Removing the arguments from the generics definition

Author: sfilipi

src/Microsoft.ML.StandardLearners/Standard/Online/AveragedLinear.cs

@@ -53,13 +53,12 @@ public abstract class AveragedLinearArguments : OnlineLinearArguments
         public Float AveragedTolerance = (Float)1e-2;
     }
 
-    public abstract class AveragedLinearTrainer<TArguments, TTransformer, TModel> : OnlineLinearTrainer<TArguments, TTransformer, TModel>
-        where TArguments : AveragedLinearArguments
+    public abstract class AveragedLinearTrainer<TTransformer, TModel> : OnlineLinearTrainer<TTransformer, TModel>
         where TTransformer : IPredictionTransformer<TModel>
         where TModel : IPredictor
     {
+        protected readonly new AveragedLinearArguments Args;
         protected IScalarOutputLoss LossFunction;
-
         protected Float Gain;
 
         // For computing averaged weights and bias (if needed)
@@ -76,15 +75,18 @@ public abstract class AveragedLinearTrainer<TArguments, TTransformer, TModel> :
         // We'll keep a few things global to prevent garbage collection
         protected int NumNoUpdates;
 
-        protected AveragedLinearTrainer(TArguments args, IHostEnvironment env, string name, SchemaShape.Column label)
+        protected AveragedLinearTrainer(AveragedLinearArguments args, IHostEnvironment env, string name, SchemaShape.Column label)
             : base(args, env, name, label)
         {
             Contracts.CheckUserArg(args.LearningRate > 0, nameof(args.LearningRate), UserErrorPositive);
             Contracts.CheckUserArg(!args.ResetWeightsAfterXExamples.HasValue || args.ResetWeightsAfterXExamples > 0, nameof(args.ResetWeightsAfterXExamples), UserErrorPositive);
+
             // Weights are scaled down by 2 * L2 regularization on each update step, so 0.5 would scale all weights to 0, which is not sensible.
             Contracts.CheckUserArg(0 <= args.L2RegularizerWeight && args.L2RegularizerWeight < 0.5, nameof(args.L2RegularizerWeight), "must be in range [0, 0.5)");
             Contracts.CheckUserArg(args.RecencyGain >= 0, nameof(args.RecencyGain), UserErrorNonNegative);
             Contracts.CheckUserArg(args.AveragedTolerance >= 0, nameof(args.AveragedTolerance), UserErrorNonNegative);
+
+            Args = args;
         }
 
         protected override void InitCore(IChannel ch, int numFeatures, LinearPredictor predictor)

src/Microsoft.ML.StandardLearners/Standard/Online/AveragedPerceptron.cs

@@ -30,13 +30,15 @@ namespace Microsoft.ML.Runtime.Learners
     //     - Feature normalization. By default, rescaling between min and max values for every feature
     //     - Prediction calibration to produce probabilities. Off by default, if on, uses exponential (aka Platt) calibration.
     /// <include file='doc.xml' path='doc/members/member[@name="AP"]/*' />
-    public sealed class AveragedPerceptronTrainer : AveragedLinearTrainer<AveragedPerceptronTrainer.Arguments, BinaryPredictionTransformer<LinearBinaryPredictor> , LinearBinaryPredictor>
+    public sealed class AveragedPerceptronTrainer : AveragedLinearTrainer<BinaryPredictionTransformer<LinearBinaryPredictor> , LinearBinaryPredictor>
     {
         public const string LoadNameValue = "AveragedPerceptron";
         internal const string UserNameValue = "Averaged Perceptron";
         internal const string ShortName = "ap";
         internal const string Summary = "Averaged Perceptron Binary Classifier.";
 
+        internal new readonly Arguments Args;
+
         public class Arguments : AveragedLinearArguments
         {
             [Argument(ArgumentType.Multiple, HelpText = "Loss Function", ShortName = "loss", SortOrder = 50)]
@@ -52,6 +54,7 @@ public class Arguments : AveragedLinearArguments
         public AveragedPerceptronTrainer(IHostEnvironment env, Arguments args)
             : base(args, env, UserNameValue, MakeLabelColumn(args.LabelColumn))
         {
+            Args = args;
             LossFunction = Args.LossFunction.CreateComponent(env);
 
