Upgrading CLI to produce ML.NET V.10 APIs and bunch of Refactoring tasks (dotnet#65)

* Added sequential grouping of columns

* reverted the file

* upgrade to v .10 and refactoring

* added null check

* fixed unit tests

* review comments

* removed the settings change

* added regions

* fixed unit tests

Author: srsaggam

src/AutoML/TrainerExtensions/TrainerExtensionUtil.cs

@@ -57,7 +57,7 @@ public static Action<T> CreateArgsFunc<T>(IEnumerable<SweepableParam> sweepParam
         }
 
         private static string[] _lightGbmTreeBoosterParamNames = new[] { "RegLambda", "RegAlpha" };
-        private const string LightGbmTreeBoosterPropName = "TreeBooster";
+        private const string LightGbmTreeBoosterPropName = "Booster";
 
         public static Action<LightGbmArguments> CreateLightGbmArgsFunc(IEnumerable<SweepableParam> sweepParams)
         {
@@ -92,7 +92,7 @@ private static IDictionary<string, object> BuildLightGbmPipelineNodeProps(IEnume
             var parentArgParams = sweepParams.Except(treeBoosterParams);
 
             var treeBoosterProps = treeBoosterParams.ToDictionary(p => p.Name, p => (object)p.ProcessedValue());
-            var treeBoosterCustomProp = new CustomProperty("LightGbmArguments.TreeBooster.Arguments", treeBoosterProps);
+            var treeBoosterCustomProp = new CustomProperty("Options.TreeBooster.Arguments", treeBoosterProps);
 
             var props = parentArgParams.ToDictionary(p => p.Name, p => (object)p.ProcessedValue());
             props[LightGbmTreeBoosterPropName] = treeBoosterCustomProp;

src/Test/TrainerExtensionsTests.cs

@@ -73,8 +73,8 @@ public void BuildPipelineNodePropsLightGbm()
   ""MaxCatThreshold"": 16,
   ""CatSmooth"": 10,
   ""CatL2"": 0.5,
-  ""TreeBooster"": {
-    ""Name"": ""LightGbmArguments.TreeBooster.Arguments"",
+  ""Booster"": {
+    ""Name"": ""Options.TreeBooster.Arguments"",
     ""Properties"": {
       ""RegLambda"": 0.5,
       ""RegAlpha"": 0.5
@@ -114,8 +114,8 @@ public void BuildParameterSetLightGbm()
             {
                 {"NumBoostRound", 1 },
                 {"LearningRate", 1 },
-                {"TreeBooster", new CustomProperty() {
-                    Name = "Microsoft.ML.LightGBM.TreeBooster",
+                {"Booster", new CustomProperty() {
+                    Name = "Options.TreeBooster.Arguments",
                     Properties = new Dictionary<string, object>()
                     {
                         {"RegLambda", 1 },

src/mlnet.Test/CodeGenTests.cs

@@ -24,9 +24,10 @@ public void TrainerGeneratorBasicNamedParameterTest()
             PipelineNode node = new PipelineNode("LightGbmBinary", PipelineNodeType.Trainer, new string[] { "Label" }, default(string), elementProperties);
             Pipeline pipeline = new Pipeline(new PipelineNode[] { node });
             CodeGenerator codeGenerator = new CodeGenerator(pipeline, null);
-            var actual = codeGenerator.GenerateTrainer();
+            var actual = codeGenerator.GenerateTrainerAndUsings();
             string expected = "LightGbm(learningRate:0.1f,numLeaves:1,labelColumn:\"Label\",featureColumn:\"Features\");";
-            Assert.AreEqual(expected, actual);
+            Assert.AreEqual(expected, actual.Item1);
+            Assert.IsNull(actual.Item2);
         }
 
         [TestMethod]
@@ -43,9 +44,11 @@ public void TrainerGeneratorBasicAdvancedParameterTest()
             PipelineNode node = new PipelineNode("LightGbmBinary", PipelineNodeType.Trainer, new string[] { "Label" }, default(string), elementProperties);
             Pipeline pipeline = new Pipeline(new PipelineNode[] { node });
             CodeGenerator codeGenerator = new CodeGenerator(pipeline, null);
-            var actual = codeGenerator.GenerateTrainer();
-            string expected = "LightGbm(new LightGbm.Options(){LearningRate=0.1f,NumLeaves=1,UseSoftmax=true,LabelColumn=\"Label\",FeatureColumn=\"Features\"});";
-            Assert.AreEqual(expected, actual);
+            var actual = codeGenerator.GenerateTrainerAndUsings();
+            string expectedTrainer = "LightGbm(new Options(){LearningRate=0.1f,NumLeaves=1,UseSoftmax=true,LabelColumn=\"Label\",FeatureColumn=\"Features\"});";
+            string expectedUsing = "using Microsoft.ML.LightGBM;\r\n";
+            Assert.AreEqual(expectedTrainer, actual.Item1);
+            Assert.AreEqual(expectedUsing, actual.Item2);
         }
 
