Internalization of more model saving infrastructure.

Author: TomFinley

src/Microsoft.ML.Data/Dirty/PredictorUtils.cs

@@ -64,7 +64,7 @@ public static void SaveText(IChannel ch, IPredictor predictor, RoleMappedSchema
         }
 
         /// <summary>
-        /// Save the model in binary format (if it can save itself)
+        /// Save the model in binary format (if it can save itself).
         /// </summary>
         public static void SaveBinary(IChannel ch, IPredictor predictor, BinaryWriter writer)
         {

src/Microsoft.ML.Data/EntryPoints/InputBase.cs

@@ -6,6 +6,7 @@
 using System.Collections.Generic;
 using Microsoft.Data.DataView;
 using Microsoft.ML.CommandLine;
+using Microsoft.ML.Core.Data;
 using Microsoft.ML.Data;
 using Microsoft.ML.Data.IO;
 using Microsoft.ML.Internal.Calibration;
@@ -18,11 +19,17 @@ namespace Microsoft.ML.EntryPoints
     [TlcModule.EntryPointKind(typeof(CommonInputs.ITransformInput))]
     public abstract class TransformInputBase
     {
+        /// <summary>
+        /// The input dataset. Used only in entry-point methods, since the normal API mechanism for feeding in a dataset to
+        /// create an <see cref="ITransformer"/> is to use the <see cref="IEstimator{TTransformer}.Fit(IDataView)"/> method.
+        /// </summary>
+        [BestFriend]
         [Argument(ArgumentType.Required, HelpText = "Input dataset", Visibility = ArgumentAttribute.VisibilityType.EntryPointsOnly, SortOrder = 1)]
-        public IDataView Data;
+        internal IDataView Data;
     }
 
-    public enum CachingOptions
+    [BestFriend]
+    internal enum CachingOptions
     {
         Auto,
         Memory,
@@ -37,10 +44,13 @@ public enum CachingOptions
     public abstract class LearnerInputBase
     {
         /// <summary>
-        /// The data to be used for training.
+        /// The data to be used for training. Used only in entry-points, since in the API the expected mechanism is
+        /// that the user iwll use the <see cref="IEstimator{TTransformer}.Fit(IDataView)"/> or some other train
+        /// method.
         /// </summary>
+        [BestFriend]
         [Argument(ArgumentType.Required, ShortName = "data", HelpText = "The data to be used for training", SortOrder = 1, Visibility = ArgumentAttribute.VisibilityType.EntryPointsOnly)]
-        public IDataView TrainingData;
+        internal IDataView TrainingData;
 
         /// <summary>
         /// Column to use for features.
@@ -49,16 +59,20 @@ public abstract class LearnerInputBase
         public string FeatureColumn = DefaultColumnNames.Features;
 
         /// <summary>
-        /// Normalize option for the feature column.
+        /// Normalize option for the feature column. Used only in entry-points, since in the API the user is expected to do this themselves.
         /// </summary>
+        [BestFriend]
         [Argument(ArgumentType.AtMostOnce, HelpText = "Normalize option for the feature column", ShortName = "norm", SortOrder = 5, Visibility = ArgumentAttribute.VisibilityType.EntryPointsOnly)]
-        public NormalizeOption NormalizeFeatures = NormalizeOption.Auto;
+        internal NormalizeOption NormalizeFeatures = NormalizeOption.Auto;
 
         /// <summary>
-        /// Whether learner should cache input training data.
+        /// Whether learner should cache input training data. Used only in entry-points, since the intended API mechanism
+        /// is that the user will use the <see cref="DataOperationsCatalog.Cache(IDataView, string[])"/> or other method
+        /// like <see cref="EstimatorChain{TLastTransformer}.AppendCacheCheckpoint(IHostEnvironment)"/>.
         /// </summary>
+        [BestFriend]
         [Argument(ArgumentType.LastOccurenceWins, HelpText = "Whether learner should cache input training data", ShortName = "cache", SortOrder = 6, Visibility = ArgumentAttribute.VisibilityType.EntryPointsOnly)]
-        public CachingOptions Caching = CachingOptions.Auto;
+        internal CachingOptions Caching = CachingOptions.Auto;
     }
 
     /// <summary>

src/Microsoft.ML.Data/Model/Onnx/OnnxNode.cs

@@ -13,7 +13,8 @@ namespace Microsoft.ML.Model.Onnx
     /// That method creates a with inputs and outputs, but this object can modify the node further
     /// by adding attributes (in ONNX parlance, attributes are more or less constant parameterizations).
     /// </summary>
-    public abstract class OnnxNode
+    [BestFriend]
+    internal abstract class OnnxNode
     {
         public abstract void AddAttribute(string argName, double value);
         public abstract void AddAttribute(string argName, long value);

src/Microsoft.ML.Data/Model/Repository.cs

@@ -24,9 +24,11 @@ public interface ICanSaveModel
     }
 
     /// <summary>
-    /// For saving to a single stream.
+    /// For saving to a single stream. Note that this interface is mostly deprecated in favor of
+    /// saving more comprehensive and composable "model" objects, via <see cref="ICanSaveModel"/>.
     /// </summary>
-    public interface ICanSaveInBinaryFormat
+    [BestFriend]
+    internal interface ICanSaveInBinaryFormat
     {
         void SaveAsBinary(BinaryWriter writer);
     }

src/Microsoft.ML.Data/Prediction/Calibrator.cs

@@ -1070,7 +1070,7 @@ private static NaiveCalibrator Create(IHostEnvironment env, ModelLoadContext ctx
             return new NaiveCalibrator(env, ctx);
         }
 
-        public void SaveAsBinary(BinaryWriter writer)
+        void ICanSaveInBinaryFormat.SaveAsBinary(BinaryWriter writer)
         {
             ModelSaveContext.Save(writer, SaveCore);
         }
@@ -1717,7 +1717,7 @@ private static PavCalibrator Create(IHostEnvironment env, ModelLoadContext ctx)
             return new PavCalibrator(env, ctx);
         }
 
-        public void SaveAsBinary(BinaryWriter writer)
+        void ICanSaveInBinaryFormat.SaveAsBinary(BinaryWriter writer)
         {
             ModelSaveContext.Save(writer, SaveCore);
         }