             OutputColumns = new[]

src/Microsoft.ML.StandardLearners/Standard/Online/LinearSvm.cs

@@ -32,7 +32,7 @@ namespace Microsoft.ML.Runtime.Learners
     /// <summary>
     /// Linear SVM that implements PEGASOS for training. See: http://ttic.uchicago.edu/~shai/papers/ShalevSiSr07.pdf
     /// </summary>
-    public sealed class LinearSvm : OnlineLinearTrainer<LinearSvm.Arguments, BinaryPredictionTransformer<LinearBinaryPredictor>, LinearBinaryPredictor>
+    public sealed class LinearSvm : OnlineLinearTrainer<BinaryPredictionTransformer<LinearBinaryPredictor>, LinearBinaryPredictor>
     {
         public const string LoadNameValue = "LinearSVM";
         public const string ShortName = "svm";
@@ -42,6 +42,8 @@ public sealed class LinearSvm : OnlineLinearTrainer<LinearSvm.Arguments, BinaryP
             + "and all the negative examples are on the other. After this mapping, quadratic programming is used to find the separating hyperplane that maximizes the "
             + "margin, i.e., the minimal distance between it and the instances.";
 
+        internal new readonly Arguments Args;
+
         public sealed class Arguments : OnlineLinearArguments
         {
             [Argument(ArgumentType.AtMostOnce, HelpText = "Regularizer constant", ShortName = "lambda", SortOrder = 50)]
@@ -89,6 +91,8 @@ public LinearSvm(IHostEnvironment env, Arguments args)
             Contracts.CheckUserArg(args.Lambda > 0, nameof(args.Lambda), UserErrorPositive);
             Contracts.CheckUserArg(args.BatchSize > 0, nameof(args.BatchSize), UserErrorPositive);
 
+            Args = args;
+
             OutputColumns = new[]
             {
                 new SchemaShape.Column(DefaultColumnNames.Score, SchemaShape.Column.VectorKind.Scalar, NumberType.R4, false),

src/Microsoft.ML.StandardLearners/Standard/Online/OnlineGradientDescent.cs

@@ -29,7 +29,7 @@ namespace Microsoft.ML.Runtime.Learners
     using TPredictor = LinearRegressionPredictor;
 
     /// <include file='doc.xml' path='doc/members/member[@name="OGD"]/*' />
-    public sealed class OnlineGradientDescentTrainer : AveragedLinearTrainer<OnlineGradientDescentTrainer.Arguments, RegressionPredictionTransformer<LinearRegressionPredictor>, LinearRegressionPredictor>
+    public sealed class OnlineGradientDescentTrainer : AveragedLinearTrainer<RegressionPredictionTransformer<LinearRegressionPredictor>, LinearRegressionPredictor>
     {
         internal const string LoadNameValue = "OnlineGradientDescent";
         internal const string UserNameValue = "Stochastic Gradient Descent (Regression)";

src/Microsoft.ML.StandardLearners/Standard/Online/OnlineLinear.cs

@@ -42,12 +42,11 @@ public abstract class OnlineLinearArguments : LearnerInputBaseWithLabel
         public int StreamingCacheSize = 1000000;
     }
 
-    public abstract class OnlineLinearTrainer<TArguments, TTransformer, TModel> : TrainerEstimatorBase<TTransformer, TModel>
+    public abstract class OnlineLinearTrainer<TTransformer, TModel> : TrainerEstimatorBase<TTransformer, TModel>
         where TTransformer : IPredictionTransformer<TModel>
         where TModel : IPredictor
-        where TArguments : OnlineLinearArguments
     {
-        protected readonly TArguments Args;
+        protected readonly OnlineLinearArguments Args;
         protected readonly string Name;
 
         // Initialized by InitCore
@@ -77,7 +76,7 @@ public abstract class OnlineLinearTrainer<TArguments, TTransformer, TModel> : Tr
 
         protected virtual bool NeedCalibration => false;
 
-        protected OnlineLinearTrainer(TArguments args, IHostEnvironment env, string name, SchemaShape.Column label)
+        protected OnlineLinearTrainer(OnlineLinearArguments args, IHostEnvironment env, string name, SchemaShape.Column label)
             : base(Contracts.CheckRef(env, nameof(env)).Register(name), MakeFeatureColumn(args.FeatureColumn), label, MakeWeightColumn(args.InitialWeights))
         {
             Contracts.CheckValue(args, nameof(args));