Address comments

Author: wschin

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

@@ -13,19 +13,20 @@ namespace Microsoft.ML
     /// </summary>
     public sealed class ModelOperationsCatalog
     {
+        /// <summary>
+        /// This is a best friend because an extension method defined in another assembly needs this field.
+        /// </summary>
+        [BestFriend]
         internal IHostEnvironment Environment { get; }
 
         public ExplainabilityTransforms Explainability { get; }
 
-        public PortabilityTransforms Portability { get; }
-
         internal ModelOperationsCatalog(IHostEnvironment env)
         {
             Contracts.AssertValue(env);
             Environment = env;
 
             Explainability = new ExplainabilityTransforms(this);
-            Portability = new PortabilityTransforms(this);
         }
 
         public abstract class SubCatalogBase
@@ -62,17 +63,6 @@ internal ExplainabilityTransforms(ModelOperationsCatalog owner) : base(owner)
             }
         }
 
-        /// <summary>
-        /// The catalog of model protability operations. Member function of this classes are able to convert the associated object to a protable format,
-        /// so that the fitted pipeline can easily be depolyed to other platforms. Currently, the only supported format is ONNX (https://github.com/onnx/onnx).
-        /// </summary>
-        public sealed class PortabilityTransforms : SubCatalogBase
-        {
-            internal PortabilityTransforms(ModelOperationsCatalog owner) : base(owner)
-            {
-            }
-        }
-
         /// <summary>
         /// Create a prediction engine for one-time prediction.
         /// </summary>

src/Microsoft.ML.Onnx/ProtabilityCatalog.cs renamed to src/Microsoft.ML.Onnx/OnnxExportExtensions.cs

@@ -1,32 +1,37 @@
-using System.Collections.Generic;
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+using System.Collections.Generic;
 using Microsoft.ML.Core.Data;
 using Microsoft.ML.Data;
 using Microsoft.ML.Model.Onnx;
 using Microsoft.ML.UniversalModelFormat.Onnx;
 
 namespace Microsoft.ML
 {
-    public static class ProtabilityCatalog
+    public static class OnnxExportExtensions
     {
         /// <summary>
         /// Convert the specified <see cref="ITransformer"/> to ONNX format. Note that ONNX uses Google's Protobuf so the returned value is a Protobuf object.
         /// </summary>
-        /// <param name="catalog">A field in <see cref="MLContext"/> which this function associated with.</param>
+        /// <param name="catalog">The class that <see cref="ConvertToOnnx(ModelOperationsCatalog, ITransformer, IDataView)"/> attached to.</param>
         /// <param name="transform">The <see cref="ITransformer"/> that will be converted into ONNX format.</param>
         /// <param name="inputData">The input of the specified transform.</param>
-        /// <returns></returns>
-        public static ModelProto ConvertToOnnx(this ModelOperationsCatalog.PortabilityTransforms catalog, ITransformer transform, IDataView inputData)
+        /// <returns>An ONNX model equivalent to the converted ML.NET model.</returns>
+        public static ModelProto ConvertToOnnx(this ModelOperationsCatalog catalog, ITransformer transform, IDataView inputData)
         {
-            var env = new MLContext(seed: 1);
-            var ctx = new OnnxContextImpl(env, "model", "ML.NET", "0", 0, "com.microsoft", OnnxVersion.Stable);
+            var env = catalog.Environment;
+            var ctx = new OnnxContextImpl(env, "model", "ML.NET", "0", 0, "ai.onnx.ml", OnnxVersion.Stable);
             var outputData = transform.Transform(inputData);
             IDataView root = null;
             IDataView sink = null;
             LinkedList<ITransformCanSaveOnnx> transforms = null;
-            using (var ch = (env as IChannelProvider).Start("ONNX conversion"))
+            using (var ch = env.Start("ONNX conversion"))
+            {
                 SaveOnnxCommand.GetPipe(ctx, ch, outputData, out root, out sink, out transforms);
-
-            return SaveOnnxCommand.ConvertTransformListToOnnxModel(ctx, root, sink, transforms, null, null);
+                return SaveOnnxCommand.ConvertTransformListToOnnxModel(ctx, ch, root, sink, transforms, null, null);
+            }
         }
     }
 }

src/Microsoft.ML.Onnx/SaveOnnxCommand.cs

@@ -116,7 +116,7 @@ public override void Run()
 
         internal static void GetPipe(OnnxContextImpl ctx, IChannel ch, IDataView end, out IDataView source, out IDataView trueEnd, out LinkedList<ITransformCanSaveOnnx> transforms)
         {
-            Contracts.AssertValue(end);
+            ch.AssertValue(end);
 
             source = trueEnd = (end as CompositeDataLoader)?.View ?? end;
             IDataTransform transform = source as IDataTransform;
@@ -136,10 +136,10 @@ internal static void GetPipe(OnnxContextImpl ctx, IChannel ch, IDataView end, ou
                 transform = (source = transform.Source) as IDataTransform;
             }
 
-            Contracts.AssertValue(source);
+            ch.AssertValue(source);
         }
 
-        internal static ModelProto ConvertTransformListToOnnxModel(OnnxContextImpl ctx, IDataView inputData, IDataView outputData,
+        internal static ModelProto ConvertTransformListToOnnxModel(OnnxContextImpl ctx, IChannel ch, IDataView inputData, IDataView outputData,
             LinkedList<ITransformCanSaveOnnx> transforms, HashSet<string> inputColumnNamesToDrop=null, HashSet<string> outputColumnNamesToDrop=null)
         {
             inputColumnNamesToDrop = inputColumnNamesToDrop ?? new HashSet<string>();
@@ -158,7 +158,10 @@ internal static ModelProto ConvertTransformListToOnnxModel(OnnxContextImpl ctx,
 
