ALINK(四十一)：模型评估（六）聚类评估 (EvalClusterBatchOp)

Java 类名：com.alibaba.alink.operator.batch.evaluation.EvalClusterBatchOp

Python 类名：EvalClusterBatchOp

功能介绍

聚类评估是对聚类算法的预测结果进行效果评估，支持下列评估指标。

参数说明

名称	中文名称	描述	类型	是否必须？	默认值
predictionCol	预测结果列名	预测结果列名	String	✓
labelCol	标签列名	输入表中的标签列名	String		null
vectorCol	向量列名	输入表中的向量列名	String		null
distanceType	距离度量方式	距离类型	String		"EUCLIDEAN"

代码示例

Python 代码

from pyalink.alink import *
import pandas as pd
useLocalEnv(1)
df = pd.DataFrame([
    [0, "0 0 0"],
    [0, "0.1,0.1,0.1"],
    [0, "0.2,0.2,0.2"],
    [1, "9 9 9"],
    [1, "9.1 9.1 9.1"],
    [1, "9.2 9.2 9.2"]
])
inOp = BatchOperator.fromDataframe(df, schemaStr='id int, vec string')
metrics = EvalClusterBatchOp().setVectorCol("vec").setPredictionCol("id").linkFrom(inOp).collectMetrics()
print("Total Samples Number:", metrics.getCount())
print("Cluster Number:", metrics.getK())
print("Cluster Array:", metrics.getClusterArray())
print("Cluster Count Array:", metrics.getCountArray())
print("CP:", metrics.getCp())
print("DB:", metrics.getDb())
print("SP:", metrics.getSp())
print("SSB:", metrics.getSsb())
print("SSW:", metrics.getSsw())
print("CH:", metrics.getVrc())

Java 代码

import org.apache.flink.types.Row;
import com.alibaba.alink.operator.batch.BatchOperator;
import com.alibaba.alink.operator.batch.evaluation.EvalClusterBatchOp;
import com.alibaba.alink.operator.batch.source.MemSourceBatchOp;
import com.alibaba.alink.operator.common.evaluation.ClusterMetrics;
import org.junit.Test;
import java.util.Arrays;
import java.util.List;
public class EvalClusterBatchOpTest {
  @Test
  public void testEvalClusterBatchOp() throws Exception {
    List <Row> df = Arrays.asList(
      Row.of(0, "0 0 0"),
      Row.of(0, "0.1,0.1,0.1"),
      Row.of(0, "0.2,0.2,0.2"),
      Row.of(1, "9 9 9"),
      Row.of(1, "9.1 9.1 9.1"),
      Row.of(1, "9.2 9.2 9.2")
    );
    BatchOperator <?> inOp = new MemSourceBatchOp(df, "id int, vec string");
    ClusterMetrics metrics = new EvalClusterBatchOp().setVectorCol("vec").setPredictionCol("id").linkFrom(inOp)
      .collectMetrics();
    System.out.println("Total Samples Number:" + metrics.getCount());
    System.out.println("Cluster Number:" + metrics.getK());
    System.out.println("Cluster Array:" + Arrays.toString(metrics.getClusterArray()));
    System.out.println("Cluster Count Array:" + Arrays.toString(metrics.getCountArray()));
    System.out.println("CP:" + metrics.getCp());
    System.out.println("DB:" + metrics.getDb());
    System.out.println("SP:" + metrics.getSp());
    System.out.println("SSB:" + metrics.getSsb());
    System.out.println("SSW:" + metrics.getSsw());
    System.out.println("CH:" + metrics.getVrc());
  }
}

运行结果

Total Samples Number: 6
Cluster Number: 2
Cluster Array: ['0', '1']
Cluster Count Array: [3.0, 3.0]
CP: 0.11547005383792497
DB: 0.014814814814814791
SP: 15.588457268119896
SSB: 364.5
SSW: 0.1199999999999996
CH: 12150.000000000042