使用CD-K算法实现RBM(代码).docx

#encoding:utf-8 import matplotlib.pylab as plt import numpy as np import random from scipy.linalg import norm import PIL.Image class Rbm: def __init__(self,n_visul, n_hidden, max_epoch = 50, batch_size = 110, penalty = 2e-4, anneal = False, w = None, v_bias = None, h_bias = None): self.n_visible = n_visul self.n_hidden = n_hidden self.max_epoch = max_epoch self.batch_size = batch_size self.penalty = penalty self.anneal = anneal if w is None: self.w = np.random.random((self.n_visible, self.n_hidden)) * 0.1 #初始化可见层到隐层的权重矩阵 if v_bias is None: self.v_bias = np.zeros((1, self.n_visible)) if h_bias is None: self.h_bias = np.zeros((1, self.n_hidden)) def sigmod(self, z): return 1.0 / (1.0 + np.exp( -z )) #定义一个激活函数 def forward(self, vis): #if(len(vis.shape) == 1): #vis = np.array([vis]) #vis = vis.transpose() #if(vis.shape[1] != self.w.shape[0]): vis = vis.transpose() pre_sigmod_input = np.dot(vis, self.w) + self.h_bias #按照矩阵乘法进行相乘 return self.sigmod(pre_sigmod_input) def backward(self, vis): #if(len(vis.shape) == 1): #vis = np.array([vis]) #vis = vis.transpose() #if(vis.shape[0] != self.w.shape[1]): back_sigmod_input = np.dot(vis, self.w.transpose()) + self.v_bias return self.sigmod(back_sigmod_input) def batch(self): eta = 0.1 momentum = 0.5 d,N = self.x.shape num_batchs = int(round(N / self.batch_size)) + 1 #训练批次大小 groups = np.ravel(np.repeat([range(0, num_batchs)], self.batch_size, axis = 0)) groups = groups[0 : N] perm = range(0, N) random.shuffle(perm) groups = groups[perm] batch_data = [] for i in range(0, num_batchs): index = groups == i batch_data.append(self.x[:, index]) return batch_data def rbmBB(self, x): self.x = x eta = 0.1 momentum = 0.5 W = self.w b = self.h_bias c = self.v_bias Wavg = W bavg = b cavg = c Winc = np.zeros((self.n_visible, self.n_hidden)) binc = np.zeros(self.n_hidden) cinc = np.zeros(self.n_visible) avgstart = self.max_epoch - 5; batch_data = self.batch() num_batch = len(batch_data) oldpenalty= self.penalty t = 1 errors = [] for epoch in ran