玩一下kmeans，调戏以下国足，顺便预测一下18世界杯冠军，18-7-15 23:00世界杯
物以类聚，人以群分，选亚洲15支球队
data = {"zhongguo":[50,50,50,40],
"riben":[28,9,29,12],
"hanguo":[17,15,27,26],
"yilang":[25,40,28,18],
"shate":[28,40,50,25],
"yilake":[50,50,40,40],
"kataer":[50,40,40,40],
"alianqiu":[50,40,50,40],
"wuzibiekesitan":[40,40,40,40],
"taiguo":[50,50,50,40],
"yuenan":[50,50,50,50],
"aman":[50,50,40,50],
"balin":[40,40,50,50],
"chaoxian":[40,32,50,50],
"yinni":[50,50,50,50]}
依次选2006年，2010年，2014年，2018年世界杯的数据作为聚类样本，打进世界杯的得分用排名衡量，预选赛小组未出线的给50，预选赛十强的给40，澳大利亚没统计，18年的排名是估计的，虽然11点是冠亚军决赛，理论上其他队伍排名已经定了，但是我不会，这样算，得分越多的越low。
k选3，初始中心选中国，日本，沙特，先计算每一条数据到三个中心点的欧氏距离，并将其归为最近点那一类，处理完所有数据后，计算每个类的中心点，更新聚类中心，重新以上步骤，知道聚类中心不再变化，代码：
import numpy as np
def name2indexf(names):
name2index = {}
for index, name in enumerate(names):
name2index[name] = index
return name2index
def cacul_eudist(vec1, vec2):
assert len(vec1) == len(vec2)
dist = np.linalg.norm(vec1 - vec2)
return dist
if __name__ == "__main__":
data = {"zhongguo":[50,50,50,40],
"riben":[28,9,29,12],
"hanguo":[17,15,27,26],
"yilang":[25,40,28,18],
"shate":[28,40,50,25],
"yilake":[50,50,40,40],
"kataer":[50,40,40,40],
"alianqiu":[50,40,50,40],
"wuzibiekesitan":[40,40,40,40],
"taiguo":[50,50,50,40],
"yuenan":[50,50,50,50],
"aman":[50,50,40,50],
"balin":[40,40,50,50],
"chaoxian":[40,32,50,50],
"yinni":[50,50,50,50]}
data_array = np.zeros(shape=(len(data.keys()),4))
name2index = name2indexf(data.keys())
for name in data.keys():
index = name2index[name]
data_array[index] = np.array(data[name])
k_center = np.array([data_array[0],data_array[1],data_array[4]])
k_center_with_near = {}
# ------cacul center and it near-------
epoch = 0
while True:
for index, item in enumerate(k_center):
k_center_with_near[index] = []
# ----------choice nearest for each data---------
for index in range(len(data.keys())):
data_item = data_array[index]
near = 0
dist_min = 100000
for i in range(len(k_center)):
dist = cacul_eudist(data_item, k_center[i])
if dist_min > dist:
dist_min = dist
near = i
k_center_with_near[near].append(index)
# ------recacul center---------------
end_tag = True
for center_near_index in k_center_with_near.keys():
contry_index = k_center_with_near[center_near_index]
center_near_data = []
for item in contry_index:
center_near_data.append(data_array[item])
center_near_data = np.array(center_near_data)
new_center = np.mean(center_near_data, axis=0)
if not (k_center[center_near_index] == new_center).all():
end_tag = False
k_center[center_near_index] = new_center
print("epoch:",epoch, "center:", k_center)
epoch += 1
test = data.keys()
if end_tag:
for item in k_center_with_near.keys():
print("classv{} include".format(item), [list(data.keys())[n] for n in (k_center_with_near[item])])
break
结果：
classv0 include ['zhongguo', 'yilake', 'kataer', 'alianqiu', 'wuzibiekesitan', 'taiguo', 'yuenan', 'aman', 'balin', 'chaoxian', 'yinni']
classv1 include ['riben', 'hanguo']
classv2 include ['yilang', 'shate']
这样算，中国队在亚洲只能算3流球队
预测克罗地亚冠军，虽然实力比法国弱一些，但是不要低估对冠军渴望的心