3层神经网络

#! /usr/bin/env python
#coding=utf-8
from sklearn.datasets import load_digits#数据集
from sklearn.preprocessing import LabelBinarizer#标签二值化
from sklearn.model_selection import train_test_split#数据集分割
import numpy as np
import pylab as pl#数据可视化
def sigmoid(x):#激活函数
    return 1/(1+np.exp(-x))
def dsigmoid(x):#sigmoid的导数
    return 1/(1+np.exp(-x))*(1-x)
class NeuralNetwork:
    def __init__(self,layers):#这里是三层网络，列表[64,100,10]表示输入，隐藏，输出层的单元个数
        #初始化权值，范围1~-1
        self.V=np.random.random((layers[0]+1,layers[1]))*2-1#隐藏层权值(65,100)，之所以是65，因为有偏置W0
        self.W=np.random.random((layers[1],layers[2]))*2-1#(100,10)
    def train(self,X,y,lr=0.1,epochs=10000):
 #lr为学习率，epochs为迭代的次数
        #为数据集添加偏置
        temp=np.ones([X.shape[0],X.shape[1]+1])
        temp[:,0:-1]=X
        X=temp#这里最后一列为偏置
        #进行权值训练更新
        for n in range(epochs+1):
            i=np.random.randint(X.shape[0])#随机选取一行数据(一个样本)进行更新
            x=X[i]
            x=np.atleast_2d(x)#转为二维数据
            L1=sigmoid(np.dot(x,self.V))#隐层输出(1,100)
            L2=sigmoid(np.dot(L1,self.W))#输出层输出(1,10)
            #delta
            L2_delta=(y[i]-L2)*dsigmoid(L2)#(1,10)
            L1_delta=L2_delta.dot(self.W.T)*dsigmoid(L1)#(1,100)，这里是数组的乘法，对应元素相乘
            #更新
            self.W+=lr*L1.T.dot(L2_delta)#(100,10)
            self.V+=lr*x.T.dot(L1_delta)#
            #每训练1000次预测准确率
            if n%1000==0: 
                predictions=[]
                for j in range(X_test.shape[0]):
                    out=self.predict(X_test[j])#用验证集去测试
                    predictions.append(np.argmax(out))#返回预测结果
                accuracy=np.mean(np.equal(predictions,y_test))#求平均值
                print('epoch:',n,'accuracy:',accuracy)
    def predict(self,x):
        #添加转置,这里是一维的
        temp=np.ones(x.shape[0]+1)
        temp[0:-1]=x
        x=temp
        x=np.atleast_2d(x)
        L1=sigmoid(np.dot(x,self.V))#隐层输出
        L2=sigmoid(np.dot(L1,self.W))#输出层输出
        return L2
digits=load_digits()#载入数据
X=digits.data#数据
y=digits.target#标签
#print y[0:10]
#数据归一化,一般是x=(x-x.min)/x.max-x.min
#X-=X.min()
#X/=X.max()-X.min()
X = (X-X.min())/(X.max()-X.min())
#创建神经网络
nm=NeuralNetwork([64,100,10])
X_train,X_test,y_train,y_test=train_test_split(X,y)#默认分割：3:1
#标签二值化
labels_train=LabelBinarizer().fit_transform(y_train)
#print labels_train[0:10]
labels_test=LabelBinarizer().fit_transform(y_test)
print ('start')
nm.train(X_train,labels_train,epochs=20000)
print ('end')