/models/model_16/biases/bc2.npy')
bf1 = np.load('../../Predictor/server_net/models/model_16/biases/bd1.npy')
otb = np.load('../../Predictor/server_net/models/model_16/biases/bout.npy')
test_im = np.reshape(test_im,(1,28,28,1))
#conv2d takes (1,28,28,1) image and (5,5,1,5) kernel and gives (1,28,28,5) output
con1 = conv2d(test_im,wc1)
# np.save('con1.npy',con1)
print "conv1"
print con1
#add_bias takes the 5 (1,28,28) inputs and 5 biases, and adds the bias to every pixel in the 5 inputs
con1 = add_bias(con1,bc1)
# np.save('con1bias.npy',con1)
print "conv1addbias"
print con1
#activ_fun takes (1,28,28,5) input and squares each and every element in the input
con1 = activ_fun(con1)
# np.save('con1act.npy',con1)
print "conv1act"
print con1
#meanpool2 takes (1,28,28,5) input and performs meanpooling on each of the 5 28x28 matrices seperately
#and gives a (1,14,14,5) output
mean1 = meanpool2(con1)
# np.save('con1mean.npy',mean1)
print "mean pooling 1"
print mean1
#conv2d takes (1,14,14,5) input and (5,5,5,10) kernel and gives (1,14,14,10) output