手写数字识别——BP神经网络实现

手写数字识别——BP神经网络实现

贴个代码，数据集是mnist，可以直接去官网下载。

#main.py
import os
from load_file import load_mnist
from neuralNetwork import neuralNetwork

if __name__ == '__main__':
    train_images_data, train_labels = load_mnist(os.getcwd(), kind='train')
    test_images_data, test_labels = load_mnist(os.getcwd(), kind='t10k')

    learningrate = 0.1
    ls = [28*28, 30, 10]
    Network = neuralNetwork(2, ls, learningrate)

    for i in range(10):
        if i%5 == 0:  Network.learningrate /= 2
        for j in range(len(train_images_data)):
            print("第" + str(i) + "轮" + "-------正在进行训练-------" + str(j))
            Network.update(train_images_data[j], train_labels[j])

    count = 0
    for i in range(len(test_images_data)):
        print('-------正在进行测试--------')
        if Network.test(test_images_data[i], test_labels[i]): count += 1

    print(count / len(test_labels))

#neuralNetwork.py
import numpy as np
from activateFunction import activate_function

class neuralNetwork:
    def __init__(self, numNeuralLayers, numNeurons_perLayer, learningrate):
        self.numBeuralLayers = numNeuralLayers
        self.numNeurons_perLayer = numNeurons_perLayer
        self.learningrate = learningrate
        self.W = []

        for i in range(self.numBeuralLayers):
            #高斯分布进行初始化
            self.W.append(np.random.randn(numNeurons_perLayer[i + 1], numNeurons_perLayer[i])  / np.sqrt(numNeurons_perLayer[i]))

    def update(self, inputs, targets):
        inputs = np.array(inputs, ndmin = 2).T
        targets = np.array(targets, ndmin = 2).T

        outputs = []
        outputs.append(inputs)

    #前向传播
        for i in range(self.numBeuralLayers):
            a = np.dot(self.W[i], outputs[i])
            z = activate_function(a)
            outputs.append(z)

    #计算损失
        self.loss = []
        for i in range(self.numBeuralLayers):
            if i == 0:
                self.loss.append(targets - outputs[-1])
            else:
                self.loss.append(np.dot(self.W[self.numBeuralLayers - i].T, self.loss[i - 1]))

    #反向传播
        for i in range(self.numBeuralLayers):
            self.W[self.numBeuralLayers - i - 1] += self.learningrate * np.dot(self.loss[i] * outputs[self.numBeuralLayers - i] * (1 -outputs[self.numBeuralLayers -i]),
                                                                                outputs[self.numBeuralLayers - i - 1].T)

    #测试
    def test(self, test_inputs, test_targets):
        inputs = test_inputs.T

        for i in range(self.numBeuralLayers):
            a = np.dot(self.W[i], inputs)
            z = activate_function(a)
            inputs = z

        return list(inputs).index(max(list(inputs))) == list(test_targets).index(1)

#activateFunction.py
import numpy as np

def activate_function(x):
    #sigmoid
    return 1.0 / (1.0 + np.exp(-x))

#load_file.py
import os
import struct
import numpy as np

def load_mnist(path, kind = 'train'):
    labels_path = os.path.join(path, '%s-labels.idx1-ubyte'% kind)
    images_path = os.path.join(path, '%s-images.idx3-ubyte'% kind)

    with open(labels_path, 'rb') as lbpath:
        magic, n = struct.unpack('>II',lbpath.read(8))
        labels = np.fromfile(lbpath, dtype = np.uint8)
    with open(images_path, 'rb') as imgpath:
        magic, num, rows, cols = struct.unpack('>IIII', imgpath.read(16))
        images = np.fromfile(imgpath,dtype = np.uint8).reshape(len(labels), 784)

        ls = []
        for i in range(labels.size):
            temp = np.zeros(10)
            temp[labels[i]] = 1
            ls.append(temp.T)

        labels = np.array(ls)

        return (images / 2550), labels