特征哈希 (FeatureHasher)

Java 类名：com.alibaba.alink.pipeline.feature.FeatureHasher

Python 类名：FeatureHasher

功能介绍

将多个特征组合成一个特征向量。

参数说明

名称	中文名称	描述	类型	是否必须？	取值范围	默认值
outputCol	输出结果列列名	输出结果列列名，必选	String	✓
selectedCols	选择的列名	计算列对应的列名列表	String[]	✓
categoricalCols	离散特征列名	离散特征列名	String[]
numFeatures	向量维度	生成向量长度	Integer			262144
reservedCols	算法保留列名	算法保留列	String[]			null
numThreads	组件多线程线程个数	组件多线程线程个数	Integer			1

代码示例

Python 代码

from pyalink.alink import *

import pandas as pd

useLocalEnv(1)

df1 = pd.DataFrame([
    [1.1, True, "2", "A"],
    [1.1, False, "2", "B"],
    [1.1, True, "1", "B"],
    [2.2, True, "1", "A"]
])

inOp = BatchOperator.fromDataframe(df1, schemaStr='double double, bool boolean, number int, str string')
binarizer = FeatureHasher().setSelectedCols(["double", "bool", "number", "str"]).setOutputCol("output").setNumFeatures(200)
binarizer.transform(inOp).print()

df2 = pd.DataFrame([
    [1.1, True, "2", "A"],
    [1.1, False, "2", "B"],
    [1.1, True, "1", "B"],
    [2.2, True, "1", "A"]
])

inOp1 = BatchOperator.fromDataframe(df2, schemaStr='double double, bool boolean, number int, str string')
inOp2 = StreamOperator.fromDataframe(df2, schemaStr='double double, bool boolean, number int, str string')

hasher = FeatureHasherBatchOp().setSelectedCols(["double", "bool", "number", "str"]).setOutputCol("output").setNumFeatures(200)
hasher.linkFrom(inOp1).print()

hasher = FeatureHasherStreamOp().setSelectedCols(["double", "bool", "number", "str"]).setOutputCol("output").setNumFeatures(200)
hasher.linkFrom(inOp2).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.FeatureHasherBatchOp;
import com.alibaba.alink.operator.batch.source.MemSourceBatchOp;
import com.alibaba.alink.operator.stream.StreamOperator;
import com.alibaba.alink.operator.stream.feature.FeatureHasherStreamOp;
import com.alibaba.alink.operator.stream.source.MemSourceStreamOp;
import com.alibaba.alink.pipeline.feature.FeatureHasher;
import org.junit.Test;

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

public class FeatureHasherTest {
	@Test
	public void testFeatureHasher() throws Exception {
		List <Row> df1 = Arrays.asList(
			Row.of(1.1, true, 2, "A"),
			Row.of(1.1, false, 2, "B"),
			Row.of(1.1, true, 1, "B"),
			Row.of(2.2, true, 1, "A")
		);
		BatchOperator <?> inOp = new MemSourceBatchOp(df1, "double double, bool boolean, number int, str string");
		FeatureHasher binarizer = new FeatureHasher().setSelectedCols("double", "bool", "number", "str").setOutputCol(
			"output").setNumFeatures(200);
		binarizer.transform(inOp).print();
		List <Row> df2 = Arrays.asList(
			Row.of(1.1, true, 2, "A"),
			Row.of(1.1, false, 2, "B"),
			Row.of(1.1, true, 1, "B"),
			Row.of(2.2, true, 1, "A")
		);
		BatchOperator <?> inOp1 = new MemSourceBatchOp(df2, "double double, bool boolean, number int, str string");
		StreamOperator <?> inOp2 = new MemSourceStreamOp(df2, "double double, bool boolean, number int, str string");
		BatchOperator <?> hasher1 = new FeatureHasherBatchOp().setSelectedCols("double", "bool", "number", "str")
			.setOutputCol("output").setNumFeatures(200);
		hasher1.linkFrom(inOp1).print();
		StreamOperator <?> hasher2 = new FeatureHasherStreamOp().setSelectedCols("double", "bool", "number", "str")
			.setOutputCol("output").setNumFeatures(200);
		hasher2.linkFrom(inOp2).print();
		StreamOperator.execute();
	}
}

运行结果

double	bool	number	str	output
1.1000	true	2	A	$200$13:2.0 38:1.1 45:1.0 195:1.0
1.1000	false	2	B	$200$13:2.0 30:1.0 38:1.1 76:1.0
1.1000	true	1	B	$200$13:1.0 38:1.1 76:1.0 195:1.0
2.2000	true	1	A	$200$13:1.0 38:2.2 45:1.0 195:1.0

double	bool	number	str	output
1.1000	true	2	A	$200$13:2.0 38:1.1 45:1.0 195:1.0
1.1000	false	2	B	$200$13:2.0 30:1.0 38:1.1 76:1.0
1.1000	true	1	B	$200$13:1.0 38:1.1 76:1.0 195:1.0
2.2000	true	1	A	$200$13:1.0 38:2.2 45:1.0 195:1.0

double	bool	number	str	output
2.2000	true	1	A	$200$13:1.0 38:2.2 45:1.0 195:1.0
1.1000	true	1	B	$200$13:1.0 38:1.1 76:1.0 195:1.0
1.1000	true	2	A	$200$13:2.0 38:1.1 45:1.0 195:1.0
1.1000	false	2	B	$200$13:2.0 30:1.0 38:1.1 76:1.0