fix multiclass with nonstandard label (dotnet#207)

daholste · Dmitry-A · commit d08e9cf2d616 · 2019-08-22T14:40:11.000-07:00
diff --git a/src/Microsoft.ML.Auto/PipelineSuggesters/PipelineSuggester.cs b/src/Microsoft.ML.Auto/PipelineSuggesters/PipelineSuggester.cs
@@ -221,7 +221,8 @@ private static IEnumerable<SuggestedTransform> CalculateTransforms(
             // this is a work-around for ML.NET bug tracked by https://github.com/dotnet/machinelearning/issues/1969
             if (task == TaskKind.MulticlassClassification)
             {
-                var transform = ValueToKeyMappingExtension.CreateSuggestedTransform(context, DefaultColumnNames.Label, DefaultColumnNames.Label);
+                var labelColumn = columns.First(c => c.Item3 == ColumnPurpose.Label).Item1;
+                var transform = ValueToKeyMappingExtension.CreateSuggestedTransform(context, labelColumn, labelColumn);
                 transforms.Add(transform);
             }
             return transforms;
diff --git a/src/Microsoft.ML.Auto/TrainerExtensions/MultiTrainerExtensions.cs b/src/Microsoft.ML.Auto/TrainerExtensions/MultiTrainerExtensions.cs
@@ -24,7 +24,7 @@ public ITrainerEstimator CreateInstance(MLContext mlContext, IEnumerable<Sweepab
             ColumnInformation columnInfo)
         {
             var binaryTrainer = _binaryLearnerCatalogItem.CreateInstance(mlContext, sweepParams, columnInfo) as ITrainerEstimatorProducingFloat;
-            return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer);
+            return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer, labelColumn: columnInfo.LabelColumn);
         }
 
         public PipelineNode CreatePipelineNode(IEnumerable<SweepableParam> sweepParams, ColumnInformation columnInfo)
@@ -46,7 +46,7 @@ public ITrainerEstimator CreateInstance(MLContext mlContext, IEnumerable<Sweepab
             ColumnInformation columnInfo)
         {
             var binaryTrainer = _binaryLearnerCatalogItem.CreateInstance(mlContext, sweepParams, columnInfo) as ITrainerEstimatorProducingFloat;
-            return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer);
+            return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer, labelColumn: columnInfo.LabelColumn);
         }
 
         public PipelineNode CreatePipelineNode(IEnumerable<SweepableParam> sweepParams, ColumnInformation columnInfo)
@@ -89,7 +89,7 @@ public ITrainerEstimator CreateInstance(MLContext mlContext, IEnumerable<Sweepab
             ColumnInformation columnInfo)
         {
             var binaryTrainer = _binaryLearnerCatalogItem.CreateInstance(mlContext, sweepParams, columnInfo) as ITrainerEstimatorProducingFloat;
-            return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer);
+            return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer, labelColumn: columnInfo.LabelColumn);
         }
 
         public PipelineNode CreatePipelineNode(IEnumerable<SweepableParam> sweepParams, ColumnInformation columnInfo)
@@ -132,7 +132,7 @@ public ITrainerEstimator CreateInstance(MLContext mlContext, IEnumerable<Sweepab
             ColumnInformation columnInfo)
         {
             var binaryTrainer = _binaryLearnerCatalogItem.CreateInstance(mlContext, sweepParams, columnInfo) as ITrainerEstimatorProducingFloat;
-            return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer);
+            return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer, labelColumn: columnInfo.LabelColumn);
         }
 
         public PipelineNode CreatePipelineNode(IEnumerable<SweepableParam> sweepParams, ColumnInformation columnInfo)
@@ -154,7 +154,7 @@ public ITrainerEstimator CreateInstance(MLContext mlContext, IEnumerable<Sweepab
             ColumnInformation columnInfo)
         {
             var binaryTrainer = _binaryLearnerCatalogItem.CreateInstance(mlContext, sweepParams, columnInfo) as ITrainerEstimatorProducingFloat;
-            return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer);
+            return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer, labelColumn: columnInfo.LabelColumn);
         }
 
         public PipelineNode CreatePipelineNode(IEnumerable<SweepableParam> sweepParams, ColumnInformation columnInfo)
@@ -176,7 +176,7 @@ public ITrainerEstimator CreateInstance(MLContext mlContext, IEnumerable<Sweepab
             ColumnInformation columnInfo)
         {
             var binaryTrainer = _binaryLearnerCatalogItem.CreateInstance(mlContext, sweepParams, columnInfo) as ITrainerEstimatorProducingFloat;
-            return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer);
+            return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer, labelColumn: columnInfo.LabelColumn);
         }
 
         public PipelineNode CreatePipelineNode(IEnumerable<SweepableParam> sweepParams, ColumnInformation columnInfo)
@@ -198,7 +198,7 @@ public ITrainerEstimator CreateInstance(MLContext mlContext, IEnumerable<Sweepab
             ColumnInformation columnInfo)
         {
             var binaryTrainer = _binaryLearnerCatalogItem.CreateInstance(mlContext, sweepParams, columnInfo) as ITrainerEstimatorProducingFloat;
-            return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer);
+            return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer, labelColumn: columnInfo.LabelColumn);
         }
 
         public PipelineNode CreatePipelineNode(IEnumerable<SweepableParam> sweepParams, ColumnInformation columnInfo)

