Java 类名：com.alibaba.alink.pipeline.tuning.GridSearchTVSplit
Python 类名：GridSearchTVSplit

功能介绍

gridsearch是通过参数数组组成的网格，对其中的每一组输入参数的组很分别进行训练，预测，评估。取得评估参数最优的模型，作为最终的返回模型
tv为训练验证，将数据按照比例切分为两份，对其中一份数据做训练，对剩余一份数据做预测和评估，得到一个评估结果。
此函数用tv方法得到每一个grid对应参数的评估结果，得到最优模型

参数说明

| 名称 | 中文名称 | 描述 | 类型 | 是否必须？ | 默认值 | | —- | —- | —- | —- | —- | —- |

| trainRatio | 训练集比例 | 训练集与验证集的划分比例，取值范围为(0, 1]。 | Double | | 0.8 |

| ParamGrid | 参数网格 | 指定参数的网格 | ParamGrid | ✓ | —- |

| Estimator | Estimator | 用于调优的Estimator | Estimator | ✓ | —- |

| TuningEvaluator | 评估指标 | 用于选择最优模型的评估指标 | TuningEvaluator | ✓ | —- |

代码示例

Python 代码

  1. from pyalink.alink import *
  3. import pandas as pd
  5. useLocalEnv(1)
  7. def adult(url):
  8.     data = (
  9.         CsvSourceBatchOp()
  10.             .setFilePath('https://alink-test-data.oss-cn-hangzhou.aliyuncs.com/adult_train.csv')
  11.             .setSchemaStr(
  12.             'age bigint, workclass string, fnlwgt bigint,'
  13.             'education string, education_num bigint,'
  14.             'marital_status string, occupation string,'
  15.             'relationship string, race string, sex string,'
  16.             'capital_gain bigint, capital_loss bigint,'
  17.             'hours_per_week bigint, native_country string,'
  18.             'label string'
  19.         )
  20.     )
  21.     return data
  23. def adult_train():
  24.     return adult('https://alink-test-data.oss-cn-hangzhou.aliyuncs.com/adult_train.csv')
  26. def adult_test():
  27.     return adult('https://alink-test-data.oss-cn-hangzhou.aliyuncs.com/adult_test.csv')
  29. def adult_numerical_feature_strs():
  30.     return [
  31.         "age", "fnlwgt", "education_num",
  32.         "capital_gain", "capital_loss", "hours_per_week"
  33.     ]
  35. def adult_categorical_feature_strs():
  36.     return [
  37.         "workclass", "education", "marital_status",
  38.         "occupation", "relationship", "race", "sex",
  39.         "native_country"
  40.     ]
  42. def adult_features_strs():
  43.     feature = adult_numerical_feature_strs()
  44.     feature.extend(adult_categorical_feature_strs())
  46.     return feature
  48. def rf_grid_search_cv(featureCols, categoryFeatureCols, label, metric):
  49.     rf = (
  50.         RandomForestClassifier()
  51.             .setFeatureCols(featureCols)
  52.             .setCategoricalCols(categoryFeatureCols)
  53.             .setLabelCol(label)
  54.             .setPredictionCol('prediction')
  55.             .setPredictionDetailCol('prediction_detail')
  56.     )
  57.     paramGrid = (
  58.         ParamGrid()
  59.             .addGrid(rf, 'SUBSAMPLING_RATIO', [1.0, 0.99, 0.98])
  60.             .addGrid(rf, 'NUM_TREES', [3, 6, 9])
  61.     )
  62.     tuningEvaluator = (
  63.         BinaryClassificationTuningEvaluator()
  64.             .setLabelCol(label)
  65.             .setPredictionDetailCol("prediction_detail")
  66.             .setTuningBinaryClassMetric(metric)
  67.     )
  68.     cv = (
  69.         GridSearchCV()
  70.             .setEstimator(rf)
  71.             .setParamGrid(paramGrid)
  72.             .setTuningEvaluator(tuningEvaluator)
  73.             .setNumFolds(2)
  74.             .enableLazyPrintTrainInfo("TrainInfo")
  75.     )
  77.     return cv
  79. def rf_grid_search_tv(featureCols, categoryFeatureCols, label, metric):
  80.     rf = (
  81.         RandomForestClassifier()
  82.             .setFeatureCols(featureCols)
  83.             .setCategoricalCols(categoryFeatureCols)
  84.             .setLabelCol(label)
  85.             .setPredictionCol('prediction')
  86.             .setPredictionDetailCol('prediction_detail')
  87.     )
  88.     paramGrid = (
  89.         ParamGrid()
  90.             .addGrid(rf, 'SUBSAMPLING_RATIO', [1.0, 0.99, 0.98])
  91.             .addGrid(rf, 'NUM_TREES', [3, 6, 9])
  92.     )
  93.     tuningEvaluator = (
  94.         BinaryClassificationTuningEvaluator()
  95.             .setLabelCol(label)
  96.             .setPredictionDetailCol("prediction_detail")
  97.             .setTuningBinaryClassMetric(metric)
  98.     )
  99.     cv = (
  100.         GridSearchTVSplit()
  101.             .setEstimator(rf)
  102.             .setParamGrid(paramGrid)
  103.             .setTuningEvaluator(tuningEvaluator)
  104.             .enableLazyPrintTrainInfo("TrainInfo")
  105.     )
  107.     return cv
  109. def tuningcv(cv_estimator, input):
  110.     return cv_estimator.fit(input)
  112. def tuningtv(tv_estimator, input):
  113.     return tv_estimator.fit(input)
  115. def main():
  116.     print('rf cv tuning')
  117.     model = tuningcv(
  118.         rf_grid_search_cv(adult_features_strs(),
  119.                           adult_categorical_feature_strs(), 'label', 'AUC'),
  120.         adult_train()
  121.     )
  123.     print('rf tv tuning')
  124.     model = tuningtv(
  125.         rf_grid_search_tv(adult_features_strs(),
  126.                           adult_categorical_feature_strs(), 'label', 'AUC'),
  127.         adult_train()
  128.     )
  130. main()

