分位数离散化训练 (QuantileDiscretizerTrainBatchOp)

Java 类名：com.alibaba.alink.operator.batch.feature.QuantileDiscretizerTrainBatchOp
Python 类名：QuantileDiscretizerTrainBatchOp

功能介绍

分位点离散可以计算选定列的分位点，然后使用这些分位点进行离散化。
生成选中列对应的q-quantile，其中可以所有列指定一个，也可以每一列对应一个

编码结果

Encode ——> INDEX

预测结果为单个token的index

Encode ——> VECTOR

预测结果为稀疏向量:

1. dropLast为true,向量中非零元个数为0或者1
2. dropLast为false,向量中非零元个数必定为1

Encode ——> ASSEMBLED_VECTOR

预测结果为稀疏向量,是预测选择列中,各列预测为VECTOR时,按照选择顺序ASSEMBLE的结果。

向量维度

Encode ——> Vector

numBuckets: 训练参数
dropLast: 预测参数
handleInvalid: 预测参数

Token index

Encode ——> Vector

1. 正常数据: 唯一的非零元为数据所在的bucket,若 dropLast为true, 最大的bucket的值会被丢掉，预测结果为全零元
2. null: 
    2.1 handleInvalid为keep: 唯一的非零元为:numBuckets - dropLast(true: 1, false: 0)
    2.2 handleInvalid为skip: null
    2.3 handleInvalid为error: 报错

参数说明

代码示例

Python 代码

from pyalink.alink import *
import pandas as pd
useLocalEnv(1)
df = pd.DataFrame([
    ["a", 1, 1, 2.0, True],
    ["c", 1, 2, -3.0, True],
    ["a", 2, 2, 2.0, False],
    ["c", 0, 0, 0.0, False]
])
batchSource = BatchOperator.fromDataframe(
    df, schemaStr='f_string string, f_long long, f_int int, f_double double, f_boolean boolean')
streamSource = StreamOperator.fromDataframe(
    df, schemaStr='f_string string, f_long long, f_int int, f_double double, f_boolean boolean')
trainOp = QuantileDiscretizerTrainBatchOp()\
    .setSelectedCols(['f_double'])\
    .setNumBuckets(8)\
    .linkFrom(batchSource)
predictBatchOp = QuantileDiscretizerPredictBatchOp()\
    .setSelectedCols(['f_double'])
predictStreamOp = QuantileDiscretizerPredictStreamOp(trainOp)\
    .setSelectedCols(['f_double'])
predictBatchOp.linkFrom(trainOp, batchSource).print()
predictStreamOp.linkFrom(streamSource) .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.QuantileDiscretizerPredictBatchOp;
import com.alibaba.alink.operator.batch.feature.QuantileDiscretizerTrainBatchOp;
import com.alibaba.alink.operator.batch.source.MemSourceBatchOp;
import com.alibaba.alink.operator.stream.StreamOperator;
import com.alibaba.alink.operator.stream.feature.QuantileDiscretizerPredictStreamOp;
import com.alibaba.alink.operator.stream.source.MemSourceStreamOp;
import org.junit.Test;
import java.util.Arrays;
import java.util.List;
public class QuantileDiscretizerTrainBatchOpTest {
    @Test
    public void testQuantileDiscretizerTrainBatchOp() throws Exception {
        List <Row> df = Arrays.asList(
            Row.of("a", 1, 1, 2.0, true),
            Row.of("c", 1, 2, -3.0, true),
            Row.of("a", 2, 2, 2.0, false),
            Row.of("c", 0, 0, 0.0, false)
        );
        BatchOperator <?> batchSource = new MemSourceBatchOp(df,
            "f_string string, f_long int, f_int int, f_double double, f_boolean boolean");
        StreamOperator <?> streamSource = new MemSourceStreamOp(df,
            "f_string string, f_long int, f_int int, f_double double, f_boolean boolean");
        BatchOperator <?> trainOp = new QuantileDiscretizerTrainBatchOp()
            .setSelectedCols("f_double")
            .setNumBuckets(8)
            .linkFrom(batchSource);
        BatchOperator <?> predictBatchOp = new QuantileDiscretizerPredictBatchOp()
            .setSelectedCols("f_double");
        predictBatchOp.linkFrom(trainOp, batchSource).print();
        StreamOperator <?> predictStreamOp = new QuantileDiscretizerPredictStreamOp(trainOp)
            .setSelectedCols("f_double");
        predictStreamOp.linkFrom(streamSource).print();
        StreamOperator.execute();
    }
}

运行结果

| f_string | f_long | f_int | f_double | f_boolean | | —- | —- | —- | —- | —- |

| a | 1 | 1 | 2 | true |

| c | 1 | 2 | 0 | true |

| a | 2 | 2 | 2 | false |

| c | 0 | 0 | 1 | false |