机器学习基石笔记：Homework #4 Regularization&Validation相关习题

原文地址：https://www.jianshu.com/p/3f7d4aa6a7cf

问题描述

程序实现

# coding: utf-8

import numpy as np
import math
import matplotlib.pyplot as plt

def sign(x):
    if(x>=0):
        return 1
    else:
        return -1

def read_data(dataFile):
    with open(dataFile,'r') as f:
        lines=f.readlines()
        data_list=[]
        for line in lines:
            line=line.strip().split()
            data_list.append([1.0] + [float(l) for l in line])
        dataArray=np.array(data_list)
        num_data=dataArray.shape[0]
        num_dim=dataArray.shape[1]-1
        dataX=dataArray[:,:-1].reshape((num_data,num_dim))
        dataY=dataArray[:,-1].reshape((num_data,1))
        return dataX,dataY

def w_reg(dataX,dataY,namuta):
    num_dim=dataX.shape[1]
    dataX_T=np.transpose(dataX)
    tmp=np.dot(np.linalg.inv(np.dot(dataX_T,dataX)+namuta*np.eye(num_dim)),dataX_T)
    return np.dot(tmp,dataY)

def pred(wREG,dataX):
    pred=np.dot(dataX,wREG)
    num_data=dataX.shape[0]
    for i in range(num_data):
        pred[i][0]=sign(pred[i][0])
    return pred

def zero_one_cost(pred,dataY):
    return np.sum(pred!=dataY)/dataY.shape[0]

if __name__=="__main__":
    # train
    dataX,dataY=read_data("hw4_train.dat")
    print("\n13")
    wREG=w_reg(dataX,dataY,namuta=10)
    Ein=zero_one_cost(pred(wREG,dataX),dataY)
    print("the Ein on the train set: ",Ein)
    # test
    testX,testY=read_data("hw4_test.dat")
    Eout=zero_one_cost(pred(wREG,testX),testY)
    print("the Eout on the test set: ",Eout)

    l=[2,1,0,-1,-2,-3,-4,-5,-6,-7,-8,-9,-10]

    print("\n14")
    Ein_list=[]
    Eout_list=[]
    for i in l:
        namuta=math.pow(10,i)
        wREG=w_reg(dataX,dataY,namuta)
        Ein_list.append(zero_one_cost(pred(wREG,dataX),dataY))
        Eout_list.append(zero_one_cost(pred(wREG,testX),testY))
    id_in=Ein_list.index(min(Ein_list))
    plt.figure()
    plt.plot(np.power(np.full(shape=(len(l),),fill_value=10,dtype=np.int32),l),Ein_list)
    plt.xlabel("namuta")
    plt.xlim((math.pow(10,l[0]),math.pow(10,l[-1])))
    plt.ylabel("Ein")
    plt.savefig("14.png")
    print("the namuta with the minimun Ein: ",math.pow(10,l[id_in]))
    print("the Eout on such namuta: ", Eout_list[id_in])

    print("\n15")
    id_out = Eout_list.index(min(Eout_list))
    plt.figure()
    plt.plot(np.power(np.full(shape=(len(l),),fill_value=10,dtype=np.int32),l),Eout_list)
    plt.xlabel("namuta")
    plt.xlim((math.pow(10,l[0]),math.pow(10,l[-1])))
    plt.ylabel("Eout")
    plt.savefig("15.png")
    print("the namuta with the minimun Eout: ", math.pow(10, l[id_out]))

    trainX=dataX[:120]
    trainY=dataY[:120]
    validX=dataX[120:]
    validY=dataY[120:]

    # validation
    print("\n16")
    Ein_list.clear()
    Eout_list.clear()
    Eval_list=[]
    for i in l:
        namuta=math.pow(10,i)
        wREG=w_reg(trainX,trainY,namuta)
        Ein_list.append(zero_one_cost(pred(wREG,trainX),trainY))
        Eout_list.append(zero_one_cost(pred(wREG,testX),testY))
        Eval_list.append(zero_one_cost(pred(wREG,validX),validY))
    id_in=Ein_list.index(min(Ein_list))
    plt.figure()
    plt.plot(np.power(np.full(shape=(len(l),),fill_value=10,dtype=np.int32),l),Ein_list)
    plt.xlabel("namuta")
    plt.xlim((math.pow(10,l[0]),math.pow(10,l[-1])))
    plt.ylabel("Ein")
    plt.savefig("16.png")
    print("the namuta with the minimun Ein: ",math.pow(10,l[id_in]))
    print("the Eout on such namuta: ", Eout_list[id_in])

    print("\n17")
    id_val=Eval_list.index(min(Eval_list))
    plt.figure()
    plt.plot(np.power(np.full(shape=(len(l),),fill_value=10,dtype=np.int32),l),Eval_list)
    plt.xlabel("namuta")
    plt.xlim((math.pow(10,l[0]),math.pow(10,l[-1])))
    plt.ylabel("Eval")
    plt.savefig("17.png")
    print("the namuta with the minimun Eval: ",math.pow(10,l[id_val]))
    print("the Eout on such namuta: ", Eout_list[id_val])

    print("\n18")
    wREG=w_reg(dataX,dataY,namuta=math.pow(10,l[id_val]))
    Ein=zero_one_cost(pred(wREG,dataX),dataY)
    Eout = zero_one_cost(pred(wREG, testX), testY)
    print("Ein: ",Ein)
    print("Eout: ",Eout)

    # 5-fold cross validation
    print("\n19")
    Eval_list.clear()
    splX=np.split(dataX,5,axis=0)
    splY=np.split(dataY,5,axis=0)
    for j in l:
        Eval = 0
        namuta=math.pow(10,j)
        for i in range(5):
            li=[a for a in range(5)]
            li.pop(i)
            trainX=np.concatenate([splX[k] for k in li],axis=0)
            trainY=np.concatenate([splY[k] for k in li],axis=0)
            wREG=w_reg(trainX,trainY,namuta)
            Eval+=zero_one_cost(pred(wREG,splX[i]),splY[i])/5
        Eval_list.append(Eval)
    id_val=Eval_list.index(min(Eval_list))
    plt.figure()
    plt.plot(np.power(np.full(shape=(len(l),),fill_value=10,dtype=np.int32),l),Eval_list)
    plt.xlabel("namuta")
    plt.xlim((math.pow(10,l[0]),math.pow(10,l[-1])))
    plt.ylabel("Ecv")
    plt.savefig("19.png")
    print("the namuta with the minimun Ecv: ",math.pow(10,l[id_val]))

    print("\n20")
    wREG=w_reg(dataX,dataY,namuta=math.pow(10,l[id_val]))
    Ein=zero_one_cost(pred(wREG,dataX),dataY)
    Eout = zero_one_cost(pred(wREG, testX), testY)
    print("Ein: ",Ein)
    print("Eout: ",Eout)

运行结果

13

14

15

16

17

18

19

20