KNN近邻算法与手写识别

# 导入库：KNeighborsClassifier
import numpy as np
from sklearn.neighbors import KNeighborsClassifier
import matplotlib.pyplot as plt
from matplotlib.colors import ListedColormap
%matplotlib inline
# 导入sklearn自带数据集
from sklearn import datasets
import  pandas as  pd

导入文件路径

path=r"C:\Users\Tsinghua-yincheng\Desktop\SZday92\KNNTest"

读取npy文件

x_train=np.load(path+"\\"+"x_train.npy") #数据
x_train.shape 

x_train

y_train=np.load(path+"\\"+"y_train.npy") #数据
y_train.shape 

y_test=np.load(path+"\\"+"y_test.npy") #数据
y_test.shape 
x_test=np.load(path+"\\"+"x_test.npy") #数据
x_test.shape 

人类的动作有以下几种

label={1:"walking",
       2:"walking up",
       3:"walking down",
       4:"sitting",
       5:"standing",
       6:"laying"}

看看有几种结果

pd.Series(y_train).unique()  #y的结果

y_train
x_train[0]
x_train[0].shape

knn=KNeighborsClassifier(100) #计算周围100个点
knn.fit(x_train,y_train)#训练数据
knn.predict(x_test[::100])#预测以100为间隔的点0 100 200...
y_test[::100]
knn.score(x_test,y_test)
y_new=knn.predict(x_test) #预测的结果
(y_new == y_test).sum()
2611/len(y_test)

np.argwhere(y_train==6) #筛选行为

plt.subplot: plt.subplot(2,3,1)也可以简写plt.subplot(231)表示把显示界面分割成2*3的网格。其中，第一个参数是行数，第二个参数是列数，第三个参数表示图形的标号。

#  78,162,640,29,0,7260   分别对应6种行为抽样数据    
x=[78,152,640,29,0,7260] 
plt.figure(figsize=(12,9))
for i,r  in enumerate(x):
    plt.subplot(3,2,(i+1))
    #取出动作对应数据
    data=x_train[r]
    color=np.random.rand(3)
    plt.plot(data,c=color,label=label[i+1])
    plt.legend()
    plt.show()

ren.m是一个模型,可以发现数据才是关键，才能训练出好的模型

from sklearn.externals import joblib
joblib.dump(knn,
            r"C:\Users\Tsinghua-yincheng\Desktop\SZday92\ren.m")
mynewknn=joblib.load(
    r"C:\Users\Tsinghua-yincheng\Desktop\SZday92\ren.m")
type(mynewknn)
mynewknn.predict(x_test[::200]) #训练好的模型
y_test[::200]

手写训练

import  pandas as pd
from sklearn.decomposition import PCA  #特征处理
from  sklearn.neighbors import KNeighborsClassifier #分类器
import time

if  __name__ =="__main__":
    train_num=20000 #训练的数据
    test_num=30000 #测试的数据
    data=pd.read_csv("train.csv")#读取数据
    print(data)
    train_data=data.values[0:train_num,1:] #抓取训练数据
    train_label=data.values[0:train_num,0] #训练的结果
    test_data=data.values[train_num:test_num,1:] #测试的数据
    test_label = data.values[train_num:test_num, 0]#测试的结果

    start=time.time()
    #pca特征训练   whiten=True白化处理  n_components特征
    pca=PCA(n_components=0.7,whiten=True)
    train_x=pca.fit_transform(train_data) #数据预处理
    test_x=pca.fit_transform(test_data)
    knn=KNeighborsClassifier(n_neighbors=10)
    knn.fit(train_x,train_label) #训练
    pre=knn.predict(test_x) #预测
    acc=  (pre==test_label).sum()/len(test_label)
    end=time.time()
    print("准确率",acc,"time",end-start)

数字手写识别

# 导入库：KNeighborsClassifier
import numpy as np
from sklearn.neighbors import KNeighborsClassifier
import matplotlib.pyplot as plt
from matplotlib.colors import ListedColormap
%matplotlib inline
# 导入sklearn自带数据集
from sklearn import datasets
import  pandas as  pd

path=r"C:\Users\Tsinghua-yincheng\Desktop\SZday92\KNNTest\datanumber\0"
zero1=plt.imread(path+"\\0_497.bmp") #读取图片，转化数组
zero1

zero1.shape
zero1.reshape(-1)#压缩成一维

plt.imshow(zero1,cmap="gray")

x_train=[]
path=r"C:\Users\Tsinghua-yincheng\Desktop\SZday92\KNNTest\datanumber\0"
for i  in range(1,501):#读取500图片，成为一个数组，500元素
    x_train.append(plt.imread(path+"\\0_%d.bmp"%(i)))
X_train=[]
Y_train=[]
X_test=[]
Y_test=[]
path=r"xxxxxxxxx"
#读取数据，加入训练集合与测试集合
for i in range(10):#就是0-9这10个数字
	for j in range(1,501):#每个里面有的数据
		if j<480:#前面是训练
			X_train.append(plt.imread(path+"\\%d\\%d_%d.bmp"%(i,i,j)).reshape(-1))
			Y_train.append(i)
		else:
			X_test.append(plt.imread(path+"\\%d\\%d_%d.bmp"%(i,i,j)).reshape(-1))
			Y_test.append(i)	
X_train = np.array(X_train)
Y_train = np.array(Y_train)
X_test=np.array(X_test)
Y_test = np.array(Y_test)
X_train.shape
X_test.shape

n=np.random.randint(1,210)
num=X_test[n].reshape((28,28))
plt.imshow(num,cmap="gray")

knn = KNeighborsClassifier()#建立KNN对象，训练模型
knn.fit(X_train,Y_train)
Y_new = knn.predict(K_test)
Y_new[::10]#0 10 20 30...
Y_test[::10]
knn.score(X_test,Y_test)#识别率
Y_new[::10]
X_test[::10]

plt.figure(figsize=(12,15))
im_datas=X_test[::10]
im_target=Y_test[::10]
im_predict=Y_new[::10]
for  i  in range(20):
    plt.subplot(5,5,(i+1))
    plt.imshow(im_datas[i].reshape((28,28)))
    plt.title("predict:%d\n"%(im_predict[i])+"true:%d"%(im_target[i]))
    plt.imsave("xxxxxxxx",zero1) #数组保存为图片

KNN测试

import pandas as pd
import sklearn
from sklearn import datasets,linear_model
from sklearn.cross_validation import  train_test_split
from sklearn.linear_model import LinearRegression
from sklearn import metrics
from sklearn.model_selection import  cross_val_predict
import ML


path=r"C:\Users\Tsinghua-yincheng\Desktop\SZday92\KNNTest\dat\iris_"
x_train=pd.read_csv(path+"xtrain.csv",index_col=False)
y_train=pd.read_csv(path+"ytrain.csv",index_col=False)
x_test=pd.read_csv(path+"xtest.csv",index_col=False)
y_test=pd.read_csv(path+"ytest.csv",index_col=False)
#mx=ML.mx_knn(x_train.values,y_train.values) #创造KNN
mx=ML.mx_dtree(x_train.values,y_train.values)
y_new=mx.predict(x_test.values) #预测
print(y_new)
print(y_test.values.reshape(-1))
print(((y_new==y_test.values.reshape(-1)).sum())/len(y_test.values))