Java 代码

  1. import com.alibaba.alink.operator.batch.source.CsvSourceBatchOp;
  2. import com.alibaba.alink.pipeline.classification.RandomForestClassifier;
  3. import com.alibaba.alink.pipeline.tuning.BinaryClassificationTuningEvaluator;
  4. import com.alibaba.alink.pipeline.tuning.GridSearchTVSplit;
  5. import com.alibaba.alink.pipeline.tuning.GridSearchTVSplitModel;
  6. import com.alibaba.alink.pipeline.tuning.ParamGrid;
  7. import org.junit.Test;
  9. public class GridSearchTVSplitTest {
  10.     @Test
  11.     public void testGridSearchTVSplit() throws Exception {
  12.         String[] featureCols = new String[] {
  13.             "age", "fnlwgt", "education_num",
  14.             "capital_gain", "capital_loss", "hours_per_week",
  15.             "workclass", "education", "marital_status",
  16.             "occupation", "relationship", "race", "sex",
  17.             "native_country"
  18.         };
  19.         String[] categoryFeatureCols = new String[] {
  20.             "workclass", "education", "marital_status",
  21.             "occupation", "relationship", "race", "sex",
  22.             "native_country"
  23.         };
  24.         String label = "label";
  25.         CsvSourceBatchOp data = new CsvSourceBatchOp()
  26.             .setFilePath("https://alink-test-data.oss-cn-hangzhou.aliyuncs.com/adult_train.csv")
  27.             .setSchemaStr(
  28.                 "age bigint, workclass string, fnlwgt bigint, education string, education_num bigint, marital_status "
  29.                     + "string, occupation string, relationship string, race string, sex string, capital_gain bigint, "
  30.                     + "capital_loss bigint, hours_per_week bigint, native_country string, label string");
  31.         RandomForestClassifier rf = new RandomForestClassifier()
  32.             .setFeatureCols(featureCols)
  33.             .setCategoricalCols(categoryFeatureCols)
  34.             .setLabelCol(label)
  35.             .setPredictionCol("prediction")
  36.             .setPredictionDetailCol("prediction_detail");
  37.         ParamGrid paramGrid = new ParamGrid()
  38.             .addGrid(rf, RandomForestClassifier.SUBSAMPLING_RATIO, new Double[] {1.0, 0.99, 0.98})
  39.             .addGrid(rf, RandomForestClassifier.NUM_TREES, new Integer[] {3, 6, 9});
  40.         BinaryClassificationTuningEvaluator tuningEvaluator = new BinaryClassificationTuningEvaluator()
  41.             .setLabelCol(label)
  42.             .setPredictionDetailCol("prediction_detail")
  43.             .setTuningBinaryClassMetric("AUC");
  44.         GridSearchTVSplit cv = new GridSearchTVSplit()
  45.             .setEstimator(rf)
  46.             .setParamGrid(paramGrid)
  47.             .setTuningEvaluator(tuningEvaluator)
  48.             .setTrainRatio(0.8)
  49.             .enableLazyPrintTrainInfo("TrainInfo");
  50.         GridSearchTVSplitModel model = cv.fit(data);
  51.     }
  52. }

