ALINK(三十)：特征工程（九）特征选择（一）主成分分析（PcaTrainBatchOp/PcaPredictBatchOp）

主成分分析训练 (PcaTrainBatchOp)

Java 类名：com.alibaba.alink.operator.batch.feature.PcaTrainBatchOp

Python 类名：PcaTrainBatchOp

功能介绍

主成分分析，是考察多个变量间相关性一种多元统计方法，研究如何通过少数几个主成分来揭示多个变量间的内部结构，即从原始变量中导出少数几个主成分，使它们尽可能多地保留原始变量的信息，且彼此间互不相关，作为新的综合指标。详细介绍请见维基百科链接wiki。

参数说明

名称	中文名称	描述	类型	是否必须？	默认值
k	降维后的维度	降维后的维度	Integer	✓
calculationType	计算类型	计算类型，包含"CORR", "COV"两种。	String		"CORR"
selectedCols	选中的列名数组	计算列对应的列名列表	String[]		null
vectorCol	向量列名	向量列对应的列名，默认值是null	String		null

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],
        [0.1,0.2,0.1],
        [0.2,0.2,0.8],
        [9.0,9.5,9.7],
        [9.1,9.1,9.6],
        [9.2,9.3,9.9]
])
# batch source 
inOp = BatchOperator.fromDataframe(df, schemaStr='x1 double, x2 double, x3 double')
trainOp = PcaTrainBatchOp()
       .setK(2)
       .setSelectedCols(["x1","x2","x3"])
predictOp = PcaPredictBatchOp()
        .setPredictionCol("pred")
# batch train
inOp.link(trainOp)
# batch predict
predictOp.linkFrom(trainOp,inOp)
predictOp.print()
# stream predict
inStreamOp = StreamOperator.fromDataframe(df, schemaStr='x1 double, x2 double, x3 double')
predictStreamOp = PcaPredictStreamOp(trainOp)
        .setPredictionCol("pred")
predictStreamOp.linkFrom(inStreamOp)
predictStreamOp.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.feature.PcaPredictBatchOp;
import com.alibaba.alink.operator.batch.feature.PcaTrainBatchOp;
import com.alibaba.alink.operator.batch.source.MemSourceBatchOp;
import com.alibaba.alink.operator.stream.StreamOperator;
import com.alibaba.alink.operator.stream.feature.PcaPredictStreamOp;
import com.alibaba.alink.operator.stream.source.MemSourceStreamOp;
import org.junit.Test;
import java.util.Arrays;
import java.util.List;
public class PcaTrainBatchOpTest {
  @Test
  public void testPcaTrainBatchOp() throws Exception {
    List <Row> df = Arrays.asList(
      Row.of(0.0, 0.0, 0.0),
      Row.of(0.1, 0.2, 0.1),
      Row.of(0.2, 0.2, 0.8),
      Row.of(9.0, 9.5, 9.7),
      Row.of(9.1, 9.1, 9.6),
      Row.of(9.2, 9.3, 9.9)
    );
    BatchOperator <?> inOp = new MemSourceBatchOp(df, "x1 double, x2 double, x3 double");
    BatchOperator <?> trainOp = new PcaTrainBatchOp()
      .setK(2)
      .setSelectedCols("x1", "x2", "x3");
    BatchOperator <?> predictOp = new PcaPredictBatchOp()
      .setPredictionCol("pred");
    inOp.link(trainOp);
    predictOp.linkFrom(trainOp, inOp);
    predictOp.print();
    StreamOperator <?> inStreamOp = new MemSourceStreamOp(df, "x1 double, x2 double, x3 double");
    StreamOperator <?> predictStreamOp = new PcaPredictStreamOp(trainOp)
      .setPredictionCol("pred");
    predictStreamOp.linkFrom(inStreamOp);
    predictStreamOp.print();
    StreamOperator.execute();
  }
}

运行结果

x1	x2	x3	pred
9.0	9.5	9.7	3.2280384305400736,1.1516225426477789E-4
0.2	0.2	0.8	0.13565076707329407,0.09003329494282108
9.2	9.3	9.9	3.250783163664603,0.0456526246528135
9.1	9.1	9.6	3.182618319978973,0.027469531992220464
0.1	0.2	0.1	0.045855205015063565,-0.012182917696915518
0.0	0.0	0.0	0.0,0.0

主成分分析预测 (PcaPredictBatchOp)

Java 类名：com.alibaba.alink.operator.batch.feature.PcaPredictBatchOp

Python 类名：PcaPredictBatchOp