代码实现bp算法示例

最新推荐文章于 2021-10-18 15:21:23 发布

光英的记忆

最新推荐文章于 2021-10-18 15:21:23 发布

阅读量1.4k

点赞数

分类专栏： tensorflow 算法文章标签： BP算法

本文链接：https://blog.csdn.net/qq_29678299/article/details/93775218

版权

本文详细介绍了BP（反向传播）算法的工作原理，并通过一个具体的示例展示了如何用代码实现这一经典的人工神经网络训练算法。内容涵盖神经网络的基本结构、权重更新规则以及误差反向传播的过程。

摘要由CSDN通过智能技术生成

# encoding:utf-8
import numpy as np
import random


class Network(object):

    def __init__(self, sizes):
        self.num_layers = len(sizes)
        print("self.num_layers", self.num_layers)
        self.sizes = sizes
        self.biases = [np.random.randn(y, 1) for y in sizes[1:]]
        print("biases", self.biases)
        # biases [array([[-0.55676967],
        #        [-1.60486518],
        #        [ 0.28285971],
        #        [ 0.50856141]]), array([[-1.02563044]])]
        print("sizes[1:]", sizes[1:])
        # [4, 1]
        self.weights = [np.random.randn(y, x)
                        for x, y in zip(sizes[:-1], sizes[1:])]
        print("weights", self.weights)
        print("sizes[:-1]", sizes[:-1])
        # [3, 4]
        #  [array([[-0.60968692,  2.86107499,  0.25978906],
        #        [ 0.20304469, -1.42226893,  1.66564988],
        #        [-0.73320158,  0.53635735,  1.49307876],
        #        [-0.27301879, -1.42973446, -0.17142598]]), array([[ 0.39620414, -1.0758598 ,
        #        0.17447957, -0.67224015]])]

    def backprop(self, x, y):
        """return a tuple
        """

        nabla_b = [np.zeros(b.shape) for b in self.biases]
        nabla_w = [np.zeros(w.shape) for w in self.weights]
        # feedforward
        activation = x
        activations = [x]  # 存放激活值

        zs = []  # list用来存放z 向量

        # 前向传递
        for b, w in zip(self.biases, self.weights):
            z = np.dot(w, activation) + b
            zs.append(z)
            activation = self.sigmoid(z)
            activations.append(activation)

        # 后向传递
        delta = self.cost_derivative(activations[-1], y) * self.sigmoid(zs[-1])

        nabla_b[-1] = delta
        nabla_w[-1] = np.dot(delta, activations[-2].transpose())

        for l in range(2, self.num_layers):
            z = zs[-l]
            sp = self.sigmoid_prime(z)
            delta = np.dot(self.weights[-l + 1].transpose(), delta) * sp