         [TestMethod]
@@ -56,9 +59,25 @@ public void TransformGeneratorBasicTest()
             PipelineNode node = new PipelineNode("Normalizing", PipelineNodeType.Transform, new string[] { "Label" }, new string[] { "Label" }, elementProperties);
             Pipeline pipeline = new Pipeline(new PipelineNode[] { node });
             CodeGenerator codeGenerator = new CodeGenerator(pipeline, null);
-            var actual = codeGenerator.GenerateTransforms();
+            var actual = codeGenerator.GenerateTransformsAndUsings();
             string expected = "Normalize(\"Label\",\"Label\")";
-            Assert.AreEqual(expected, actual[0]);
+            Assert.AreEqual(expected, actual[0].Item1);
+            Assert.IsNull(actual[0].Item2);
+        }
+
+        [TestMethod]
+        public void TransformGeneratorUsingTest()
+        {
+            var context = new MLContext();
+            var elementProperties = new Dictionary<string, object>();
+            PipelineNode node = new PipelineNode("OneHotEncoding", PipelineNodeType.Transform, new string[] { "Label" }, new string[] { "Label" }, elementProperties);
+            Pipeline pipeline = new Pipeline(new PipelineNode[] { node });
+            CodeGenerator codeGenerator = new CodeGenerator(pipeline, null);
+            var actual = codeGenerator.GenerateTransformsAndUsings();
+            string expectedTransform = "Categorical.OneHotEncoding(new []{new OneHotEncodingEstimator.ColumnInfo(\"Label\",\"Label\")})";
+            var expectedUsings = "using Microsoft.ML.Transforms.Categorical;\r\n";
+            Assert.AreEqual(expectedTransform, actual[0].Item1);
+            Assert.AreEqual(expectedUsings, actual[0].Item2);
         }
 
         [TestMethod]
@@ -79,19 +98,6 @@ public void ClassLabelGenerationBasicTest()
             Assert.AreEqual(expected2, actual[1]);
         }
 
-        [TestMethod]
-        public void GenerateUsingsBasicTest()
-        {
-            var context = new MLContext();
-            var elementProperties = new Dictionary<string, object>();
-            PipelineNode node = new PipelineNode("TypeConverting", PipelineNodeType.Transform, new string[] { "Label" }, new string[] { "Label" }, elementProperties);
-            Pipeline pipeline = new Pipeline(new PipelineNode[] { node });
-            CodeGenerator codeGenerator = new CodeGenerator(pipeline, null);
-            var actual = codeGenerator.GenerateUsings();
-            string expected = "using Microsoft.ML.Transforms.Conversions;\r\n";
-            Assert.AreEqual(expected, actual);
-        }
-
         [TestMethod]
         public void ColumnGenerationTest()
         {
@@ -122,14 +128,16 @@ public void TrainerComplexParameterTest()
 
             var elementProperties = new Dictionary<string, object>()
             {
-                {"TreeBooster", new CustomProperty(){Properties= new Dictionary<string, object>(), Name = "TreeBooster"} },
+                {"Booster", new CustomProperty(){Properties= new Dictionary<string, object>(), Name = "TreeBooster"} },
             };
             PipelineNode node = new PipelineNode("LightGbmBinary", PipelineNodeType.Trainer, new string[] { "Label" }, default(string), elementProperties);
             Pipeline pipeline = new Pipeline(new PipelineNode[] { node });
             CodeGenerator codeGenerator = new CodeGenerator(pipeline, null);
-            var actual = codeGenerator.GenerateTrainer();
-            string expected = "LightGbm(new LightGbm.Options(){TreeBooster=new TreeBooster(){},LabelColumn=\"Label\",FeatureColumn=\"Features\"});";
-            Assert.AreEqual(expected, actual);
+            var actual = codeGenerator.GenerateTrainerAndUsings();
+            string expectedTrainer = "LightGbm(new Options(){Booster=new TreeBooster(){},LabelColumn=\"Label\",FeatureColumn=\"Features\"});";
+            var expectedUsings = "using Microsoft.ML.LightGBM;\r\n";
+            Assert.AreEqual(expectedTrainer, actual.Item1);
+            Assert.AreEqual(expectedUsings, actual.Item2);
 
         }
 

src/mlnet/CodeGenerator/CodeGenerator.cs

@@ -21,24 +21,25 @@ public CodeGenerator(Pipeline pipelineToDeconstruct, ColumnInferenceResult colum
             this.pipeline = pipelineToDeconstruct;
             this.columnInferenceResult = columnInferenceResult;
         }
-        internal IList<string> GenerateTransforms()
+        internal IList<(string, string)> GenerateTransformsAndUsings()
         {
             var nodes = pipeline.Nodes.Where(t => t.NodeType == PipelineNodeType.Transform);
-            var results = new List<string>();
+            var results = new List<(string, string)>();
             foreach (var node in nodes)
             {
                 ITransformGenerator generator = TransformGeneratorFactory.GetInstance(node);
-                results.Add(generator.GenerateTransformer());
+                results.Add((generator.GenerateTransformer(), generator.GenerateUsings()));
             }
 
             return results;
         }
 
-        internal string GenerateTrainer()
+        internal (string, string) GenerateTrainerAndUsings()
         {
             ITrainerGenerator generator = TrainerGeneratorFactory.GetInstance(pipeline);
             var trainerString = generator.GenerateTrainer();
-            return trainerString;
+            var trainerUsings = generator.GenerateUsings();
+            return (trainerString, trainerUsings);
         }
 
         internal IList<string> GenerateClassLabels()
@@ -149,44 +150,6 @@ private static string ConstructColumnDefinition(Column column)
             return def;
         }
 
