CLI tool - make validation dataset optional and support for crossvali…

…dation in generated code (dotnet#83) * Added sequential grouping of columns * reverted the file * bug fixes, more logic to templates to support cross-validate * formatting and fix type in consolehelper * Added logic in templates * revert settings
Dmitry-A · Aug 22, 2019 · fd71c47 · fd71c47
1 parent 6b179df
commit fd71c47
Show file tree

Hide file tree

Showing 7 changed files with 221 additions and 168 deletions.
diff --git a/src/mlnet/CodeGenerator/TrainerGenerators.cs b/src/mlnet/CodeGenerator/TrainerGenerators.cs
@@ -178,7 +178,7 @@ internal class LinearSvm : TrainerGeneratorBase
             internal override string MethodName => "LinearSupportVectorMachines";
 
             //ClassName of the options to trainer
-            internal override string OptionsName => "LinearSvm.Options";
+            internal override string OptionsName => "LinearSvmTrainer.Options";
 
             //The named parameters to the trainer.
             internal override IDictionary<string, string> NamedParameters

diff --git a/src/mlnet/Commands/CommandDefinitions.cs b/src/mlnet/Commands/CommandDefinitions.cs
@@ -18,23 +18,19 @@ public static System.CommandLine.Command New()
         {
             var newCommand = new System.CommandLine.Command("new", "ML.NET CLI tool for code generation",
 
-                            handler: CommandHandler.Create</*FileInfo,*/ FileInfo,/* FileInfo,*/ FileInfo, TaskKind, string>((/*FileInfo dataset,*/ FileInfo trainDataset, /*FileInfo validationDataset,*/ FileInfo testDataset, TaskKind mlTask, string labelColumnName) =>
+                            handler: CommandHandler.Create<FileInfo, FileInfo, TaskKind, string>((FileInfo trainDataset, FileInfo testDataset, TaskKind mlTask, string labelColumnName) =>
                                  {
                                      NewCommand.Run(new Options()
                                      {
-                                         /*Dataset = dataset,*/
                                          TrainDataset = trainDataset,
-                                         /*ValidationDataset = validationDataset,*/
                                          TestDataset = testDataset,
                                          MlTask = mlTask,
                                          LabelName = labelColumnName
                                      });
 
                                  }))
             {
-                                //Dataset(),
                                 TrainDataset(),
-                                //ValidationDataset(),
                                 TestDataset(),
                                 MlTask(),
                                 LabelName(),
@@ -51,10 +47,6 @@ public static System.CommandLine.Command New()
                 {
                     return "Option required : --train-dataset";
                 }
-                if (sym.Children["--test-dataset"] == null)
-                {
-                    return "Option required : --test-dataset";
-                }
                 if (sym.Children["--ml-task"] == null)
                 {
                     return "Option required : --ml-task";
@@ -69,21 +61,14 @@ public static System.CommandLine.Command New()
 
             return newCommand;
 
-            //Option Dataset() =>
-            //   new Option("--dataset", "Dataset file path.",
-            //              new Argument<FileInfo>().ExistingOnly());
 
             Option TrainDataset() =>
                new Option("--train-dataset", "Train dataset file path.",
                           new Argument<FileInfo>().ExistingOnly());
 
-            //Option ValidationDataset() =>
-            //   new Option("--validation-dataset", "Test dataset file path.",
-            //              new Argument<FileInfo>().ExistingOnly());
-
             Option TestDataset() =>
                new Option("--test-dataset", "Test dataset file path.",
-                          new Argument<FileInfo>().ExistingOnly());
+                          new Argument<FileInfo>(defaultValue: default(FileInfo)).ExistingOnly());
 
             Option MlTask() =>
                new Option("--ml-task", "Type of ML task.",
@@ -93,27 +78,6 @@ Option LabelName() =>
                new Option("--label-column-name", "Name of the label column.",
                           new Argument<string>());
 
-            //Option ColumnSeperator() =>
-            //   new Option("--column-separator", "Column separator in dataset file.",
-            //              new Argument<string>(defaultValue: default(string)));
-
-            //Option ExplorationTimeout() =>
-            //   new Option("--exploration-timeout", "Timeout for exploring the best models.",
-            //              new Argument<int>(defaultValue: 10));
-
-            //Option Name() =>
-            //   new Option("--name", "Name of the project file.",
-            //              new Argument<string>(defaultValue: "SampleProject"));
-
-            //Option ShowOutput() =>
-            //   new Option("--show-output", "Show output on the console",
-            //              new Argument<bool>(defaultValue: true));
-
-            //Option LabelIndex() =>
-            //  new Option("--label-column-index", "Index of the label column.",
-            //             new Argument<int>(defaultValue: -1));
-
-
         }
 
         private static string[] GetMlTaskSuggestions()

diff --git a/src/mlnet/Commands/NewCommand.cs b/src/mlnet/Commands/NewCommand.cs
@@ -24,17 +24,21 @@ internal static void Run(Options options)
             // For Version 0.1 It is required that the data set has header. 
             var columnInference = context.Data.InferColumns(options.TrainDataset.FullName, label, true, groupColumns: false);
             var textLoader = context.Data.CreateTextLoader(columnInference);
-            var trainData = textLoader.Read(options.TrainDataset.FullName);
 
-            var validationData = textLoader.Read(options.TestDataset.FullName);
-            Pipeline pipelineToDeconstruct = null;
+            IDataView trainData = textLoader.Read(options.TrainDataset.FullName);
+            IDataView validationData = options.TestDataset == null ? null : textLoader.Read(options.TestDataset.FullName);
 
-            var result = ExploreModels(options, context, label, trainData, validationData, pipelineToDeconstruct);
-            pipelineToDeconstruct = result.Item1;
+            //Explore the models
+            Pipeline pipeline = null;
+            var result = ExploreModels(options, context, label, trainData, validationData, pipeline);
+
+            //Get the best pipeline
+            pipeline = result.Item1;
             var model = result.Item2;
+
             //Path can be overriden from args
             GenerateModel(model, @"./BestModel", "model.zip", context);
-            RunCodeGen(options, columnInference, pipelineToDeconstruct);
+            RunCodeGen(options, columnInference, pipeline);
         }
 
         private static void GenerateModel(ITransformer model, string ModelPath, string modelName, MLContext mlContext)
@@ -116,7 +120,7 @@ private static void RunCodeGen(Options options, ColumnInferenceResult columnInfe
             MLCodeGen codeGen = new MLCodeGen()
             {
                 Path = options.TrainDataset.FullName,
-                TestPath = options.TestDataset.FullName,
+                TestPath = options.TestDataset?.FullName,
                 Columns = columns,
                 Transforms = transforms,
                 HasHeader = columnInference.HasHeader,