EP2.numpy实现浅层神经网络

最新推荐文章于 2021-09-05 19:01:11 发布

原创最新推荐文章于 2021-09-05 19:01:11 发布 · 187 阅读

0 ·

CC 4.0 BY-SA版权

文章标签：

#deeplearning

本篇博客详细介绍了如何利用numpy从零开始实现一个简单的浅层神经网络，包括权重初始化、前向传播、损失计算及反向传播等关键步骤。通过实例代码，帮助读者深入理解神经网络的工作原理。


import numpy as np

def sigmoid(x):
  return 1/(1+np.exp(-x)) 

def train():
    hidden_units=4
    learning_rate=0.01
    m=train_x.shape[1]
    iteration_nums=10000
    input_dim=train_x.shape[0]
    output_dim=train_y.shape[0] 
    w1=np.zeros(shape=(hidden_units,input_dim))
    b1=np.zeros(shape=(hidden_units,1))
    w2=np.zeros(shape=(output_dim,hidden_units))
    b1=np.zeros(shape=(output_dim,1))
    for i in range(iteration_nums):
#前向传播(交叉熵损失)
        z1 = np.dot(w1,train_x) + b1
        a1 = np.tanh(z1)
        z2 = np.dot(w2, a1) + b2
        a2 = sigmoid(z2)
        cost=-1/m*np.sum(train_y*np.log(a2)+(1-train_y)*np.log(1-a2))
        print("cost after iteration %i:%f" % (i, cost))
#后向迭代更新参数
        dz2 = a2-train_y
        dw2 = np.dot(dz2,a1.T)/m
        db2 = np.sum(dz2)/m
        dz1 = np.dot(w2.T,dz2)*(1-np.power(a1,2))
        dw1 = np.dot(dz1,train_x.T)/m
        db1 = np.sum(dz1)/m
        w1=w1-dw1*learning_rate
        b1=b1-db1*learning_rate
        w2=w2-dw2*learning_rate
        b2=b2-db2*learning_rate
    return w1,b1,w2,b2

def predict():
    train()
    z1 = np.dot(w1,test_x) + b1
    a1 = np.tanh(z1)
    z2 = np.dot(w2, a1) + b2
    a2 = sigmoid(z2)
    return test_y