ALINK(二十)：数据处理数值型数据处理向量标准化 VectorNormalizeBatchOp/向量标准化训练 VectorStandardScalerTrainBatchOp /向量

Posted 2021-06-17 秋华

向量标准化 (VectorNormalizeBatchOp)

Java 类名：com.alibaba.alink.operator.batch.dataproc.vector.VectorNormalizeBatchOp

Python 类名：VectorNormalizeBatchOp

功能介绍

对 Vector 进行正则化操作。

指定参数范数的阶，例如p = 2, 对于向量<x1, x2, x3>，计算向量的平方和再开二次方记为norm，最终计算结果为<x1/norm, x2/norm, x3/norm>

参数说明

名称	中文名称	描述	类型	是否必须？	默认值
selectedCol	选中的列名	计算列对应的列名	String	✓
outputCol	输出结果列	输出结果列列名，可选，默认null	String		null
reservedCols	算法保留列名	算法保留列	String[]		null
p	范数的阶	范数的阶，默认2	Double		2.0
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)
df = pd.DataFrame([
    ["1:3,2:4,4:7", 1],
    ["0:3,5:5", 3],
    ["2:4,4:5", 4]
])
data = BatchOperator.fromDataframe(df, schemaStr="vec string, id bigint")
VectorNormalizeBatchOp().setSelectedCol("vec").setOutputCol("vec_norm").linkFrom(data).collectToDataframe()

Java 代码

import org.apache.flink.types.Row;
import com.alibaba.alink.operator.batch.BatchOperator;
import com.alibaba.alink.operator.batch.dataproc.vector.VectorNormalizeBatchOp;
import com.alibaba.alink.operator.batch.source.MemSourceBatchOp;
import org.junit.Test;
import java.util.Arrays;
import java.util.List;
public class VectorNormalizeBatchOpTest {
  @Test
  public void testVectorNormalizeBatchOp() throws Exception {
    List <Row> df = Arrays.asList(
      Row.of("1:3,2:4,4:7", 1),
      Row.of("0:3,5:5", 3),
      Row.of("2:4,4:5", 4)
    );
    BatchOperator <?> data = new MemSourceBatchOp(df, "vec string, id int");
    new VectorNormalizeBatchOp().setSelectedCol("vec").setOutputCol("vec_norm").linkFrom(data).print();
  }
}

运行结果

vec	id	vec_norm
1:3,2:4,4:7	1	1:0.34874291623145787 2:0.46499055497527714 4:0.813733471206735
0:3,5:5	3	0:0.5144957554275265 5:0.8574929257125441
2:4,4:5	4	2:0.6246950475544243 4:0.7808688094430304

 

 

 

 

向量标准化训练 (VectorStandardScalerTrainBatchOp)

Java 类名：com.alibaba.alink.operator.batch.dataproc.vector.VectorStandardScalerTrainBatchOp

Python 类名：VectorStandardScalerTrainBatchOp

功能介绍

标准化是对向量数据进行按正态化处理的组件

生成向量标准化的模型，在VectorStandardScalerPredictBatchOp中加载，对数据做标准化处理。

参数说明

名称	中文名称	描述	类型	是否必须？	默认值
selectedCol	选中的列名	计算列对应的列名	String	✓
withMean	是否使用均值	是否使用均值，默认使用	Boolean		true
withStd	是否使用标准差	是否使用标准差，默认使用	Boolean		true

代码示例

Python 代码

from pyalink.alink import *
import pandas as pd
useLocalEnv(1)
df = pd.DataFrame([
    ["a", "10.0, 100"],
    ["b", "-2.5, 9"],
    ["c", "100.2, 1"],
    ["d", "-99.9, 100"],
    ["a", "1.4, 1"],
    ["b", "-2.2, 9"],
    ["c", "100.9, 1"]
])
data = BatchOperator.fromDataframe(df, schemaStr="col string, vector string")
trainOp = VectorStandardScalerTrainBatchOp().setSelectedCol("vector")
model = trainOp.linkFrom(data)
VectorStandardScalerPredictBatchOp().linkFrom(model, data).collectToDataframe()

