代码实现三层神经网络的手写字训练及测试

import numpy
import scipy.special
import matplotlib.pyplot
class network:
    def __init__(self , inputnodes, hiddennodes, outputnodes,learningrate ):
        self.inputnodes=inputnodes
        self.hiddennodes=hiddennodes
        self.outputnodes=outputnodes
        self.lr=learningrate
        # 计算输入层隐藏层的权重系数
        self.wih = numpy.random.normal(0.0, pow(self.hiddennodes, -0.5),(self.hiddennodes, self.inputnodes))
        # 计算隐藏层输出层的权重系数
        self.who = numpy.random.normal(0.0, pow(self.outputnodes, -0.5),(self.outputnodes, self.hiddennodes))
        # 定义激活函数，使用scipy中的expit函数
        self.activation_function = lambda x:scipy.special.expit(x)
    # 训练神经网络
    def train(self, inputs_list, targets_list):
        # 将输入转化为2维矩阵
        inputs = numpy.array(inputs_list, ndmin=2).T
        targets = numpy.array(targets_list, ndmin=2).T
        # 计算隐藏层的输入
        hidden_inputs = numpy.dot(self.wih, inputs)
        # 计算激活函数处理后隐藏层的输入变成输出
        hidden_outputs = self.activation_function(hidden_inputs)
        # 计算输出层的输入
        final_inputs = numpy.dot(self.who, hidden_outputs)
        #计算激活函数处理后输出层的输入变成输出
        final_outputs = self.activation_function(final_inputs)
        output_errors = targets - final_outputs
        # 计算输出层隐藏层的误差矩阵
        hidden_errors = numpy.dot(self.who.T, output_errors)
        # 更新输出层隐藏层的权重
        self.who += self.lr * numpy.dot((output_errors *final_outputs * (1.0 - final_outputs)),numpy.transpose(hidden_outputs))
        # 更新隐藏层隐藏层的权重
        self.wih += self.lr * numpy.dot((hidden_errors *hidden_outputs * (1.0 - hidden_outputs)), numpy.transpose(inputs))
        pass
            # query the neural network
    def query(self,inputs_list):
        # 将输入转化为二维矩阵
        inputs = numpy.array(inputs_list, ndmin=2).T
        # 计算隐藏层的输入
        hidden_inputs = numpy.dot(self.wih, inputs)
        # 计算激活函数处理后隐藏层的输入变成输出
        hidden_outputs = self.activation_function(hidden_inputs)
        # 计算输出层的输入
        final_inputs = numpy.dot(self.who, hidden_outputs)
        # 计算激活函数处理后输出层的输入变成输出
        final_outputs = self.activation_function(final_inputs)
        return final_outputs
        pass

#加载数据进行训练
def train_datas(output_nodes,network):
    with open('/home/opprash/Desktop/mnist_train_100.csv', 'r') as f:
        data_list = f.readlines()
        f.close()
    #可视化csv中的数字的意义
    # all_values = data_list[1].split(',')
    # image_array = numpy.asfarray(all_values[1:]).reshape((28, 28))
    # matplotlib.pyplot.imshow(image_array, cmap='Greys', interpolation='None')
    # matplotlib.pyplot.show()
    # scaled_input = (numpy.asfarray(all_values[1:]) / 255.0 * 0.99) + 0.01
    # print(scaled_input)

    # 训练神经网络
    # 遍历训练集的所有数据
    for record in data_list:
        # 以逗号分割
        all_values = record.split(',')
        # 将数据压缩到0到1之间，原始数据是在0-255之间
        inputs = (numpy.asfarray(all_values[1:]) / 255.0 * 0.99) + 0.01
        # 创建目标输出值（全部0.01，除了所需的值标签是0.99）
        targets = numpy.zeros(output_nodes) + 0.01
        # all_values[0] 是此记录的目标标签
        targets[int(all_values[0])] = 0.99
        network.train(inputs, targets)
        pass

#加载数据进行测试
def test_data(network):
    with open('/home/opprash/Desktop/mnist_test_10.csv', 'r') as f:
        data_list = f.readlines()
        f.close()
    scorecard = []
    for record in data_list:
        # 以逗号分割
        all_values = record.split(',')
        # 正确答案是第一个值
        correct_label = int(all_values[0])
        print( "真实的标签",correct_label)
        # 将数据压缩到0到1之间，原始数据是在0-255之间
        inputs = (numpy.asfarray(all_values[1:]) / 255.0 * 0.99) + 0.01
        # 使用网络进行查询
        outputs = network.query(inputs)
        # 最高值的索引就是对应的标签
        label = numpy.argmax(outputs)
        print("神经网络预测的答案：",label)
        # 追加正确或者不正确的list
        if (label == correct_label):
        # 网络的答案匹配正确的答案，添加1
            scorecard.append(1)
        else:
        # 网络的答案匹配不正确的答案，添加0
            scorecard.append(0)
        scorecard_array = numpy.asarray(scorecard)
    print("准确率 = ", scorecard_array.sum()/scorecard_array.size)

if __name__ == '__main__':
    # 定义输入，隐藏层，输出层节点个数
    input_nodes = 784
    hidden_nodes = 100
    output_nodes = 10
    # 学习率 0.3
    learning_rate = 0.3
    # 实例化神经网络
    n = network(input_nodes,hidden_nodes,output_nodes,learning_rate)
    train_datas(output_nodes,n)
    test_data(n)