手写BP神经网络

---

一、BP神经网络的前向传播和梯度更新

二、程序设计

1、准备工作

import numpy as np
from sklearn import datasets
from sklearn.model_selection import train_test_split
learning_rate = 0.01

#输出转化为独热编码(one-hot)的形式
def convert_to_one_hot(y, C):  
    return np.eye(C)[y.reshape(-1)]
def sigmoid(x):
    return 1/(1+np.exp(-x))

#数据处理
#输入数据每一列是一个样本，标签使用独热编码
iris = datasets.load_iris()
train_data, test_data, train_label, test_label = train_test_split(iris.data, iris.target, test_size=0.2,random_state=0)
train_label = convert_to_one_hot(train_label,3)
test_label = convert_to_one_hot(test_label,3)
print("X_train="+str(train_data.shape))
print("X_test="+str(test_data.shape))
print("Y_train="+str(train_label.shape))
print("Y_test"+str(test_label.shape))

def compute_cost(A2,Y):
    m = A2.shape[1]
    cost = np.sum(np.power(A2-Y,2))/(2*m)
    cost = float(np.squeeze(cost))
    return cost

2、 初始化参数

def initialize_parameters(n_x,n_h,n_y):
    np.random.seed(1)
    W1 = np.random.randn(n_x,n_h)*0.01
    b1 = np.zeros(shape=(1,n_h))
    
    W2 = np.random.randn(n_h,n_y)*0.01
    b2 = np.zeros(shape=(1,n_y))
    parameters = {
        "W1":W1, "W2":W2, "b1":b1, "b2":b2
    }
    return parameters

3、前向传播

def forward_propagation(X,parameters):
    W1 = parameters["W1"]
    W2 = parameters["W2"]
    b1 = parameters["b1"]
    b2 = parameters["b2"]
    
    Z1 = np.dot(X,W1) + b1
    A1 = np.tanh(Z1)
    Z2 = np.dot(A1,W2) + b2
    A2 = sigmoid(Z2)
    cache = {
        "Z1":Z1, "Z2":Z2, "A1":A1, "A2":A2
    }
    return (A2,cache)

4、反向传播

def backward_propagation(parameters,cache,X,Y):
    W1 = parameters["W1"]
    W2 = parameters["W2"]
    A1 = cache["A1"]
    A2 = cache["A2"]
    
    W2_delta = (Y-A2)*A2*(1-A2)
    dW2 = np.dot(A1.T, W2_delta )
    W1_delta = W2_delta.dot(W2.T)*(1-np.power(A1,2))
    dW1 = np.dot(X.T,W1_delta)
    
    grads = {
        "W2_delta":W2_delta, "dW2":dW2, "W1_delta":W1_delta, "dW1":dW1
    }
    return grads

5、更新权重值

def update_parameters(parameters, grads, learning_rate):
    W1 = parameters["W1"]
    W2 = parameters["W2"]
    b1 = parameters["b1"]
    b2 = parameters["b2"]
    dW1 = grads["dW1"]
    dW2 = grads["dW2"]
    W1_delta = grads["W1_delta"]
    W2_delta = grads["W2_delta"]
    
    b1 = b1 + W1_delta*learning_rate
    b2 = b2 + W2_delta*learning_rate
    W1 = W1 + dW1*learning_rate
    W2 = W2 + dW2*learning_rate
    
    parameters = {
        "W1":W1, "W2":W2, "b1":b1, "b2":b2
    }
    return parameters

6、训练模型

parameters = initialize_parameters(4,40,3)
AA2=train_label
Y=train_label
cost = 0
for i in range(1000):
    A2,cache = forward_propagation(train_data,parameters)
    cost = compute_cost(A2,train_label)
    grads = backward_propagation(parameters,cache,train_data,train_label)
    parameters = update_parameters(parameters,grads,learning_rate)
    AA2=A2
    cost=cost
    print("cost in "+str(i)+"th loop is :"+str(cost))

7、测试

AA2 = np.around(AA2,0)
accuracy = 1 - (np.sum(np.power(AA2-train_label,2))/2)/train_data.shape[0]
print("your training accuracy is:"+str(accuracy))
print("your training cost is:"+str(cost))