运行结果

TrainInfo
Metric information:
Metric name: AUC
Larger is better: true
Tuning information:

| AUC | stage | param | value | stage 2 | param 2 | value 2 | | —- | —- | —- | —- | —- | —- | —- |

| 0.9146840488554084 | RandomForestClassifier | numTrees | 9 | RandomForestClassifier | subsamplingRatio | 0.99 |

| 0.9125303612307454 | RandomForestClassifier | numTrees | 9 | RandomForestClassifier | subsamplingRatio | 1.0 |

| 0.9098037018824784 | RandomForestClassifier | numTrees | 6 | RandomForestClassifier | subsamplingRatio | 0.98 |

| 0.9091959987727252 | RandomForestClassifier | numTrees | 6 | RandomForestClassifier | subsamplingRatio | 0.99 |

| 0.9078221190139827 | RandomForestClassifier | numTrees | 9 | RandomForestClassifier | subsamplingRatio | 0.98 |

| 0.9025009982071417 | RandomForestClassifier | numTrees | 6 | RandomForestClassifier | subsamplingRatio | 1.0 |

| 0.9015572276497046 | RandomForestClassifier | numTrees | 3 | RandomForestClassifier | subsamplingRatio | 1.0 |

| 0.893774245901412 | RandomForestClassifier | numTrees | 3 | RandomForestClassifier | subsamplingRatio | 0.98 |

| 0.892706752181599 | RandomForestClassifier | numTrees | 3 | RandomForestClassifier | subsamplingRatio | 0.99 |

