二分K均值聚类训练 (BisectingKMeansTrainBatchOp)

Java 类名：com.alibaba.alink.operator.batch.clustering.BisectingKMeansTrainBatchOp

Python 类名：BisectingKMeansTrainBatchOp

功能介绍

二分k均值算法是k-means聚类算法的一个变体，主要是为了改进k-means算法随机选择初始质心的随机性造成聚类结果不确定性的问题。

二分k均值方法是一种自顶向下的聚类。初始阶段该算法将所有节点视为属于同一个cluster。在每次迭代中，它选择一个类别，然后使用Kmeans算法对该类中的所有数据点进行二分类。当没有类可拆分时或者达到最大聚类个数时，算法终止。

参数说明

训练

名称	中文名称	描述	类型	是否必须？	取值范围	默认值
vectorCol	向量列名	向量列对应的列名	String	✓	所选列类型为 [DENSE_VECTOR, SPARSE_VECTOR, STRING, VECTOR]
distanceType	距离度量方式	聚类使用的距离类型	String		“EUCLIDEAN”, “COSINE”	“EUCLIDEAN”
k	聚类中心点数目	聚类中心点数目	Integer			4
maxIter	最大迭代步数	最大迭代步数，默认为 10。	Integer			10
minDivisibleClusterSize	最小可分裂的聚类大小	最小可分裂的聚类大小	Integer			1
randomSeed	随机数种子	随机数种子	Integer			0

代码示例

Python 代码

from pyalink.alink import *

import pandas as pd

useLocalEnv(1)

df = pd.DataFrame([
    [0, "0 0 0"],
    [1, "0.1,0.1,0.1"],
    [2, "0.2,0.2,0.2"],
    [3, "9 9 9"],
    [4, "9.1 9.1 9.1"],
    [5, "9.2 9.2 9.2"]
])

inBatch = BatchOperator.fromDataframe(df, schemaStr='id int, vec string')
inStream = StreamOperator.fromDataframe(df, schemaStr='id int, vec string')

kmeansTrain = BisectingKMeansTrainBatchOp()\
    .setVectorCol("vec")\
    .setK(2)\
    .linkFrom(inBatch)
kmeansTrain.lazyPrint(10)

predictBatch = BisectingKMeansPredictBatchOp()\
    .setPredictionCol("pred")\
    .linkFrom(kmeansTrain, inBatch)
predictBatch.print()

predictStream = BisectingKMeansPredictStreamOp(kmeansTrain)\
    .setPredictionCol("pred")\
    .linkFrom(inStream)
predictStream.print()

StreamOperator.execute()

Java 代码

import org.apache.flink.types.Row;

import com.alibaba.alink.operator.batch.BatchOperator;
import com.alibaba.alink.operator.batch.clustering.BisectingKMeansPredictBatchOp;
import com.alibaba.alink.operator.batch.clustering.BisectingKMeansTrainBatchOp;
import com.alibaba.alink.operator.batch.source.MemSourceBatchOp;
import com.alibaba.alink.operator.stream.StreamOperator;
import com.alibaba.alink.operator.stream.clustering.BisectingKMeansPredictStreamOp;
import com.alibaba.alink.operator.stream.source.MemSourceStreamOp;
import org.junit.Test;

import java.util.Arrays;
import java.util.List;

public class BisectingKMeansTrainBatchOpTest {
	@Test
	public void testBisectingKMeansTrainBatchOp() throws Exception {
		List <Row> df = Arrays.asList(
			Row.of(0, "0 0 0"),
			Row.of(1, "0.1,0.1,0.1"),
			Row.of(2, "0.2,0.2,0.2"),
			Row.of(3, "9 9 9"),
			Row.of(4, "9.1 9.1 9.1"),
			Row.of(5, "9.2 9.2 9.2")
		);
		BatchOperator <?> inBatch = new MemSourceBatchOp(df, "id int, vec string");
		StreamOperator <?> inStream = new MemSourceStreamOp(df, "id int, vec string");
		BatchOperator <?> kmeansTrain = new BisectingKMeansTrainBatchOp()
			.setVectorCol("vec")
			.setK(2)
            .linkFrom(inBatch);
		kmeansTrain.lazyPrint(10);

		BatchOperator <?> predictBatch = new BisectingKMeansPredictBatchOp()
			.setPredictionCol("pred")
		    .linkFrom(kmeansTrain, inBatch);
		predictBatch.print();

		StreamOperator <?> predictStream = new BisectingKMeansPredictStreamOp(kmeansTrain)
			.setPredictionCol("pred")
            .linkFrom(inStream);
		predictStream.print();
		StreamOperator.execute();
	}
}

运行结果

模型结果

model_id	model_info
0	{“vectorCol”:“"vec"”,“distanceType”:“"EUCLIDEAN"”,“k”:“2”,“vectorSize”:“3”}
1048576	{“clusterId”:1,“size”:6,“center”:{“data”:[4.6,4.6,4.6]},“cost”:364.61999999999995}
2097152	{“clusterId”:2,“size”:3,“center”:{“data”:[0.1,0.1,0.1]},“cost”:0.06}
3145728	{“clusterId”:3,“size”:3,“center”:{“data”:[9.099999999999998,9.099999999999998,9.099999999999998]},“cost”:0.060000000000172804}

预测结果

id	vec	pred
0	0 0 0	0
1	0.1,0.1,0.1	0
2	0.2,0.2,0.2	0
3	9 9 9	1
4	9.1 9.1 9.1	1
5	9.2 9.2 9.2	1

ALinkLab