Java 类名：com.alibaba.alink.operator.stream.onlinelearning.FtrlModelFilterStreamOp
Python 类名：FtrlModelFilterStreamOp

功能介绍

该组件是对ftrl 实时训练出来的模型进行实时过滤，将指标不好的模型丢弃掉，仅保留达到用户要求的模型。

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 | | —- | —- | —- | —- | —- | —- | —- |

| labelCol | 标签列名 | 输入表中的标签列名 | String | ✓ | | |

| accuracyThreshold | 模型过滤的Accuracy阈值 | 模型过滤的Accuracy阈值 | Double | | | 0.5 |

| aucThreshold | 模型过滤的Auc阈值 | 模型过滤的Auc阈值 | Double | | | 0.5 |

| positiveLabelValueString | 正样本 | 正样本对应的字符串格式。 | String | | | null |

| vectorCol | 向量列名 | 向量列对应的列名，默认值是null | String | | 所选列类型为 [DENSE_VECTOR, SPARSE_VECTOR, STRING, VECTOR] | null |

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

  1. trainData0 = RandomTableSourceBatchOp() \
  2.             .setNumCols(5) \
  3.             .setNumRows(100) \
  4.             .setOutputCols(["f0", "f1", "f2", "f3", "label"]) \
  5.             .setOutputColConfs("label:weight_set(1.0,1.0,2.0,5.0)")
  7. model = LogisticRegressionTrainBatchOp() \
  8.             .setFeatureCols(["f0", "f1", "f2", "f3"]) \
  9.             .setLabelCol("label") \
  10.             .setMaxIter(10).linkFrom(trainData0)
  12. trainData1 = RandomTableSourceStreamOp() \
  13.             .setNumCols(5) \
  14.             .setMaxRows(10000) \
  15.             .setOutputCols(["f0", "f1", "f2", "f3", "label"]) \
  16.             .setOutputColConfs("label:weight_set(1.0,1.0,2.0,5.0)") \
  17.             .setTimePerSample(0.1)
  19. models = FtrlTrainStreamOp(model, None) \
  20.             .setFeatureCols(["f0", "f1", "f2", "f3"]) \
  21.             .setLabelCol("label") \
  22.             .setTimeInterval(10) \
  23.             .setAlpha(0.1) \
  24.             .setBeta(0.1) \
  25.             .setL1(0.1) \
  26.             .setL2(0.1)\
  27.             .setVectorSize(4)\
  28.             .setWithIntercept(True) \
  29.             .linkFrom(trainData1)
  31. FtrlModelFilterStreamOp().setAucThreshold(0.5).setAccuracyThreshold(0.5) \
  32.             .setPositiveLabelValueString("1.0") \
  33.             .setLabelCol("label").linkFrom(models, trainData1).print()
  35. StreamOperator.execute()

Java 代码

  1. package com.alibaba.alink.operator.stream.ml.onlinelearning;
  3. import com.alibaba.alink.operator.batch.BatchOperator;
  4. import com.alibaba.alink.operator.batch.classification.LogisticRegressionTrainBatchOp;
  5. import com.alibaba.alink.operator.batch.source.RandomTableSourceBatchOp;
  6. import com.alibaba.alink.operator.stream.StreamOperator;
  7. import com.alibaba.alink.operator.stream.onlinelearning.FtrlModelFilterStreamOp;
  8. import com.alibaba.alink.operator.stream.onlinelearning.FtrlPredictStreamOp;
  9. import com.alibaba.alink.operator.stream.onlinelearning.FtrlTrainStreamOp;
  10. import com.alibaba.alink.operator.stream.source.RandomTableSourceStreamOp;
  11. import org.junit.Test;
  13. public class FtrlTrainTestTest {
  14.     @Test
  15.     public void FtrlClassification() throws Exception {
  16.         StreamOperator.setParallelism(2);
  17.         BatchOperator trainData0 = new RandomTableSourceBatchOp()
  18.             .setNumCols(5)
  19.             .setNumRows(100L)
  20.             .setOutputCols(new String[]{"f0", "f1", "f2", "f3", "label"})
  21.             .setOutputColConfs("label:weight_set(1.0,1.0,2.0,5.0)");
  23.         BatchOperator model = new LogisticRegressionTrainBatchOp()
  24.             .setFeatureCols(new String[]{"f0", "f1", "f2", "f3"})
  25.             .setLabelCol("label")
  26.             .setMaxIter(10).linkFrom(trainData0);
  28.         StreamOperator trainData1 = new RandomTableSourceStreamOp()
  29.             .setNumCols(5)
  30.             .setMaxRows(1000L)
  31.             .setOutputCols(new String[]{"f0", "f1", "f2", "f3", "label"})
  32.             .setOutputColConfs("label:weight_set(1.0,1.0,2.0,5.0)")
  33.             .setTimePerSample(0.1);
  35.         StreamOperator smodel = new FtrlTrainStreamOp(model)
  36.             .setFeatureCols(new String[]{"f0", "f1", "f2", "f3"})
  37.             .setLabelCol("label")
  38.             .setTimeInterval(10)
  39.             .setAlpha(0.1)
  40.             .setBeta(0.1)
  41.             .setL1(0.1)
  42.             .setL2(0.1)
  43.             .setVectorSize(4)
  44.             .setWithIntercept(true)
  45.             .linkFrom(trainData1);
  47.         new FtrlModelFilterStreamOp().setAucThreshold(0.5).setAccuracyThreshold(0.5)
  48.             .setPositiveLabelValueString("1.0")
  49.             .setLabelCol("label").linkFrom(smodel, trainData1).print();
  51.         StreamOperator.execute();
  52.     }
  53. }

输出结果

| alinkmodelstreamtimestamp | alinkmodelstreamcount | model_id | model_info | label_value | | —- | —- | —- | —- | —- |

| 2021-06-10 19:39:41.169 | 4 | 1048576 | {“featureColNames”:[“f0”,”f1”,”f2”,”f3”],”featureColTypes”:null,”coefVector”:{“data”:[0.8344505544432526,0.12691581782275618,0.218543815526658,0.22769775985648064,0.05203808913915911]},”coefVectors”:null,”convergenceInfo”:null} | null |

| 2021-06-10 19:39:41.169 | 4 | 0 | {“hasInterceptItem”:”true”,”modelName”:””Logistic Regression””,”labelCol”:null,”linearModelType”:””LR””,”vectorSize”:”4”} | null |

| 2021-06-10 19:39:41.169 | 4 | 2251799812636673 | null | 1.0000 |

| 2021-06-10 19:39:41.169 | 4 | 2251799812636672 | null | 2.0000 |

| 2021-06-10 19:40:11.319 | 4 | 0 | {“hasInterceptItem”:”true”,”modelName”:””Logistic Regression””,”labelCol”:null,”linearModelType”:””LR””,”vectorSize”:”4”} | null |

| 2021-06-10 19:40:11.319 | 4 | 1048576 | {“featureColNames”:[“f0”,”f1”,”f2”,”f3”],”featureColTypes”:null,”coefVector”:{“data”:[0.9795436401762672,0.2366945713649036,0.3644473752499545,0.2654714469214479,0.23195535286616062]},”coefVectors”:null,”convergenceInfo”:null} | null |

| 2021-06-10 19:40:11.319 | 4 | 2251799812636672 | null | 2.0000 |

| 2021-06-10 19:40:11.319 | 4 | 2251799812636673 | null | 1.0000 |