运行结果

  1. rf cv tuning
  2. com.alibaba.alink.pipeline.tuning.GridSearchCV
  3. [ {
  4.   "param" : [ {
  5.     "stage" : "RandomForestClassifier",
  6.     "paramName" : "numTrees",
  7.     "paramValue" : 3
  8.   }, {
  9.     "stage" : "RandomForestClassifier",
  10.     "paramName" : "subsamplingRatio",
  11.     "paramValue" : 1.0
  12.   } ],
  13.   "metric" : 0.8922549257899725
  14. }, {
  15.   "param" : [ {
  16.     "stage" : "RandomForestClassifier",
  17.     "paramName" : "numTrees",
  18.     "paramValue" : 3
  19.   }, {
  20.     "stage" : "RandomForestClassifier",
  21.     "paramName" : "subsamplingRatio",
  22.     "paramValue" : 0.99
  23.   } ],
  24.   "metric" : 0.8920255970548456
  25. }, {
  26.   "param" : [ {
  27.     "stage" : "RandomForestClassifier",
  28.     "paramName" : "numTrees",
  29.     "paramValue" : 3
  30.   }, {
  31.     "stage" : "RandomForestClassifier",
  32.     "paramName" : "subsamplingRatio",
  33.     "paramValue" : 0.98
  34.   } ],
  35.   "metric" : 0.8944982480437225
  36. }, {
  37.   "param" : [ {
  38.     "stage" : "RandomForestClassifier",
  39.     "paramName" : "numTrees",
  40.     "paramValue" : 6
  41.   }, {
  42.     "stage" : "RandomForestClassifier",
  43.     "paramName" : "subsamplingRatio",
  44.     "paramValue" : 1.0
  45.   } ],
  46.   "metric" : 0.8923867598288401
  47. }, {
  48.   "param" : [ {
  49.     "stage" : "RandomForestClassifier",
  50.     "paramName" : "numTrees",
  51.     "paramValue" : 6
  52.   }, {
  53.     "stage" : "RandomForestClassifier",
  54.     "paramName" : "subsamplingRatio",
  55.     "paramValue" : 0.99
  56.   } ],
  57.   "metric" : 0.9012141767959505
  58. }, {
  59.   "param" : [ {
  60.     "stage" : "RandomForestClassifier",
  61.     "paramName" : "numTrees",
  62.     "paramValue" : 6
  63.   }, {
  64.     "stage" : "RandomForestClassifier",
  65.     "paramName" : "subsamplingRatio",
  66.     "paramValue" : 0.98
  67.   } ],
  68.   "metric" : 0.8993774036693788
  69. }, {
  70.   "param" : [ {
  71.     "stage" : "RandomForestClassifier",
  72.     "paramName" : "numTrees",
  73.     "paramValue" : 9
  74.   }, {
  75.     "stage" : "RandomForestClassifier",
  76.     "paramName" : "subsamplingRatio",
  77.     "paramValue" : 1.0
  78.   } ],
  79.   "metric" : 0.8981738808130779
  80. }, {
  81.   "param" : [ {
  82.     "stage" : "RandomForestClassifier",
  83.     "paramName" : "numTrees",
  84.     "paramValue" : 9
  85.   }, {
  86.     "stage" : "RandomForestClassifier",
  87.     "paramName" : "subsamplingRatio",
  88.     "paramValue" : 0.99
  89.   } ],
  90.   "metric" : 0.9029671873892725
  91. }, {
  92.   "param" : [ {
  93.     "stage" : "RandomForestClassifier",
  94.     "paramName" : "numTrees",
  95.     "paramValue" : 9
  96.   }, {
  97.     "stage" : "RandomForestClassifier",
  98.     "paramName" : "subsamplingRatio",
  99.     "paramValue" : 0.98
  100.   } ],
  101.   "metric" : 0.905228896323363
  102. } ]
  103. rf tv tuning
  104. com.alibaba.alink.pipeline.tuning.GridSearchTVSplit
  105. [ {
  106.   "param" : [ {
  107.     "stage" : "RandomForestClassifier",
  108.     "paramName" : "numTrees",
  109.     "paramValue" : 3
  110.   }, {
  111.     "stage" : "RandomForestClassifier",
  112.     "paramName" : "subsamplingRatio",
  113.     "paramValue" : 1.0
  114.   } ],
  115.   "metric" : 0.9022694229691741
  116. }, {
  117.   "param" : [ {
  118.     "stage" : "RandomForestClassifier",
  119.     "paramName" : "numTrees",
  120.     "paramValue" : 3
  121.   }, {
  122.     "stage" : "RandomForestClassifier",
  123.     "paramName" : "subsamplingRatio",
  124.     "paramValue" : 0.99
  125.   } ],
  126.   "metric" : 0.8963559966080328
  127. }, {
  128.   "param" : [ {
  129.     "stage" : "RandomForestClassifier",
  130.     "paramName" : "numTrees",
  131.     "paramValue" : 3
  132.   }, {
  133.     "stage" : "RandomForestClassifier",
  134.     "paramName" : "subsamplingRatio",
  135.     "paramValue" : 0.98
  136.   } ],
  137.   "metric" : 0.9041948454957178
  138. }, {
  139.   "param" : [ {
  140.     "stage" : "RandomForestClassifier",
  141.     "paramName" : "numTrees",
  142.     "paramValue" : 6
  143.   }, {
  144.     "stage" : "RandomForestClassifier",
  145.     "paramName" : "subsamplingRatio",
  146.     "paramValue" : 1.0
  147.   } ],
  148.   "metric" : 0.8982021117392784
  149. }, {
  150.   "param" : [ {
  151.     "stage" : "RandomForestClassifier",
  152.     "paramName" : "numTrees",
  153.     "paramValue" : 6
  154.   }, {
  155.     "stage" : "RandomForestClassifier",
  156.     "paramName" : "subsamplingRatio",
  157.     "paramValue" : 0.99
  158.   } ],
  159.   "metric" : 0.9031851535310546
  160. }, {
  161.   "param" : [ {
  162.     "stage" : "RandomForestClassifier",
  163.     "paramName" : "numTrees",
  164.     "paramValue" : 6
  165.   }, {
  166.     "stage" : "RandomForestClassifier",
  167.     "paramName" : "subsamplingRatio",
  168.     "paramValue" : 0.98
  169.   } ],
  170.   "metric" : 0.9034443322241488
  171. }, {
  172.   "param" : [ {
  173.     "stage" : "RandomForestClassifier",
  174.     "paramName" : "numTrees",
  175.     "paramValue" : 9
  176.   }, {
  177.     "stage" : "RandomForestClassifier",
  178.     "paramName" : "subsamplingRatio",
  179.     "paramValue" : 1.0
  180.   } ],
  181.   "metric" : 0.8993474753000145
  182. }, {
  183.   "param" : [ {
  184.     "stage" : "RandomForestClassifier",
  185.     "paramName" : "numTrees",
  186.     "paramValue" : 9
  187.   }, {
  188.     "stage" : "RandomForestClassifier",
  189.     "paramName" : "subsamplingRatio",
  190.     "paramValue" : 0.99
  191.   } ],
  192.   "metric" : 0.9090250137144916
  193. }, {
  194.   "param" : [ {
  195.     "stage" : "RandomForestClassifier",
  196.     "paramName" : "numTrees",
  197.     "paramValue" : 9
  198.   }, {
  199.     "stage" : "RandomForestClassifier",
  200.     "paramName" : "subsamplingRatio",
  201.     "paramValue" : 0.98
  202.   } ],
  203.   "metric" : 0.9129786771786127
  204. } ]