def greedy_recon(dbnet, test):
temp = test
top_inputsize = dbnet['top']['v_num'] - dbnet['top']['label_n']
for i in range(len(dbnet['layer'])): #bottom up
temp = ReLU(temp, dbnet['layer'][i]['w'])
top = ReLU(temp, dbnet['top']['w'][:top_inputsize, :])
label = ReLU(top, np.transpose(dbnet['top']['w'][top_inputsize:, :]))
temp = np.append(temp, label, axis=1)
temp = ReLU(temp, dbnet['top']['w'])
temp = ReLU(temp, np.transpose(dbnet['top']['w']))
temp = temp[:top_inputsize]
for i in range(len(dbnet['layer']) - 1, -1, -1): #up down
temp = ReLU(temp, np.transpose(dbnet['layer'][i]['w']))
recon = temp
mu.plot_digit(recon)
predict = np.argmax(label)
return predict, recon
def greedy_recon(dbnet, test):
temp = test
top_inputsize = dbnet['top']['v_num'] - dbnet['top']['label_n']
for i in range(len(dbnet['layer'])): #bottom up
temp = sigmoid_sampling(temp, dbnet['layer'][i]['w'], dbnet['layer'][i]['b'])
top = sigmoid_sampling(temp, dbnet['top']['w'][:top_inputsize, :], dbnet['top']['b'])
label = sigmoid_sampling(top, np.transpose(dbnet['top']['w'][top_inputsize:, :]), dbnet['top']['a'][top_inputsize:])
temp = np.append(temp, label, axis=1)
temp = sigmoid_sampling(temp, dbnet['top']['w'], dbnet['top']['b'])
temp = sigmoid_sampling(temp, np.transpose(dbnet['top']['w']), dbnet['top']['a'])
temp = temp[:top_inputsize]
for i in range(len(dbnet['layer'])-1, 0, -1): #up down
temp = sigmoid_sampling(temp, np.transpose(dbnet['layer'][i]['w']), dbnet['layer'][i]['a'])
recon = sigmoid(temp, np.transpose(dbnet['layer'][0]['w']), dbnet['layer'][0]['a'])
mu.plot_digit(recon)
predict = np.argmax(label)
return predict, recon
def dbn_recon(dbnet, test):
temp = test
top_inputsize = dbnet['top']['v_num'] - dbnet['top']['label_n']
for i in range(len(dbnet['layer'])): #bottom up
temp = ReLU(temp, dbnet['layer'][i]['w_up'])
top = ReLU(temp, dbnet['top']['w'][:top_inputsize, :])
label = ReLU(top, np.transpose(dbnet['top']['w'][top_inputsize:, :]))
temp = np.append(temp, label, axis=1)
temp = ReLU(temp, dbnet['top']['w'])
temp = ReLU(temp, np.transpose(dbnet['top']['w']))
temp = temp[:top_inputsize]
for i in range(len(dbnet['layer'])-1, -1, -1): #up down
temp = ReLU(temp, dbnet['layer'][i]['w_down'])
recon = temp
mu.plot_digit(recon)
predict = np.argmax(label)
return predict, recon
def dbn_recon(dbnet, test):
temp = test
top_inputsize = dbnet['top']['v_num'] - dbnet['top']['label_n']
for i in range(len(dbnet['layer'])): #bottom up
temp = sigmoid_sampling(temp, dbnet['layer'][i]['w_up'],
dbnet['layer'][i]['b'])
top = sigmoid_sampling(temp, dbnet['top']['w'][:top_inputsize, :],
dbnet['top']['b'])
label = sigmoid_sampling(
top, np.transpose(dbnet['top']['w'][top_inputsize:, :]),
dbnet['top']['a'][top_inputsize:])
temp = np.append(temp, label, axis=1)
temp = sigmoid_sampling(temp, dbnet['top']['w'], dbnet['top']['b'])
temp = sigmoid_sampling(temp, np.transpose(dbnet['top']['w']),
dbnet['top']['a'])
temp = temp[:top_inputsize]
for i in range(len(dbnet['layer']) - 1, 0, -1): #up down
temp = sigmoid_sampling(temp, dbnet['layer'][i]['w_down'],
dbnet['layer'][i]['a'])
recon = sigmoid(temp, dbnet['layer'][0]['w_down'], dbnet['layer'][0]['a'])
mu.plot_digit(recon)
predict = np.argmax(label)
return predict, recon
def plot_recon(digit_img, para):
data_v = np.array(digit_img).astype(float)
data_h, gibbs_v, gibbs_h = sampling(para, data_v)
mu.plot_digit(gibbs_v)
def plot_recon(digit_img, para):
data_v = np.array(digit_img).astype(float)
data_h, gibbs_v, gibbs_h = sampling_nb(para, data_v)
mu.plot_digit(gibbs_v)
