bp.zip_BP 分类 matlab_BP语音_神经网络

import random import numpy as np import matplotlib.pyplot as plt random.seed(0) # 使random函数每一次生成的随机数值都相等 def rand(a, b): return (b - a) * random.random() + a # 生成a到b的随机数 def make_matrix(m, n, fill=0.0): mat = [] for i in range(m): mat.append([fill] * n) # 生成m行n列的0矩阵 return mat def sigmoid(x): #return (1.0 - np.exp(-x)) / (1.0 + np.exp(-x)) #sigmoid双极型型激励函数 return (np.exp(x) - np.exp(-x)) / (np.exp(x) + np.exp(-x)) # tanh型激励函数 def sigmoid_derivative(x): return 0.5* (1 - x*x)#sigmoid激励函数的导数 #return 1.0 - x * x # tanh激励函数的导数 class BPNeuralNetwork: # 初始化参数 def __init__(self): self.input_n = 0 self.hidden_n = 0 self.hidden_second_n = 0 self.output_n = 0 self.input_cells = [] self.hidden_cells = [] self.hidden_second_cells = [] self.output_cells = [] self.input_weights = [] self.hidden_weights=[] self.output_weights = [] self.bais_h=[] self.bais_hs=[] self.bais_o =[] def setup(self, ni, nh, nhs, no): self.input_n = ni + 1 self.hidden_n = nh self.hidden_second_n = nhs self.output_n = no # 初始化神经元,建立相应的向量 self.input_cells = [1.0] * self.input_n self.hidden_cells = [1.0] * self.hidden_n self.hidden_second_cells = [1.0] * self.hidden_second_n self.output_cells = [1.0] * self.output_n # 初始化话权矩阵 self.input_weights = make_matrix(self.input_n, self.hidden_n) # 初始化输入权函数矩阵初始为零矩阵 self.hidden_weights = make_matrix(self.hidden_n,self.hidden_second_n) self.output_weights = make_matrix(self.hidden_second_n, self.output_n) # 初始化输出权函数矩阵 self.bais_h=make_matrix(self.hidden_n,1) self.bais_o=make_matrix(self.output_n,1) self.bais_hs = make_matrix(self.hidden_second_n, 1) # 随机激励 for i in range(self.input_n): for h in range(self.hidden_n): self.input_weights[i][h] = rand(-1.0,1.0) # 对输入权矩阵随机赋值 for h in range(self.hidden_n): for o in range(self.hidden_second_n): self.hidden_weights[h][o] = rand(-1.0,1.0) # 对输出权矩阵随机赋值 for o in range(self.hidden_second_n): for k in range(self.output_n): self.output_weights[o][k] = rand(-1.0, 1.0) # 对输出权矩阵随机赋值 for jj in range(self.hidden_n): self.bais_h[jj]= rand(-1.0,1.0) for dd in range(self.hidden_second_n): self.bais_hs[dd] = rand(-1.0, 1.0) for pp in range(self.output_n): self.bais_o[pp] = rand(-1.0, 1.0) # 前馈预测函数 def predict(self, inputs): # 刺激输入层 for i in range(self.input_n - 1): self.input_cells[i] = inputs[i] # 刺激隐藏 for j in range(self.hidden_n): total = 0.0 for i in range(self.input_n): total += self.input_cells[i] * self.input_weights[i][j] self.hidden_cells[j] = sigmoid(total+self.bais_h[j]) for k in range(self.hidden_second_n): total = 0.0 for j in range(self.hidden_n): total += self.hidden_cells[j] * self.hidden_weights[j][k] self.hidden_second_cells[k] = sigmoid(total + self.bais_hs[k]) for o in range(self.output_n): total = 0.0 for k in range(self.hidden_second_n): total += self.hidden_second_cells[k] * self.output_weights[k][o] self.output_cells[o] = sigmoid(total+self.bais_o[o]) return self.output_cells[:] # 返回输出层的值 # 反馈函数 def back_propagate(self, case, label, learn): # 前项传递 self.predict(case) # 给前馈预测函数赋值 learn_o=learn_h=learn_hs=learn # 获得输出层误差 output_deltas = [0.0] * self.output_n # 输出增量的向量 for o in range(self.output_n): error = label[o] - self.output_cells[o] # 训练值减输出值得到输出误差 c=output_deltas[o] output_deltas[o] = sigmoid_derivative(self.output_cells[o]) * error # 样本输出减输出层输出再乘增量 if(output_deltas[o]<c): learn_o = learn_o * 0.7 elif(output_deltas[o]>c): learn_o = 1.05 * learn_o else: learn_o = learn_o # 获得隐藏第二层误差 hidden_second_deltas = [0.0] * self.hidden_second_n for h in range(self.hidden_second_n): error = 0.0 for o in range(self.output_n): error += output_deltas[o] * self.output_weights[h][o] cc=hidden_second_deltas[h] hidden_second_deltas[h] = sigmoid_derivative(self.hidden_second_cells[h]) * error if (hidden_second_deltas[h]< cc): learn_hs = learn_hs * 0.7 elif (hidden_second_deltas[h] > cc): learn_hs = 1.05 * learn_hs else: learn_hs = learn_hs # 获得隐藏层误差 hidden_deltas = [0.0] * self.hidden_n for j in range(self.hidden_n): error = 0.0 for h in range(self.hidden_second_n): error += hidden_second_deltas[h] * self.hidden_weights[j][h] ccc=hidden_deltas[j] hidden_deltas[j] = sigmoid_derivative(self.hidden_cells[j]) * error if (hidden_deltas[j]< ccc): learn_h = learn_h * 0.7 elif (hidden_deltas[j] > ccc): learn_h = 1.05 * learn_h else: learn_h = learn_h # 更新输出层权系数 for h in range(self.hidden_second_n): for o in range(self.output_n): change = output_deltas[o] * self.hidden_second_cells[h] self.output_weights[h][o] = self.output_weights[h][o] +learn_o * change # 更新隐藏第二层权系数 for j in range(self.hidden_n): for h in range(self.hidden_second_n): change = hidden_second_deltas[h] * self.hidden_cells[j] self.hidden_weights[j][h] = self.hidden_weights[j][h] + learn_hs * change # 更新输入权重 for i in range(self.input_n): for h in range(self.hidden_n): change = hidden_deltas[h] * self.input_cells[i] self.input_weights[i][h]=self.input_weights[i][h]+learn_h * change for i in range(self.hidden_n): self.bais_h[i] = 0.05 * learn_h * hidden_deltas[i]+0.95*self.bais_h[i] for k in range(self.hidden_second_n): self.bais_hs[k] = 0.05 * learn_hs * hidden_second_deltas[k] + 0.95 * self.bais_hs[k] for j in range(self.output_n): self.bais_o[j] = 0.05*learn_o*output_deltas[j]+0.95*self.bais_o[j] # 得到全局变量方差 error = 0.0 for o in range(len(label)): error += 0.5 * (label[o] - self.output_cells[o]) ** 2 return error def train(self, cases, labels, limit, learn): for j in range(limit): error = 0.0 for i in range(len(cases)): label = labels[i] case = cases[i] error += self.back_propagate(case, label, learn) def test(self): #a = np.loadtxt('test_data.txt') #test_d = np.matrix(a).T #xxx = test_d [0].T #yyy = test_d [