BP网络——前馈神经网络
(Back Propgation Networks)
本文将实现一个最简单的三层神经网络
其中,损失函数是线性函数,激励函数是sigmoid函数。
代码的实现中,采用的是随机梯度下降法。
计算导数的方法可以参考图片,有兴趣的小伙伴可以参考,自行推导:
随机梯度下降法的原理
代码如下所示:
#!/user/bin/env python3
# -*- coding: utf-8 -*-
import random
import numpy as np
#定义神经网络结构
class Network(object):
def __init__(self, sizes):
# 网络层数
self.num_layers = len(sizes)
# 每层神经元的个数
self.sizes = sizes
# 初始化每层的偏置
self.biases = [np.random.randn(y, 1) for y in sizes[1:]]
# 初始化每层的权重
self.weights = [np.random.randn(y, x) for x, y in zip(sizes[:-1], sizes[1:])]
def feedforward(self, a):
for b, w in zip(self.biases, self.weights):
a = sigmoid(np.dot(w, a) + b)
return a
# 梯度下降
def SGD(self, training_data, epochs, mini_batch_size, eta, test_data=None):
if test_data:
n_test = len(test_data)
# 训练数据总个数
n = len(training_data)
# 开始训练,循环每一个epochs
for j in range(epochs): # 在python2.7中为xrange
# 洗牌 打乱训练数据
random.shuffle(training_data)
# mini_batch
mini_batches = [training_data[k:k+mini_batch_size] for k in range(0, n, mini_batch_size)]
# 训练mini_batch
for mini_batch in mini_batches:
self.update_mini_batch(mini_batch, eta)
if test_data:
print("Epoch {0}: {1} / {2}".format(j, self.evaluate(test_data), n_test))
print("Epoch {0} complete".format(j))
def update_mini_batch(self, mini_batch, eta):
# 保存每层偏导
nabla_b = [np.zeros(b.shape) for b in self.biases]
nabla_w = [np.zeros(w.shape) for w in self.weights]
# 训练一个mini_batch
for x, y in mini_batch:
delta_nabla_b, delta_nabla_w = self.update(x, y)
# 保存一次训练网络中每层的偏导
nabla_b = [nb + dnb for nb, dnb in zip(nabla_b, delta_nabla_b)]
nabla_w = [nw + dnw for nw, dnw in zip(nabla_w, delta_nabla_w)]
# 更新权重和偏置 Wn+1 = Wn - eta * nw
self.weights = [w - (eta / len(mini_batch)) * nw for w, nw in zip(self.weights, nabla_w)]
self.biases = [b - (eta / len(mini_batch)) * nb for b, nb in zip(self.biases, nabla_b)]
# 前向传播
def update(self, x, y):
# 保存每层偏导
nabla_b = [np.zeros(b.shape) for b in self.biases]
nabla_w = [np.zeros(w.shape) for w in self.weights]
activation = x # 保存的输入(训练数据)
# 保存每一层的激励值a=sigmoid(z)
activations = [x]
# 保存每一层的z=wx+b
zs = []
# 前向传播
for b, w in zip(self.biases, self.weights):
# 计算每层的z
z = np.dot(w, activation) + b
# 保存每层的z
zs.append(z)
# 计算每层的a
activation = sigmoid(z)
# 保存每一层的a
activations.append(activation)
# 反向更新
# 计算最后一层的误差
delta = self.cost_derivative(activations[-1], y) * sigmoid_prime(zs[-1])
# 最后一层权重和偏置的导数
nabla_b[-1] = delta
nabla_w[-1] = np.dot(delta, activations[-2].T)
# 倒数第二层一直到第一层 权重和偏置的导数
for l in range(2, self.num_layers):
z = zs[-l]
sp = sigmoid_prime(z)
# 当前层的误差
delta = np.dot(self.weights[-l+1].T, delta) * sp
# 当前层的偏置和权重的导数
nabla_b[-l] = delta
nabla_w[-l] = np.dot(delta, activations[-l-1].T)
return (nabla_b, nabla_w)
def evaluate(self, test_data):
test_results = [(np.argmax(self.feedforward(x)), y) for (x, y) in test_data]
return sum(int(x == y) for x, y in test_results)
def cost_derivative(self, output_activation, y):
return (output_activation - y)
# sigmoid激励函数
def sigmoid(z):
return 1.0 / (1.0 + np.exp(-z))
def sigmoid_prime(z):
return sigmoid(z) * (1-sigmoid(z))
if __name__ == "__main__":
import mnist_loader
training_data, validation_data, test_data = mnist_loader.load_data_wrapper()
# 28*28=784个像素, 可以定义30个神经元, 共有10种分类
net = Network([784, 30, 10])
net.SGD(training_data, 30, 10, 0.5, test_data=test_data)
由于进行预测的时候用的是mnist数据集
所以需要下载mnist.pkl.gz
mnist.pkl.gz下载地址为:mnist.pkl.gz
小伙伴千万不要下错哦,之前我下的就是个数据不够的,结果害得我查了好久为什么代码会出错,哭唧唧。。。
注:mnist_loader.py是一个python文件(如下)
#!/user/bin/env python3
# -*- coding: utf-8 -*-
"""
mnist_loader
~~~~~~~~~~~~
A library to load the MNIST image data. For details of the data
structures that are returned, see the doc strings for ``load_data``
and ``load_data_wrapper``. In practice, ``load_data_wrapper`` is the
function usually called by our neural network code.
"""
#### Libraries
# Standard library
import pickle
import gzip
import numpy as np
def load_data():
f = gzip.open('data/mnist.pkl.gz', 'rb')
training_data, validation_data, test_data = pickle.load(f, encoding='bytes')
f.close()
return (training_data, validation_data, test_data)
def load_data_wrapper():
tr_d, va_d, te_d = load_data()
training_inputs = [np.reshape(x, (784, 1)) for x in tr_d[0]]
training_results = [vectorized_result(y) for y in tr_d[1]]
training_data = list(zip(training_inputs, training_results))
validation_inputs = [np.reshape(x, (784, 1)) for x in va_d[0]]
validation_data = list(zip(validation_inputs, va_d[1]))
test_inputs = [np.reshape(x, (784, 1)) for x in te_d[0]]
test_data = list(zip(test_inputs, te_d[1]))
return (training_data, validation_data, test_data)
def vectorized_result(j):
e = np.zeros((10, 1))
e[j] = 1.0
return e
大家一定要注意python版本的问题,我用的是python3,和python2还是有差别的,比如以下几点:
①python3中没有xrange,python3中的range实际上等同于python2中的xrange。
②好像pickle的使用也有所不同,在python2中是cpickle,但python3中是pickle;
很多人出现了一个问题,就是python3使用pickle.load()方法时,需要加上pickle.load(f, encoding='bytes')
encoding编码方式才能解决。
③python2中的zip方法和python3不同,python3的zip()不是列表,所以在上述代码的load_data_wrapper()方法中,zip出现的地方需要加上list()方法。
注:
代码仍然有很多可以进行提升的部分,见下一篇博客:反向传播(BP)网络的mnist实例(提升篇)