向量近似最近邻训练 (VectorApproxNearestNeighborTrainBatchOp)

Java 类名：com.alibaba.alink.operator.batch.similarity.VectorApproxNearestNeighborTrainBatchOp

Python 类名：VectorApproxNearestNeighborTrainBatchOp

功能介绍

该功能由训练和预测组成，训练时指定距离计算方式，生成最近邻模型

可选择的距离计算方式包含EUCLIDEAN和JACCARD两种，同时支持KDTREE和LSH两种近似方法。

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
idCol	id列名	id列名	String	✓
selectedCol	选中的列名	计算列对应的列名	String	✓	所选列类型为 [DENSE_VECTOR, SPARSE_VECTOR, STRING, VECTOR]
metric	距离度量方式	距离类型	String		“EUCLIDEAN”, “JACCARD”	“EUCLIDEAN”
numHashTables	哈希表的数目	哈希表的数目	Integer			1
numProjectionsPerTable	每个哈希表中的哈希函数个数	每个哈希表中的哈希函数个数	Integer			1
projectionWidth	桶的宽度	桶的宽度	Double			1.0
seed	采样种子	采样种子	Long			0
solver	近似方法	近似方法，包括KDTREE和LSH	String		“KDTREE”, “LSH”, “LOCAL_LSH”	“KDTREE”

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, "1 1 1"],
    [2, "2 2 2"]
])

inOp = BatchOperator.fromDataframe(df, schemaStr='id int, vec string')
train = VectorApproxNearestNeighborTrainBatchOp().setIdCol("id").setSelectedCol("vec").linkFrom(inOp)
predict = VectorApproxNearestNeighborPredictBatchOp().setSelectedCol("vec").setTopN(3).linkFrom(train, inOp)
predict.print()

Java 代码

import org.apache.flink.types.Row;

import com.alibaba.alink.operator.batch.BatchOperator;
import com.alibaba.alink.operator.batch.similarity.VectorApproxNearestNeighborPredictBatchOp;
import com.alibaba.alink.operator.batch.similarity.VectorApproxNearestNeighborTrainBatchOp;
import com.alibaba.alink.operator.batch.source.MemSourceBatchOp;
import org.junit.Test;

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

public class VectorApproxNearestNeighborTrainBatchOpTest {
	@Test
	public void testVectorApproxNearestNeighborTrainBatchOp() throws Exception {
		List <Row> df = Arrays.asList(
			Row.of(0, "0 0 0"),
			Row.of(1, "1 1 1"),
			Row.of(2, "2 2 2")
		);
		BatchOperator <?> inOp = new MemSourceBatchOp(df, "id int, vec string");
		BatchOperator <?> train = new VectorApproxNearestNeighborTrainBatchOp().setIdCol("id").setSelectedCol("vec")
			.linkFrom(inOp);
		BatchOperator <?> predict = new VectorApproxNearestNeighborPredictBatchOp().setSelectedCol("vec").setTopN(3)
			.linkFrom(train, inOp);
		predict.print();
	}
}

运行结果

id	vec
0	{“ID”:“[0,1,2]”,“METRIC”:“[0.0,1.7320508075688772,3.4641016151377544]”}
1	{“ID”:“[1,2,0]”,“METRIC”:“[0.0,1.7320508075688772,1.7320508075688772]”}
2	{“ID”:“[2,1,0]”,“METRIC”:“[0.0,1.7320508075688772,3.4641016151377544]”}