人工智能与实践 bilibili.zip

import numpy as np from matplotlib import pyplot as plt def sigmoid(x): return 1/(1+np.exp(-x)) # def init_ws(number_of_previous_hidden_layers, number_of_current_hidden_layers): # return np.arange(number_of_previous_hidden_layers*number_of_current_hidden_layers) # def init_bs(number_of_current_hidden_layers): # return np.arange(number_of_current_hidden_layers) # def forward_propagation(ws, a, bs, number_of_previous_hidden_layers, number_of_current_hidden_layers): # zs = [] # a_s = [] # z = 0 # for i in range(number_of_current_hidden_layers): # w_start_index = i*number_of_previous_hidden_layers # for j in range(number_of_previous_hidden_layers): # z += a[j]*ws[w_start_index]+bs[i] # w_start_index += 1 # zs.append(z) # a_s.append(sigmoid(z)) # return zs, a_s m = 100 virulence = np.random.rand(m) volume = np.random.rand(m) virulence.sort() volume.sort() for i in range(m): if virulence[i] < 0.3 or virulence[i] > 0.7: virulence[i] = 0 else: virulence[i] = 1 w_1_1_1 = np.random.rand() w_1_2_1 = np.random.rand() w_1_1_2 = np.random.rand() w_2_1_2 = np.random.rand() b_1_1 = np.random.rand() b_2_1 = np.random.rand() b_1_2 = np.random.rand() def forward_progation(xs): z1_1 = w_1_1_1*xs+b_1_1 a1_1 = sigmoid(z1_1) z2_1 = w_1_2_1*xs+b_2_1 a2_1 = sigmoid(z2_1) z1_2 = w_1_1_2*a1_1+b_1_2+w_2_1_2*a2_1+b_1_2 a1_2 = sigmoid(z1_2) return z1_1, a1_1, z2_1, a2_1, z1_2, a1_2 z1_1, a1_1, z2_1, a2_1, z1_2, a1_2 = forward_progation(volume) alpha = 0.5 def BP(x, y, a1, w2, a2): deda2 = 2*(a2-y) da2dz2 = a2*(1-a2) dz2dw2 = a1 dedw2 = deda2*da2dz2*dz2dw2 dz2db2 = 1 dedb2 = deda2*da2dz2*dz2db2 dz2da1 = w2 da1dz1 = a1*(1-a1) dz1dw1 = x dedw1 = deda2*da2dz2*dz2da1*da1dz1*dz1dw1 dz1db1 = 1 dedb1 = deda2*da2dz2*dz2da1*da1dz1*dz1db1 return dedw1, dedb1, dedw2, dedw1 for j in range(5500): for i in range(m): x = volume[i] y = virulence[i] z1_1, a1_1, z2_1, a2_1, z1_2, a1_2 = forward_progation(x) dedw1_1_1, dedb1_1, dedw1_1_2, dedb1_2 = BP(x, y, a1_1, w_1_1_2, a1_2) w_1_1_1 = w_1_1_1 - alpha*dedw1_1_1 w_1_1_2 = w_1_1_2 - alpha*dedw1_1_2 b_1_1 = b_1_1 - alpha*dedb1_1 b_1_2 = b_1_2 - alpha*dedb1_2 dedw1_2_1, dedb2_1, dedw2_1_2, dedb1_2 = BP(x, y, a2_1, w_2_1_2, a1_2) w_1_2_1 = w_1_2_1 - alpha*dedw1_2_1 w_2_1_2 = w_2_1_2 - alpha*dedw2_1_2 b_2_1 = b_2_1 - alpha*dedb2_1 b_1_2 = b_1_2 - alpha*dedb1_2 if j % 100 == 0: plt.clf() plt.xlabel("volume") plt.ylabel("Probability of toxicity") plt.scatter(volume, virulence) plt.xlim(0, 1.1) plt.ylim(0, 1.1) z1_1, a1_1, z2_1, a2_1, z1_2, a1_2 = forward_progation(volume) plt.plot(volume, a1_2) plt.pause(0.01) plt.show()