K-means基于的简单实现

简单实现k-means算法和non standard k-means方法。基本思想就在于找到一组center以后,对新的数据进行聚类,直到误差不再变化。

def dis(a,b):
    return ((a[0]-b[0])**2+(a[1]-b[1])**2)**(0.5)
def Error(x,y):
        assert(len(x)==len(y))
        z = (0,0)
        m = mean(dis(a,b) for a,b in zip(x,y))
        s = mean(dis(a,z)+dis(b,z) for a,b in zip(x,y))/2
        return m/s
def k_means(x,k=2,initial=None):
    c =  initial if initial else [x[i] for i in range(k)]
    old_center =  [(0,0)]* len(initial) if initial else k
    while Error(old_center,c)>0.001:
        cluster = defaultdict(list)
        old_center = c.copy()
        for point in x:
            d,index = min((dis(center,point),i) for i,center in enumerate(c)) 
            cluster[index].append((point))
        for index,points in cluster.items():
            c[index] = mean(p[0] for p in points),mean(p[1] for p in points)
    return c
def bin_split(x,M):
    def split(centers):
        new_center = []
        for center in centers:
            new_center.append((center[0]*(1+e),center[1]*(1+e)))
            new_center.append((center[0]*(1-e),center[1]*(1-e)))
        return new_center

    m,e = 1,0.05
    centers = [(mean(a[0] for a in x),mean(a[1] for a in x))]

    while m<M:
        centers = split(centers)
        centers = k_means(x,initial = centers)
        m<<=1
    return centers