-        internal string GenerateUsings()
-        {
-            var trainerNodes = pipeline.Nodes.Where(t => t.NodeType == PipelineNodeType.Trainer);
-            var transformNodes = pipeline.Nodes.Where(t => t.NodeType == PipelineNodeType.Transform);
-
-            StringBuilder sb = new StringBuilder();
-
-            foreach (var node in trainerNodes)
-            {
-                if (Enum.TryParse<TrainerName>(node.Name, out TrainerName nodeName))
-                {
-                    if (nodeName == TrainerName.LightGbmBinary || nodeName == TrainerName.LightGbmMulti || nodeName == TrainerName.LightGbmRegression)
-                    {
-                        sb.Append("using Microsoft.ML.LightGBM;");
-                        sb.Append("\r\n");
-                    }
-                }
-            }
-
-            foreach (var node in transformNodes)
-            {
-                if (Enum.TryParse<EstimatorName>(node.Name, out EstimatorName nodeName))
-                {
-                    if (nodeName == EstimatorName.OneHotEncoding || nodeName == EstimatorName.OneHotHashEncoding)
-                    {
-                        sb.Append("using Microsoft.ML.Transforms.Categorical;");
-                        sb.Append("\r\n");
-                    }
-                    if (nodeName == EstimatorName.TypeConverting)
-                    {
-                        sb.Append("using Microsoft.ML.Transforms.Conversions;");
-                        sb.Append("\r\n");
-                    }
-                }
-            }
-            return sb.ToString();
-        }
-
         private static string Normalize(string inputColumn)
         {
             //check if first character is int

src/mlnet/CodeGenerator/TrainerGeneratorBase.cs

@@ -25,6 +25,7 @@ internal abstract class TrainerGeneratorBase : ITrainerGenerator
         internal abstract string OptionsName { get; }
         internal abstract string MethodName { get; }
         internal abstract IDictionary<string, string> NamedParameters { get; }
+        internal abstract string Usings { get; }
 
         /// <summary>
         /// Generates an instance of TrainerGenerator
@@ -136,5 +137,12 @@ public string GenerateTrainer()
             return sb.ToString();
         }
 
+        public string GenerateUsings()
+        {
+            if (hasAdvancedSettings)
+                return Usings;
+
+            return null;
+        }
     }
 }

src/mlnet/CodeGenerator/TrainerGeneratorFactory.cs

@@ -12,6 +12,7 @@ namespace Microsoft.ML.CLI
     internal interface ITrainerGenerator
     {
         string GenerateTrainer();
+        string GenerateUsings();
     }
     internal static class TrainerGeneratorFactory
     {
@@ -33,30 +34,31 @@ internal static ITrainerGenerator GetInstance(Pipeline pipeline)
                     case TrainerName.AveragedPerceptronBinary:
                         return new AveragedPerceptron(node);
                     case TrainerName.FastForestBinary:
+                        return new FastForestClassification(node);
                     case TrainerName.FastForestRegression:
-                        return new FastForest(node);
+                        return new FastForestRegression(node);
                     case TrainerName.FastTreeBinary:
+                        return new FastTreeClassification(node);
                     case TrainerName.FastTreeRegression:
-                        return new FastTree(node);
+                        return new FastTreeRegression(node);
                     case TrainerName.FastTreeTweedieRegression:
                         return new FastTreeTweedie(node);
-
                     case TrainerName.LinearSvmBinary:
                         return new LinearSvm(node);
                     case TrainerName.LogisticRegressionBinary:
-                    case TrainerName.LogisticRegressionMulti:
-                        return new LogisticRegression(node);
+                        return new LogisticRegressionBinary(node);
                     case TrainerName.OnlineGradientDescentRegression:
                         return new OnlineGradientDescentRegression(node);
                     case TrainerName.OrdinaryLeastSquaresRegression:
                         return new OrdinaryLeastSquaresRegression(node);
                     case TrainerName.PoissonRegression:
                         return new PoissonRegression(node);
                     case TrainerName.SdcaBinary:
-                    case TrainerName.SdcaMulti:
-                        return new StochasticDualCoordinateAscent(node);
+                        return new StochasticDualCoordinateAscentBinary(node);
+                    case TrainerName.SdcaRegression:
+                        return new StochasticDualCoordinateAscentRegression(node);
                     case TrainerName.StochasticGradientDescentBinary:
-                        return new StochasticGradientDescent(node);
+                        return new StochasticGradientDescentClassification(node);
                     case TrainerName.SymSgdBinary:
                         return new SymbolicStochasticGradientDescent(node);
                     default: