import numpy as np
import matplotlib.pyplot as plt
from sklearn import datasets

X,z = datasets.make_blobs(n_samples=100,centers=4,cluster_std=0.8,random_state=0)
plt.axes(aspect=1).scatter(X[:,0],X[:,1],c=z,s=30,edgecolor='k',cmap='rainbow')
plt.show()

from sklearn.neighbors import KNeighborsClassifier as Knn
knn = Knn()
knn.fit(X,z)

nmesh = 200
mx,my = np.meshgrid(np.linspace(X[:,0].min(),X[:,0].max(),nmesh),np.linspace(X[:,1].min(),X[:,1].max(),nmesh))
mX = np.stack([mx.ravel(),my.ravel()],1)
mz = knn.predict(mX).reshape(nmesh,nmesh)
plt.axes(xlim=[X[:,0].min(),X[:,0].max()],ylim=[X[:,1].min(),X[:,1].max()],aspect=1)
plt.contourf(mx,my,mz,alpha=0.1,cmap='rainbow')
plt.contour(mx,my,mz,colors='#222222')
plt.scatter(X[:,0],X[:,1],c=z,edgecolor='k',cmap='rainbow')
plt.show()

knn = Knn(n_neighbors=1,p=1)
# หรือ knn = Knn(1,p=1) ก็ได้ เพราะ n_neighbors เป็นคีย์เวิร์ดลำดับแรกอยู่แล้ว

import time
X,z = datasets.make_blobs(n_samples=10000,n_features=10,centers=4,random_state=0)
for al in ['ball_tree','kd_tree','brute','auto']:
    t1 = time.time()
    knn = Knn(algorithm=al)
    knn.fit(X,z)
    knn.predict(X)
    print(u'%s: %.3f วินาที'%(al,time.time()-t1))

ball_tree: 0.670 วินาที
kd_tree: 0.585 วินาที
brute: 3.151 วินาที
auto: 0.585 วินาที

X,z = datasets.make_blobs(n_samples=20000,n_features=20,random_state=0)
for j in [1,2,3]:
    t1 = time.time()
    Knn(n_jobs=j).fit(X,z).predict(X)
    print(u'n_jobs=%s: %.3f วินาที'%(j,time.time()-t1))

n_jobs=1: 8.393 วินาที
n_jobs=2: 4.825 วินาที
n_jobs=3: 4.235 วินาที

X,z = datasets.make_blobs(n_samples=8,centers=2,random_state=5)
knn = Knn(n_neighbors=3)
knn.fit(X,z)
plt.axes(aspect=1)
plt.scatter(X[:,0],X[:,1],c=z,edgecolor='k',cmap='summer')
plt.show()
k = knn.kneighbors(X)
print(k[0])
print(k[1])
for i in range(8):
    print(', '.join(['%d > %.2f'%(k[1][i][j],k[0][i][j]) for j in range(3)]))

[[ 0.          1.12763922  1.91394086]
 [ 0.          0.68420112  1.16141094]
 [ 0.          1.43561929  1.9628876 ]
 [ 0.          0.77421595  1.16141094]
 [ 0.          2.24451596  2.95862282]
 [ 0.          1.12763922  1.35891489]
 [ 0.          0.77421595  1.43561929]
 [ 0.          0.68420112  1.35891489]]
[[0 5 6]
 [1 7 3]
 [2 6 3]
 [3 6 1]
 [4 0 6]
 [5 0 7]
 [6 3 2]
 [7 1 5]]
0 > 0.00, 5 > 1.13, 6 > 1.91
1 > 0.00, 7 > 0.68, 3 > 1.16
2 > 0.00, 6 > 1.44, 3 > 1.96
3 > 0.00, 6 > 0.77, 1 > 1.16
4 > 0.00, 0 > 2.24, 6 > 2.96
5 > 0.00, 0 > 1.13, 7 > 1.36
6 > 0.00, 3 > 0.77, 2 > 1.44
7 > 0.00, 1 > 0.68, 5 > 1.36

print(knn.kneighbors(X,n_neighbors=8,return_distance=0))

[[0 5 6 1 3 7 4 2]
 [1 7 3 5 6 0 2 4]
 [2 6 3 1 0 4 7 5]
 [3 6 1 7 2 0 5 4]
 [4 0 6 5 2 3 1 7]
 [5 0 7 1 3 6 4 2]
 [6 3 2 1 0 5 7 4]
 [7 1 5 3 0 6 2 4]]

print(type(knn.kneighbors_graph(X))) # ให้แสดงชนิด
print(knn.kneighbors_graph(X).toarray()) # แปลงเป็นอาเรย์ธรรมดา

<class 'scipy.sparse.csr.csr_matrix'>
[[ 1.  0.  0.  0.  0.  1.  1.  0.]
 [ 0.  1.  0.  1.  0.  0.  0.  1.]
 [ 0.  0.  1.  1.  0.  0.  1.  0.]
 [ 0.  1.  0.  1.  0.  0.  1.  0.]
 [ 1.  0.  0.  0.  1.  0.  1.  0.]
 [ 1.  0.  0.  0.  0.  1.  0.  1.]
 [ 0.  0.  1.  1.  0.  0.  1.  0.]
 [ 0.  1.  0.  0.  0.  1.  0.  1.]]

print(knn.kneighbors_graph(X,mode='distance',n_neighbors=8).toarray())

[[ 0.          2.00350493  3.21293728  2.07395977  2.24451596  1.12763922
   1.91394086  2.23635633]
 [ 2.00350493  0.          3.12214116  1.16141094  4.05546124  1.46770732
   1.80934825  0.68420112]
 [ 3.21293728  3.12214116  0.          1.9628876   3.45228072  3.83210647
   1.43561929  3.80417203]
 [ 2.07395977  1.16141094  1.9628876   0.          3.58705789  2.16656867
   0.77421595  1.84527381]
 [ 2.24451596  4.05546124  3.45228072  3.58705789  0.          3.35721827
   2.95862282  4.43043208]
 [ 1.12763922  1.46770732  3.83210647  2.16656867  3.35721827  0.
   2.40100554  1.35891489]
 [ 1.91394086  1.80934825  1.43561929  0.77421595  2.95862282  2.40100554
   0.          2.46612122]
 [ 2.23635633  0.68420112  3.80417203  1.84527381  4.43043208  1.35891489
   2.46612122  0.        ]]

X,z = datasets.make_blobs(n_samples=200,centers=2,cluster_std=2.5,random_state=12)
nmesh = 200
mx,my = np.meshgrid(np.linspace(X[:,0].min(),X[:,0].max(),nmesh),np.linspace(X[:,1].min(),X[:,1].max(),nmesh))
mX = np.stack([mx.ravel(),my.ravel()],1)
for i in [0,1]:
    n = 3+27*i
    knn = Knn(n)
    knn.fit(X,z)
    k = knn.kneighbors(X)
    for j in [0,1]:
        if(j==1):
            mz = knn.predict_proba(mX)[:,1].reshape(nmesh,nmesh)
        else:
            mz = knn.predict(mX).reshape(nmesh,nmesh)
        plt.subplot(221+i+2*j,xlim=[X[:,0].min(),X[:,0].max()],ylim=[X[:,1].min(),X[:,1].max()],aspect=1)
        plt.scatter(X[:,0],X[:,1],10,c=z,edgecolor='k',cmap='winter')
        plt.contourf(mx,my,mz,100,cmap='winter',zorder=0)
        if(j==0):
            plt.title('n=%d'%n)
        else:
            plt.ylabel('proba')
plt.show()