一份为KREpresentives和LSHK代表准备的套餐
krepresentatives的Python项目详细描述
用于聚类分类数据的k-Representatives和LSH-k-Representatives算法的Python实现:
与k-模式算法不同,k-代表和LSH-k-代表定义了保持聚类所有分类值频率的“代表”。在
安装:
使用pip:
pip install krepresentatives
导入包:
^{pr2}$生成一个简单的分类数据集:
X= np.array([[0,0],[0,1],[0,0],[1,1],[2,2],[2,3],[2,3]])y= np.array([0,0,0,0,1,1,1])
k代表:
kreps= kRepresentatives(X,y,n_init=5,n_clusters=2 ,verbose=3) kreps.fit_predict()
内置评估指标:
kreps.CalcScore()
输出:
kRepresentatives Init 0 Iter 0 Cost: 8.00 Move: 0 Num empty: 0 Timelapse: 0.00 Iter 1 Cost: 4.83 Move: 0 Num empty: 0 Timelapse: 0.00 Iter 2 Cost: 4.83 Move: 0 Num empty: 0 Timelapse: 0.00 kRepresentatives Init 1 Iter 0 Cost: 9.48 Move: 0 Num empty: 0 Timelapse: 0.00 Iter 1 Cost: 6.50 Move: 1 Num empty: 0 Timelapse: 0.00 Iter 2 Cost: 5.33 Move: 0 Num empty: 0 Timelapse: 0.00 Iter 3 Cost: 5.33 Move: 0 Num empty: 0 Timelapse: 0.00 kRepresentatives Init 2 Iter 0 Cost: 9.08 Move: 0 Num empty: 0 Timelapse: 0.00 Iter 1 Cost: 7.60 Move: 0 Num empty: 0 Timelapse: 0.00 Iter 2 Cost: 7.60 Move: 0 Num empty: 0 Timelapse: 0.00 kRepresentatives Init 3 Iter 0 Cost: 9.31 Move: 0 Num empty: 0 Timelapse: 0.00 Iter 1 Cost: 6.50 Move: 1 Num empty: 0 Timelapse: 0.00 Iter 2 Cost: 5.33 Move: 0 Num empty: 0 Timelapse: 0.00 Iter 3 Cost: 5.33 Move: 0 Num empty: 0 Timelapse: 0.00 kRepresentatives Init 4 Iter 0 Cost: 9.42 Move: 0 Num empty: 0 Timelapse: 0.00 Iter 1 Cost: 7.60 Move: 0 Num empty: 0 Timelapse: 0.00 Iter 2 Cost: 7.60 Move: 0 Num empty: 0 Timelapse: 0.00 Score: 4.833333333333334 Time: 0.0015569399999999956 Purity: 1.00 NMI: 1.00 ARI: 1.00 Sil: 0.52 Acc: 1.00 Recall: 1.00 Precision: 1.00
参数:
X:分类数据集
y: 对象标签(仅供评估)
n_init:初始化次数
n_clusters:目标簇数
最大迭代次数:最大迭代次数
冗长:
随机状态:
输出:
集群代表:最终代表名单
标签:预测标签
成本:物体到质心的距离平方和
迭代次数
epoch_costs_u:初始化的平均时间
LSH-k-代表:待更新
参考文献:
[1]San,Ohn Mar,Van Nam Huynh和Yoshiteru Nakamori。”分类数据聚类k-means算法的另一种扩展〉,《国际应用数学和计算机科学杂志》14(2004):241-247。 [2] 待更新
- 项目
标签: