add kmeans/readme.md

KMeans/Readme.md

@@ -0,0 +1,60 @@
+###KMeans算法，二分KMeans算法
+
+- 封装成class
+- 参考sklearn.cluster.KMeans的接口设计
+
+
+###依赖库
+- Numpy
+- Matplotlib (可视化结果)
+
+
+###使用方法
+
+```
+import cPickle
+from kmeans import KMeans,biKMeans
+X,y = cPickle.load(open('data.pkl','r'))
+clf = KMeans(n_clusters=10,initCent='random',max_iter=200)
+#clf = KMeans(n_clusters=10,initCent=X[0:10],max_iter=200)
+#clf = biKMeans(n_clusters=10)
+clf.fit(X)
+clf.predict(X)
+
+```
+
+###可视化
+
+```
+
+import numpy as np
+import matplotlib.pyplot as plt
+from kmeans import biKMeans
+n_clusters = 10
+clf = biKMeans(n_clusters)
+clf.fit(X)
+cents = clf.centroids
+labels = clf.labels
+sse = clf.sse
+#画出聚类结果，每一类用一种颜色
+colors = ['b','g','r','k','c','m','y','#e24fff','#524C90','#845868']
+for i in range(n_clusters):
+	index = np.nonzero(labels==i)[0]
+	x0 = X[index,0]
+	x1 = X[index,1]
+	y_i = y[index]
+	for j in range(len(x0)):
+		plt.text(x0[j],x1[j],str(int(y_i[j])),color=colors[i],\
+				fontdict={'weight': 'bold', 'size': 9})
+	plt.scatter(cents[i,0],cents[i,1],marker='x',color=colors[i],linewidths=12)
+plt.title("SSE={:.2f}".format(sse))
+plt.axis([-30,30,-30,30])
+plt.show()
+
+```
+
+得到下图：
+![](http://img.blog.csdn.net/20150820180422017)
+
+
+