使用层次聚类解决了这个问题。设置默认半径运行相同的数据点。降到了6分。
#mean shift clustering.
#this lets the program decide number of groups involved in the given dataset
import numpy as np
import matplotlib.pyplot as plt
import random as r
#setting all the points as centroids
centroids = {}
def auto_cluster(radius,data):
global centroids
for i in range(len(data)):
centroids[i] = data[i]
while True:
new_centroids=[]
#checking all the points whether it is in radius and assign to to that centroid
for j in centroids:
in_radius=[]
centroid=centroids[j]
for point in data:
if np.linalg.norm(point-centroid)<radius:
in_radius.append(point)
#finding mean
new_centroid=np.average(in_radius,axis=0)
new_centroids.append(tuple(new_centroid))
#collect all the final centroids for each grp
uniques=sorted(list(set(new_centroids)))
prev_centroids=dict(centroids)
centroids={}
#fil with new centroids
for i in range(len(uniques)):
centroids[i]=np.array(uniques[i])
opt=True
#chech whether the centroid is optimized
for i in centroids:
if not np.array_equal(centroids[i],prev_centroids[i]):
opt=False
if not opt:
break
if opt:break
return centroids
if __name__=="__main__":
data = [[61, 148], [61, 149], [61, 150], [62, 147], [62, 148], [62, 149], [62, 150], [63, 147], [63, 148],
[63, 149], [63, 150], [64, 147], [64, 148], [64, 149], [64, 150], [65, 147], [65, 148], [65, 149],
[65, 150], [149, 436], [149, 437], [149, 438], [150, 366], [150, 367], [150, 368], [150, 436], [150, 437],
[150, 438], [150, 439], [151, 366], [151, 367], [151, 368], [151, 436], [151, 437], [151, 438], [151, 439],
[152, 366], [152, 367], [152, 368], [152, 436], [152, 437], [152, 438], [152, 439], [175, 147], [175, 148],
[175, 149], [175, 150], [175, 264], [175, 265], [175, 266], [175, 267], [176, 147], [176, 148], [176, 149],
[176, 150], [176, 264], [176, 265], [176, 266], [176, 267], [177, 147], [177, 148], [177, 149], [177, 150],
[177, 264], [177, 265], [177, 266], [177, 267], [178, 147], [178, 148], [178, 149], [178, 264], [178, 265],
[178, 266], [230, 366], [230, 367], [230, 368], [230, 369], [231, 366], [231, 367], [231, 368], [231, 369],
[232, 366], [232, 367], [232, 368], [232, 369], [233, 366], [233, 367], [233, 368]]
data = np.array(data)
centroids = {}
cent = auto_cluster(radius=5,data=data)
print centroids
print(len(cent)) # no. of centroids
# plots
[plt.scatter(x[0], x[1], s=50, c='g') for x in data]
for c in cent:
plt.scatter(cent[c][0], cent[c][1], s=200, marker='*')
plt.show()