Original file line number	Diff line number	Diff line change
`@@ -221,7 +221,8 @@ private static IEnumerable<SuggestedTransform> CalculateTransforms(`
`221`	`221`	`// this is a work-around for ML.NET bug tracked by https://github.com/dotnet/machinelearning/issues/1969`
`222`	`222`	`if (task == TaskKind.MulticlassClassification)`
`223`	`223`	`{`
`224`		`- var transform = ValueToKeyMappingExtension.CreateSuggestedTransform(context, DefaultColumnNames.Label, DefaultColumnNames.Label);`
	`224`	`+ var labelColumn = columns.First(c => c.Item3 == ColumnPurpose.Label).Item1;`
	`225`	`+ var transform = ValueToKeyMappingExtension.CreateSuggestedTransform(context, labelColumn, labelColumn);`
`225`	`226`	`transforms.Add(transform);`
`226`	`227`	`}`
`227`	`228`	`return transforms;`
Original file line number	Diff line number	Diff line change
`@@ -24,7 +24,7 @@ public ITrainerEstimator CreateInstance(MLContext mlContext, IEnumerable<Sweepab`
`24`	`24`	`ColumnInformation columnInfo)`
`25`	`25`	`{`
`26`	`26`	`var binaryTrainer = _binaryLearnerCatalogItem.CreateInstance(mlContext, sweepParams, columnInfo) as ITrainerEstimatorProducingFloat;`
`27`		`- return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer);`
	`27`	`+ return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer, labelColumn: columnInfo.LabelColumn);`
`28`	`28`	`}`
`29`	`29`
`30`	`30`	`public PipelineNode CreatePipelineNode(IEnumerable<SweepableParam> sweepParams, ColumnInformation columnInfo)`
`@@ -46,7 +46,7 @@ public ITrainerEstimator CreateInstance(MLContext mlContext, IEnumerable<Sweepab`
`46`	`46`	`ColumnInformation columnInfo)`
`47`	`47`	`{`
`48`	`48`	`var binaryTrainer = _binaryLearnerCatalogItem.CreateInstance(mlContext, sweepParams, columnInfo) as ITrainerEstimatorProducingFloat;`
`49`		`- return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer);`
	`49`	`+ return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer, labelColumn: columnInfo.LabelColumn);`
`50`	`50`	`}`
`51`	`51`
`52`	`52`	`public PipelineNode CreatePipelineNode(IEnumerable<SweepableParam> sweepParams, ColumnInformation columnInfo)`
`@@ -89,7 +89,7 @@ public ITrainerEstimator CreateInstance(MLContext mlContext, IEnumerable<Sweepab`
`89`	`89`	`ColumnInformation columnInfo)`
`90`	`90`	`{`
`91`	`91`	`var binaryTrainer = _binaryLearnerCatalogItem.CreateInstance(mlContext, sweepParams, columnInfo) as ITrainerEstimatorProducingFloat;`
`92`		`- return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer);`
	`92`	`+ return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer, labelColumn: columnInfo.LabelColumn);`
`93`	`93`	`}`
`94`	`94`
`95`	`95`	`public PipelineNode CreatePipelineNode(IEnumerable<SweepableParam> sweepParams, ColumnInformation columnInfo)`
`@@ -132,7 +132,7 @@ public ITrainerEstimator CreateInstance(MLContext mlContext, IEnumerable<Sweepab`
`132`	`132`	`ColumnInformation columnInfo)`
`133`	`133`	`{`
`134`	`134`	`var binaryTrainer = _binaryLearnerCatalogItem.CreateInstance(mlContext, sweepParams, columnInfo) as ITrainerEstimatorProducingFloat;`
`135`		`- return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer);`
	`135`	`+ return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer, labelColumn: columnInfo.LabelColumn);`
`136`	`136`	`}`
`137`	`137`
`138`	`138`	`public PipelineNode CreatePipelineNode(IEnumerable<SweepableParam> sweepParams, ColumnInformation columnInfo)`
`@@ -154,7 +154,7 @@ public ITrainerEstimator CreateInstance(MLContext mlContext, IEnumerable<Sweepab`
`154`	`154`	`ColumnInformation columnInfo)`
`155`	`155`	`{`
`156`	`156`	`var binaryTrainer = _binaryLearnerCatalogItem.CreateInstance(mlContext, sweepParams, columnInfo) as ITrainerEstimatorProducingFloat;`
`157`		`- return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer);`
	`157`	`+ return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer, labelColumn: columnInfo.LabelColumn);`
`158`	`158`	`}`
`159`	`159`
`160`	`160`	`public PipelineNode CreatePipelineNode(IEnumerable<SweepableParam> sweepParams, ColumnInformation columnInfo)`
`@@ -176,7 +176,7 @@ public ITrainerEstimator CreateInstance(MLContext mlContext, IEnumerable<Sweepab`
`176`	`176`	`ColumnInformation columnInfo)`
`177`	`177`	`{`
`178`	`178`	`var binaryTrainer = _binaryLearnerCatalogItem.CreateInstance(mlContext, sweepParams, columnInfo) as ITrainerEstimatorProducingFloat;`
`179`		`- return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer);`
	`179`	`+ return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer, labelColumn: columnInfo.LabelColumn);`
`180`	`180`	`}`
`181`	`181`
`182`	`182`	`public PipelineNode CreatePipelineNode(IEnumerable<SweepableParam> sweepParams, ColumnInformation columnInfo)`
`@@ -198,7 +198,7 @@ public ITrainerEstimator CreateInstance(MLContext mlContext, IEnumerable<Sweepab`
`198`	`198`	`ColumnInformation columnInfo)`
`199`	`199`	`{`
`200`	`200`	`var binaryTrainer = _binaryLearnerCatalogItem.CreateInstance(mlContext, sweepParams, columnInfo) as ITrainerEstimatorProducingFloat;`
`201`		`- return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer);`
	`201`	`+ return mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryTrainer, labelColumn: columnInfo.LabelColumn);`
`202`	`202`	`}`
`203`	`203`
`204`	`204`	`public PipelineNode CreatePipelineNode(IEnumerable<SweepableParam> sweepParams, ColumnInformation columnInfo)`