In [1]:
import numpy as np
In [2]:
def createData():
group=np.array([[1.,.9],[1.,1.],[.1,.2],[.0,.1]])
labels=['A','A','B','B']
return group,labels
In [3]:
group,labels=createData()
group
Out[3]:
In [4]:
import matplotlib.pyplot as plt
%matplotlib inline
In [5]:
group[:,0] #np.array可以这么取元素，list不可以
Out[5]:
In [6]:
group[:,1]
Out[6]:
In [7]:
plt.scatter(group[:,0],group[:,1])
Out[7]:
In [8]:
np.tile([1,2,3],3)
Out[8]:
In [9]:
np.tile([1,2,3],(3,1))
Out[9]:
In [10]:
diff=np.array([1,2,3])-np.array([1,1,1])
diff
Out[10]:
In [11]:
diff=np.array([[1,2,3],[2,3,4]])-np.array([[1,1,1],[1,1,1]])
diff
Out[11]:
In [12]:
diff
Out[12]:
In [13]:
diff**2
Out[13]:
In [38]:
sum(diff**2) # python自带的sum求得是同一列的数的和
Out[38]:
In [15]:
np.sum(diff**2,axis=1) # np.sum
Out[15]:
In [16]:
len(group)
Out[16]:
In [17]:
# 使用np.argsort，排序返回的是从小到大排序后的索引
def knnclassify(newinput,datas,lables,k):
#distances=[]
#input_x,input_y=newinput
#for x,y in datas: 这样只能计算二维的
#distances.append((x-input_x)**2+(y-input_y)**2)
diff=np.tile(newinput,(len(datas),1))-datas
distances=np.sum(diff**2,axis=1)
sort_distances=np.argsort(distances)
print distances
print sort_distances
classCount={}
for i in range(k):
votelabel=lables[sort_distances[i]]
classCount[votelabel]=classCount.get(votelabel,0)+1
print classCount
maxCount=0
for key,value in classCount.items():
if value>maxCount:
maxCount=value
maxIndex=key
#print maxIndex
return maxIndex
In [18]:
group
Out[18]:
In [19]:
knnclassify([0,0],group,labels,2)
Out[19]:
In [20]:
knnclassify([.1,.3],group,labels,3)
Out[20]:
In [21]:
knnclassify([1.2,1.],group,labels,3)
Out[21]:
- 准备数据，原数据太大，这里切前3000行进行测试。
In [22]:
import pandas as pd
datas=pd.read_csv('train.csv')
datas.shape
Out[22]:
In [23]:
digitdatas=datas[:3000]
digitdatas.shape
Out[23]:
In [24]:
digitdatas.to_csv('digit_datas.csv',index=False)
In [25]:
all_datas=pd.read_csv('digit_datas.csv')
print all_datas.shape
all_datas.head()
Out[25]:
In [26]:
y=all_datas['label'].values
x=all_datas.drop(['label'],axis=1).values
print y[:10]
x[:10]
Out[26]:
In [27]:
set(x[0])
Out[27]:
In [28]:
# 大于1的像素都转为1
x[x>0]=1
In [29]:
set(x[0])
Out[29]:
In [30]:
from sklearn.cross_validation import train_test_split
In [31]:
x_train,x_test,y_train,y_test=train_test_split(x,y,test_size=0.3)
x_train.shape,len(y_train),x_test.shape,len(y_test)
Out[31]:
In [32]:
x_train[:1]
Out[32]:
In [33]:
y_train[:1]
Out[33]:
In [34]:
y_train[1750]
Out[34]:
In [35]:
# 使用np.argsort，排序返回的是从小到大排序后的索引
def knnClassify(newinput,datas,lables,k):
#distances=[]
#input_x,input_y=newinput
#for x,y in datas: 这样只能计算二维的
#distances.append((x-input_x)**2+(y-input_y)**2)
diff=np.tile(newinput,(len(datas),1))-datas
distances=np.sum(diff**2,axis=1)
sort_distances=np.argsort(distances)
#print distances
#print sort_distances
classCount={}
for i in range(k):
#print i,sort_distances[i]
#print lables[sort_distances[i]]
votelabel=lables[sort_distances[i]]
classCount[votelabel]=classCount.get(votelabel,0)+1
#print classCount
maxCount=0
for key,value in classCount.items():
if value>maxCount:
maxCount=value
maxIndex=key
#print maxIndex
return maxIndex
In [36]:
matchCount=0
for i in range(len(x_test)):
predict=knnClassify(x_test[i],x_train,y_train,3)
#print predict,y_test[i]
if predict==y_test[i]:
matchCount+=1
else:
print predict,y_test[i]
accuracy=float(matchCount)/len(x_test)
print accuracy