Java 代码

import org.apache.flink.types.Row;
import com.alibaba.alink.operator.batch.BatchOperator;
import com.alibaba.alink.operator.batch.dataproc.vector.VectorStandardScalerPredictBatchOp;
import com.alibaba.alink.operator.batch.dataproc.vector.VectorStandardScalerTrainBatchOp;
import com.alibaba.alink.operator.batch.source.MemSourceBatchOp;
import org.junit.Test;
import java.util.Arrays;
import java.util.List;
public class VectorStandardScalerTrainBatchOpTest {
  @Test
  public void testVectorStandardScalerTrainBatchOp() throws Exception {
    List <Row> df = Arrays.asList(
      Row.of("a", "10.0, 100"),
      Row.of("b", "-2.5, 9"),
      Row.of("c", "100.2, 1"),
      Row.of("d", "-99.9, 100"),
      Row.of("a", "1.4, 1"),
      Row.of("b", "-2.2, 9"),
      Row.of("c", "100.9, 1")
    );
    BatchOperator <?> data = new MemSourceBatchOp(df, "col string, vector string");
    BatchOperator <?> trainOp = new VectorStandardScalerTrainBatchOp().setSelectedCol("vector");
    BatchOperator <?> model = trainOp.linkFrom(data);
    new VectorStandardScalerPredictBatchOp().linkFrom(model, data).print();
  }
}

运行结果

col1	vec
a	-0.07835182408093559,1.4595814453461897
c	1.2269606224811418,-0.6520885789229323
b	-0.2549018445693762,-0.4814485769617911
a	-0.20280511721213143,-0.6520885789229323
c	1.237090541689495,-0.6520885789229323
b	-0.25924323851581327,-0.4814485769617911
d	-1.6687491397923802,1.4595814453461897

 

 

 

 

 

 

 

 

向量标准化预测 (VectorStandardScalerPredictBatchOp)

Java 类名：com.alibaba.alink.operator.batch.dataproc.vector.VectorStandardScalerPredictBatchOp

Python 类名：VectorStandardScalerPredictBatchOp

功能介绍

标准化是对向量数据进行按正态化处理的组件

加载VectorStandardScalerTrainBatchOp中生成的模型，对向量数据做标准化预处理。

参数说明

名称	中文名称	描述	类型	是否必须？	默认值
outputCol	输出结果列	输出结果列列名，可选，默认null	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)
df = pd.DataFrame([
    ["a", "10.0, 100"],
    ["b", "-2.5, 9"],
    ["c", "100.2, 1"],
    ["d", "-99.9, 100"],
    ["a", "1.4, 1"],
    ["b", "-2.2, 9"],
    ["c", "100.9, 1"]
])
data = BatchOperator.fromDataframe(df, schemaStr="col string, vector string")
trainOp = VectorStandardScalerTrainBatchOp().setSelectedCol("vector")
model = trainOp.linkFrom(data)
VectorStandardScalerPredictBatchOp().linkFrom(model, data).collectToDataframe()

Java 代码

import org.apache.flink.types.Row;
import com.alibaba.alink.operator.batch.BatchOperator;
import com.alibaba.alink.operator.batch.dataproc.vector.VectorStandardScalerPredictBatchOp;
import com.alibaba.alink.operator.batch.dataproc.vector.VectorStandardScalerTrainBatchOp;
import com.alibaba.alink.operator.batch.source.MemSourceBatchOp;
import org.junit.Test;
import java.util.Arrays;
import java.util.List;
public class VectorStandardScalerPredictBatchOpTest {
  @Test
  public void testVectorStandardScalerPredictBatchOp() throws Exception {
    List <Row> df = Arrays.asList(
      Row.of("a", "10.0, 100"),
      Row.of("b", "-2.5, 9"),
      Row.of("c", "100.2, 1"),
      Row.of("d", "-99.9, 100"),
      Row.of("a", "1.4, 1"),
      Row.of("b", "-2.2, 9"),
      Row.of("c", "100.9, 1")
    );
    BatchOperator <?> data = new MemSourceBatchOp(df, "col string, vector string");
    BatchOperator <?> trainOp = new VectorStandardScalerTrainBatchOp().setSelectedCol("vector");
    BatchOperator <?> model = trainOp.linkFrom(data);
    new VectorStandardScalerPredictBatchOp().linkFrom(model, data).print();
  }
}

 

运行结果

col1	vec
a	-0.07835182408093559,1.4595814453461897
c	1.2269606224811418,-0.6520885789229323
b	-0.2549018445693762,-0.4814485769617911
a	-0.20280511721213143,-0.6520885789229323
c	1.237090541689495,-0.6520885789229323
b	-0.25924323851581327,-0.4814485769617911
d	-1.6687491397923802,1.4595814453461897