Ftrl在线训练 (FtrlTrainStreamOp)

Java 类名：com.alibaba.alink.operator.stream.onlinelearning.FtrlTrainStreamOp

Python 类名：FtrlTrainStreamOp

功能介绍

Ftrl 是一个经典的在线学习算法，支持在线实时训练模型，并按照用户指定的频率输出模型。该算法支持稀疏和稠密两种训练数据。

算法原理

Ftrl 算法的参数更新公式如下：

该更新公式使得每一个维度上的特征更新学习率都是单独考虑，而不是使用全局学习率，保证累积加和限定了新的迭代结果W不要离“已迭代过的解”太远。具体细节请参阅文献[1]。

算法使用

Ftrl 在点击率预估，推荐等领域应用广泛。

• 备注 ：该组件训练的时候 FeatureCols 和 VectorCol 是两个互斥参数，只能有一个参数来描述算法的输入特征。

文献

[1] McMahan, H. Brendan, et al. “Ad click prediction: a view from the trenches.” Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining. 2013.

参数说明

代码示例

以下代码仅用于示意，可能需要修改部分代码或者配置环境后才能正常运行！

Python 代码

trainData0 = RandomTableSourceBatchOp() \
            .setNumCols(5) \
            .setNumRows(100) \
            .setOutputCols(["f0", "f1", "f2", "f3", "label"]) \
            .setOutputColConfs("label:weight_set(1.0,1.0,2.0,5.0)")
model = LogisticRegressionTrainBatchOp() \
            .setFeatureCols(["f0", "f1", "f2", "f3"]) \
            .setLabelCol("label") \
            .setMaxIter(10).linkFrom(trainData0)
trainData1 = RandomTableSourceStreamOp() \
            .setNumCols(5) \
            .setMaxRows(10000) \
            .setOutputCols(["f0", "f1", "f2", "f3", "label"]) \
            .setOutputColConfs("label:weight_set(1.0,1.0,2.0,5.0)") \
            .setTimePerSample(0.1)
models = FtrlTrainStreamOp(model) \
            .setFeatureCols(["f0", "f1", "f2", "f3"]) \
            .setLabelCol("label") \
            .setTimeInterval(10) \
            .setAlpha(0.1) \
            .setBeta(0.1) \
            .setL1(0.1) \
            .setL2(0.1)\
            .setWithIntercept(True) \
            .linkFrom(trainData1).print()
StreamOperator.execute()

Java 代码

package com.alibaba.alink.operator.stream.ml.onlinelearning;
import com.alibaba.alink.operator.batch.BatchOperator;
import com.alibaba.alink.operator.batch.classification.LogisticRegressionTrainBatchOp;
import com.alibaba.alink.operator.batch.source.RandomTableSourceBatchOp;
import com.alibaba.alink.operator.stream.StreamOperator;
import com.alibaba.alink.operator.stream.onlinelearning.FtrlTrainStreamOp;
import com.alibaba.alink.operator.stream.source.RandomTableSourceStreamOp;
import org.junit.Test;
public class FtrlTrainTestTest {
    @Test
    public void FtrlClassification() throws Exception {
        StreamOperator.setParallelism(2);
        BatchOperator trainData0 = new RandomTableSourceBatchOp()
            .setNumCols(5)
            .setNumRows(100L)
            .setOutputCols(new String[]{"f0", "f1", "f2", "f3", "label"})
            .setOutputColConfs("label:weight_set(1.0,1.0,2.0,5.0)");
        BatchOperator model = new LogisticRegressionTrainBatchOp()
            .setFeatureCols(new String[]{"f0", "f1", "f2", "f3"})
            .setLabelCol("label")
            .setMaxIter(10).linkFrom(trainData0);
        StreamOperator trainData1 = new RandomTableSourceStreamOp()
            .setNumCols(5)
            .setMaxRows(100L)
            .setOutputCols(new String[]{"f0", "f1", "f2", "f3", "label"})
            .setOutputColConfs("label:weight_set(1.0,1.0,2.0,5.0)")
            .setTimePerSample(0.1);
        new FtrlTrainStreamOp(model)
            .setFeatureCols(new String[]{"f0", "f1", "f2", "f3"})
            .setLabelCol("label")
            .setTimeInterval(10)
            .setAlpha(0.1)
            .setBeta(0.1)
            .setL1(0.1)
            .setL2(0.1)
            .setWithIntercept(true)
            .linkFrom(trainData1).print();
        StreamOperator.execute();
    }
}

运行结果