推荐组件：精排 (RecommendationRanking)

Java 类名：com.alibaba.alink.pipeline.recommendation.RecommendationRanking

Python 类名：RecommendationRanking

功能介绍

该组件功能是对召回的结果进行排序，并输出排序后的TopK个object，此处排序算法用户可以通过创建PipelineModel的方式定制，具体使用方式参见代码示例。

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
mTableCol	MTable 列名	召回列表列	String	✓
modelFilePath	模型的文件路径	模型的文件路径	String			null
outputCol	输出结果列	输出结果列列名，可选，默认null	String			null
overwriteSink	是否覆写已有数据	是否覆写已有数据	Boolean			false
rankingCol	用来排序的得分列	用来排序的得分列	String			null
reservedCols	算法保留列名	算法保留列	String[]			null
topN	前N的数据	挑选最近的N个数据	Integer		x >= 1	10
modelStreamFilePath	模型流的文件路径	模型流的文件路径	String			null
modelStreamScanInterval	扫描模型路径的时间间隔	描模型路径的时间间隔，单位秒	Integer			10
modelStreamStartTime	模型流的起始时间	模型流的起始时间。默认从当前时刻开始读。使用yyyy-mm-dd hh:mm:ss.fffffffff格式，详见Timestamp.valueOf(String s)	String			null

代码示例

Python 代码

from pyalink.alink import *

import pandas as pd

useLocalEnv(1)

import pandas as pd

data = pd.DataFrame([["u6", "0.0 1.0", 0.0, 1.0, 1, "{\"data\":{\"iid\":[18,19,88]},\"schema\":\"iid INT\"}"]])
predData = BatchOperator.fromDataframe(data, schemaStr='uid string, uf string, f0 double, f1 double, labels int, ilist string')
predData = predData.link(ToMTableBatchOp().setSelectedCol("ilist"))
data = pd.DataFrame([
            ["u0", "1.0 1.0", 1.0, 1.0, 1, 18],
			["u1", "1.0 1.0", 1.0, 1.0, 0, 19],
			["u2", "1.0 0.0", 1.0, 0.0, 1, 88],
			["u3", "1.0 0.0", 1.0, 0.0, 0, 18],
			["u4", "0.0 1.0", 0.0, 1.0, 1, 88],
			["u5", "0.0 1.0", 0.0, 1.0, 0, 19],
			["u6", "0.0 1.0", 0.0, 1.0, 1, 88]]);
trainData = BatchOperator.fromDataframe(data, schemaStr='uid string, uf string, f0 double, f1 double, labels int, iid string')
oneHotCols = ["uid", "f0", "f1", "iid"]
multiHotCols = ["uf"]
pipe = Pipeline() \
    .add( \
        OneHotEncoder() \
            .setSelectedCols(oneHotCols) \
			.setOutputCols(["ovec"])) \
    .add( \
		MultiHotEncoder().setDelimiter(" ") \
            .setSelectedCols(multiHotCols) \
            .setOutputCols(["mvec"])) \
	.add( \
		VectorAssembler() \
			.setSelectedCols(["ovec", "mvec"]) \
			.setOutputCol("vec")) \
	.add(
		LogisticRegression() \
            .setVectorCol("vec") \
			.setLabelCol("labels") \
			.setReservedCols(["uid", "iid"]) \
			.setPredictionDetailCol("detail") \
			.setPredictionCol("pred")) \
	.add( \
		JsonValue() \
			.setSelectedCol("detail") \
			.setJsonPath(["$.1"]) \
			.setOutputCols(["score"]))
lrModel = pipe.fit(trainData)
rank = RecommendationRanking()\
			.setModelData(lrModel.save())\
			.setMTableCol("ilist")\
			.setOutputCol("il")\
			.setTopN(2)\
			.setRankingCol("score")\
			.setReservedCols(["uid", "labels"])
rank.transform(predData).print()

Java 代码

import org.apache.flink.types.Row;

import com.alibaba.alink.operator.batch.BatchOperator;
import com.alibaba.alink.operator.batch.source.MemSourceBatchOp;
import com.alibaba.alink.pipeline.Pipeline;
import com.alibaba.alink.pipeline.classification.LogisticRegression;
import com.alibaba.alink.pipeline.dataproc.JsonValue;
import com.alibaba.alink.pipeline.dataproc.vector.VectorAssembler;
import com.alibaba.alink.pipeline.feature.MultiHotEncoder;
import com.alibaba.alink.pipeline.feature.OneHotEncoder;
import org.junit.Test;

import java.util.Arrays;

public class RecommendationRankingTest {

	 @Test
	 public void test() throws Exception {

		Row[] predArray = new Row[] {
			Row.of("u6", "0.0 1.0", 0.0, 1.0, 1, "{\"data\":{\"iid\":[18,19,88]},"
				+ "\"schema\":\"iid INT\"}")
		};

		Row[] trainArray = new Row[] {
			Row.of("u0", "1.0 1.0", 1.0, 1.0, 1, 18),
			Row.of("u1", "1.0 1.0", 1.0, 1.0, 0, 19),
			Row.of("u2", "1.0 0.0", 1.0, 0.0, 1, 88),
			Row.of("u3", "1.0 0.0", 1.0, 0.0, 1, 18),
			Row.of("u4", "0.0 1.0", 0.0, 1.0, 1, 88),
			Row.of("u5", "0.0 1.0", 0.0, 1.0, 1, 19),
			Row.of("u6", "0.0 1.0", 0.0, 1.0, 1, 88)
		};
		BatchOperator <?> trainData = new MemSourceBatchOp(Arrays.asList(trainArray),
				new String[] {"uid", "uf", "f0", "f1", "labels", "iid"});
		BatchOperator <?> predData =  new MemSourceBatchOp(Arrays.asList(predArray),
				new String[] {"uid", "uf", "f0", "f1", "labels", "ilist"});

		String[] oneHotCols = new String[] {"uid", "f0", "f1", "iid"};
		String[] multiHotCols = new String[] {"uf"};

		Pipeline pipe = new Pipeline()
			.add(
				new OneHotEncoder()
					.setSelectedCols(oneHotCols)
					.setOutputCols("ovec"))
			.add(
				new MultiHotEncoder().setDelimiter(" ")
					.setSelectedCols(multiHotCols)
					.setOutputCols("mvec"))
			.add(
				new VectorAssembler()
					.setSelectedCols("ovec", "mvec")
					.setOutputCol("vec"))
			.add(
				new LogisticRegression()
					.setVectorCol("vec")
					.setLabelCol("labels")
					.setReservedCols("uid", "iid")
					.setPredictionDetailCol("detail")
					.setPredictionCol("pred"))
			.add(
				new JsonValue()
					.setSelectedCol("detail")
					.setJsonPath("$.1")
					.setOutputCols("score"));
		RecommendationRanking rank = new RecommendationRanking()
			.setModelData(pipe.fit(trainData).save())
			.setMTableCol("ilist")
			.setOutputCol("ilist")
			.setTopN(2)
			.setRankingCol("score")
			.setReservedCols("uid", "labels");
		rank.transform(predData).print();
	}
}

运行结果

uid	uf	f0	f1	labels	ilist
u6	0.0 1.0	0.0000	1.0000	1	{“data”:{“iid”:[18,88],“score”:[0.9999999999999553,0.9999999999999472]},“schema”:“iid INT,score DOUBLE”}