头歌平台-人工智能导论实验（神经网络）

BP神经网络

import numpy as np

def loaddataset(filename):
    fp = open(filename)

    #存放数据
    dataset = []

    #存放标签
    labelset = []
    for i in fp.readlines():
        a = i.strip().split()

        #每个数据行的最后一个是标签数据
        dataset.append([float(j) for j in a[:len(a)-1]])
        labelset.append(int(float(a[-1])))
    return dataset, labelset


#x为输入层神经元个数，y为隐层神经元个数，z输出层神经元个数
def parameter_initialization(x, y, z):

    #隐层阈值
    value1 = np.random.randint(-5, 5, (1, y)).astype(np.float64)

    #输出层阈值
    value2 = np.random.randint(-5, 5, (1, z)).astype(np.float64)

    #输入层与隐层的连接权重
    weight1 = np.random.randint(-5, 5, (x, y)).astype(np.float64)

    #隐层与输出层的连接权重
    weight2 = np.random.randint(-5, 5, (y, z)).astype(np.float64)

    return weight1, weight2, value1, value2

#返回sigmoid函数值
def sigmoid(z):
#********** Begin **********#
    return 1 / (1 + np.exp(-z))
#********** End **********#

# '''
# weight1:输入层与隐层的连接权重
# weight2:隐层与输出层的连接权重
# value1:隐层阈值
# value2:输出层阈值
# '''
def trainning(dataset, labelset, weight1, weight2, value1, value2):
    #dataset：数据集 labelset：标签数据
    #x为步长
    x = 0.01
    for i in range(len(dataset)):
        #输入数据
        inputset = np.mat(dataset[i]).astype(np.float64)
        #数据标签
        outputset = np.mat(labelset[i]).astype(np.float64)
        #隐层输入
        input1 = np.dot(inputset, weight1).astype(np.float64)
        #隐层输出
        #********** Begin **********#
        output2 = sigmoid(input1 - value1).astype(np.float64)
        #********** End **********#
        #输出层输入
        input2 = np.dot(output2, weight2).astype(np.float64)
        #输出层输出
        output3 = sigmoid(input2 - value2).astype(np.float64)

        #更新公式由矩阵运算表示
        a = np.multiply(output3, 1 - output3)
        g = np.multiply(a, outputset - output3)
        b = np.dot(g, np.transpose(weight2))
        c = np.multiply(output2, 1 - output2)
        e = np.multiply(b, c)

        value1_change = -x * e
        value2_change = -x * g
        weight1_change = x * np.dot(np.transpose(inputset), e)
        weight2_change = x * np.dot(np.transpose(output2), g)

        #更新连接权重、阈值参数value1、value2、weight1、weight2
        #********** Begin **********#
        value1 += value1_change
        value2 += value2_change
        weight1 += weight1_change
        weight2 += weight2_change
        #********** End **********#
    return weight1, weight2, value1, value2

def testing(dataset, labelset, weight1, weight2, value1, value2):
    #记录预测正确的个数
    rightcount = 0
    for i in range(len(dataset)):
        #计算每一个样例通过该神经网路后的预测值
        inputset = np.mat(dataset[i]).astype(np.float64)
        outputset = np.mat(labelset[i]).astype(np.float64)
        output2 = sigmoid(np.dot(inputset, weight1) - value1)
        output3 = sigmoid(np.dot(output2, weight2) - value2)

        #确定其预测标签：输出大于 0.5 置 flag 为 1，否则置 flag 为 0
        #********** Begin **********#
        if output3 > 0.5:
            flag = 1
        else:
            flag = 0
        #********** End **********#
        if labelset[i] == flag:
              rightcount += 1
    #返回正确率
    return rightcount / len(dataset)

Hopfield 神经网络

import numpy as np
import random
from random import randint
from matplotlib import pyplot as plt
#根据Hebb学习规则计算神经元之间的连接权值
def calcWeight(savedsample):
    N = len(savedsample[0])
    P = len(savedsample)
    mat = [0]*N
    returnMat = []
    for i in range(N):
        m = mat[:]
        returnMat.append(m)
    for i in range(N):
        for j in range(N):
            if i==j:
                continue
            sum = 0
            for u in range(P):
                sum += savedsample[u][i] * savedsample[u][j]
            returnMat[i][j] = sum/float(N)
    return returnMat
#根据神经元的输入计算神经元的输出（静态突触）
def calcXi(inMat , weighMat):
    #假设计算第t次循环后神经元的输出时，输入的参数inMat表示第t-1次循环后神经元的输出。即用上一次循环的输出做本次循环的输入。
    #weighMat权值矩阵
    returnMat = inMat
    choose = []
    for i in range(len(inMat)//5):
        #随机改变N/5个神经元的值，该参数可调，也可同时改变所有神经元的值
        choose.append(random.randint(0,len(inMat)-1))
    for i in choose:
        sum = 0
        #网络动态演变
        #********** Begin **********#
        for j in range(len(inMat)):
            sum += weighMat[i][j] * inMat[j]
        if sum>=0:
            returnMat[i] = 1
        else: returnMat[i] = -1
        #********** End **********#
    return returnMat
    
#加噪函数，在记忆样本的基础上增加30%的噪声：
def addnoise(mytest_data,n):
    #mytest_data：网络矩阵 n:节点数 
    #********** Begin **********#
    for x in range(n):
        for y in range(n):
            if random.randint(0, 10) > 7:
                mytest_data[x * n + y] = -mytest_data[x * n + y]
    #********** End **********#
    return mytest_data
    
#标准输出函数：data 为联想记忆后的矩阵，N 为方阵行数：
def regularout(data,N):
    for j in range(N):
        ch = ""
        for i in range(N):
        #矩阵元素值为 1 则输出“ X ”，否则输出“ ”
        #********** Begin **********#
            ch += " " if data[j*N+i] == -1 else "X"
        #********** Begin **********#
        print(ch)
    

第4关：卷积神经网络

import numpy as np

def my_conv(inputmatrix,kernel):
    output_size = (len(inputmatrix)-len(kernel)+1)
    res = np.zeros([output_size,output_size],np.float32)
    for i in range(len(res)):
        for j in range(len(res)):
            res[i][j] = compute_conv(inputmatrix,kernel,i,j)
    return res

#两个矩阵的卷积算法
def compute_conv(inputmatrix,kernel,i,j):
    #inputmatrix为输入矩阵，kernel为卷积核，i,j分别是每次卷积运算的首元素地址
    res = 0
    #********** Begin **********#
    for kk in range(3):
        for k in range(3):
            res +=inputmatrix[i+kk][j+k]*kernel[kk][k]  #这句是关键代码，实现了两个矩阵的点乘操作
    #********** End **********#
    return res