机器学习是AI人工智能的重要组成部分。机器学习不仅包含了KNN、SVM、决策树等模型与方法,更有近年来越来越潮流的DP(深度学习)。这回讲一讲如何利用python构建KNN模型源码并实现识别手写体数字。
目录
1.导入第三方函数库
import numpy as np
import pandas as pd
import matplotlib.pylab as plt
%matplotlib inline
import os
2.读取训练数据
# 获取数据文件
fileList = os.listdir('C:/Users/15405/Desktop/digits/digits/trainingDigits/')
# 定义数据标签列表
trainingIndex = []
# 添加数据标签
for filename in fileList:
trainingIndex.append(int(filename.split('_')[0]))
# 定义矩阵数据格式
trainingData = np.zeros((len(trainingIndex),32*32))
# 获取矩阵数据
index = 0
for filename in fileList:
with open('C:/Users/15405/Desktop/digits/digits/trainingDigits/%s'%filename, 'rb') as f:
# 定义一个空矩阵
vect = np.zeros((1,1024))
# 循环32行
for i in range(32):
# 读取每一行数据
line = f.readline()
# 遍历每行数据索引 line[j] 即为数据
for j in range(32):
vect[0,32*i+j] = int(line[j])
trainingData[index,:] = vect
index+=1
3.读取测试数据
fileList2 = os.listdir('C:/Users/15405/Desktop/digits/digits/testDigits/')
# 定义数据标签列表
testIndex = []# 获取数据标签
for filename2 in fileList2:
testIndex.append(int(filename2.split('_')[0]))
# 定义矩阵数据格式
testData = np.zeros((len(testIndex),1024))
# 获取矩阵数据
index = 0
for filename2 in fileList2:
with open('C:/Users/15405/Desktop/digits/digits/testDigits/%s'%filename2, 'rb') as f:
# 定义一个空矩阵
vect = np.zeros((1,1024))
# 循环32行
for i in range(32):
# 读取每一行数据
line = f.readline()
# 遍历每行数据索引 line[j] 即为数据
for j in range(32):
vect[0,32*i+j] = int(line[j])
testData[index,:] = vect
index+=1
4.建立模型
def KNN(n_neighbors):
from sklearn.neighbors import KNeighborsClassifier
knn = KNeighborsClassifier(n_neighbors=n_neighbors)#超参数N
#利用训练数据拟合模型
knn.fit(trainingData,trainingIndex)
# 预测数据
predict_data = knn.predict(testData)
print("超参数n="+str(n_neighbors)+"时,模型的准确率:"+str(knn.score(testData,testIndex)))
return knn.score(testData,testIndex)
5.绘制超参数N与模型准确率的关系曲线
correct=[]
for i in range(1,10):
correct.append(KNN(i))
#绘制平滑曲线
import matplotlib.pyplot as plt
from scipy.interpolate import make_interp_spline
xnew = np.linspace(1,10,300) #将自变量划分为300段
power_smooth = make_interp_spline(range(1,10),correct)(xnew)
plt.plot(xnew,power_smooth)
plt.show()