             // Create graph nodes, outputs and intermediate values.
             foreach (var trans in transforms)
+            {
+                ch.Assert(trans.CanSaveOnnx(ctx));
                 trans.SaveAsOnnx(ctx);
+            }
 
             // Add graph outputs.
             for (int i = 0; i < outputData.Schema.Count; ++i)
@@ -255,7 +258,7 @@ private void Run(IChannel ch)
                     nameof(Arguments.LoadPredictor), "We were explicitly told to load the predictor but one was not present.");
             }
 
-            var model = ConvertTransformListToOnnxModel(ctx, source, end, transforms, _inputsToDrop, _outputsToDrop);
+            var model = ConvertTransformListToOnnxModel(ctx, ch, source, end, transforms, _inputsToDrop, _outputsToDrop);
 
             using (var file = Host.CreateOutputFile(_outputModelPath))
             using (var stream = file.CreateWriteStream())

test/BaselineOutput/Common/Onnx/Cluster/BreastCancer/Kmeans.txt

@@ -2,7 +2,7 @@
   "irVersion": "3",
   "producerName": "ML.NET",
   "producerVersion": "##VERSION##",
-  "domain": "com.microsoft",
+  "domain": "ai.onnx.ml",
   "graph": {
     "node": [
       {

test/BaselineOutput/Common/Onnx/Regression/Adult/SimplePipeline.txt

@@ -2,7 +2,7 @@
   "irVersion": "3",
   "producerName": "ML.NET",
   "producerVersion": "##VERSION##",
-  "domain": "com.microsoft",
+  "domain": "ai.onnx.ml",
   "graph": {
     "node": [
       {

test/Microsoft.ML.Tests/OnnxConversionTest.cs

@@ -1,4 +1,5 @@
-using System.IO;
+using System;
+using System.IO;
 using System.Linq;
 using System.Runtime.InteropServices;
 using System.Text.RegularExpressions;
@@ -31,7 +32,7 @@ public OnnxConversionTest(ITestOutputHelper output) : base(output)
         /// <summary>
         /// In this test, we convert a trained <see cref="TransformerChain"/> into ONNX <see cref="UniversalModelFormat.Onnx.ModelProto"/> file and then
         /// call <see cref="OnnxScoringEstimator"/> to evaluate that file. The outputs of <see cref="OnnxScoringEstimator"/> are checked against the original
-        /// ML.NET model's outputs. 
+        /// ML.NET model's outputs.
         /// </summary>
         [Fact]
         public void SimpleEndToEndOnnxConversionTest()
@@ -52,12 +53,12 @@ public void SimpleEndToEndOnnxConversionTest()
             var transformedData = model.Transform(data);
 
             // Step 2: Convert ML.NET model to ONNX format and save it as a file.
-            var onnxModel = mlContext.Model.Portability.ConvertToOnnx(model, data);
+            var onnxModel = mlContext.Model.ConvertToOnnx(model, data);
             var onnxFileName = "model.onnx";
             var onnxModelPath = GetOutputPath(onnxFileName);
             SaveOnnxModel(onnxModel, onnxModelPath, null);
 
-            if (RuntimeInformation.IsOSPlatform(OSPlatform.Windows))
+            if (RuntimeInformation.IsOSPlatform(OSPlatform.Windows) && Environment.Is64BitProcess)
             {
                 // Step 3: Evaluate the saved ONNX model using the data used to train the ML.NET pipeline.
                 string[] inputNames = onnxModel.Graph.Input.Select(valueInfoProto => valueInfoProto.Name).ToArray();
@@ -70,7 +71,8 @@ public void SimpleEndToEndOnnxConversionTest()
                 CompareSelectedR4ScalarColumns("Score", "Score0", transformedData, onnxResult, 2);
             }
 
-            // Step 5: Check ONNX model's text format.
+            // Step 5: Check ONNX model's text format. This test will be not necessary if Step 3 and Step 4 can run on Linux and
+            // Mac to support cross-platform tests.
             var subDir = Path.Combine("..", "..", "BaselineOutput", "Common", "Onnx", "Regression", "Adult");
             var onnxTextName = "SimplePipeline.txt";
             var onnxTextPath = GetOutputPath(subDir, onnxTextName);
@@ -86,11 +88,6 @@ private class BreastCancerFeatureVector
             public float[] Features;
         }
 
-        private void CreateDummyExamplesToMakeComplierHappy()
-        {
-            var dummyExample = new BreastCancerFeatureVector() { Features = null };
-        }
-
         [Fact]
         public void KmeansOnnxConversionTest()
         {
@@ -102,24 +99,24 @@ public void KmeansOnnxConversionTest()
             // Now read the file (remember though, readers are lazy, so the actual reading will happen when the data is accessed).
             var data = mlContext.Data.ReadFromTextFile<BreastCancerFeatureVector>(dataPath,
                 hasHeader: true,
-                separatorChar: '\t' );
+                separatorChar: '\t');
 
             var pipeline = mlContext.Transforms.Normalize("Features").
                 Append(mlContext.Clustering.Trainers.KMeans(features: "Features", advancedSettings: settings =>
                 {
-                      settings.MaxIterations = 1;
-                      settings.K = 4;
-                      settings.NumThreads = 1;
-                      settings.InitAlgorithm = Trainers.KMeans.KMeansPlusPlusTrainer.InitAlgorithm.KMeansPlusPlus;
+                    settings.MaxIterations = 1;
+                    settings.K = 4;
+                    settings.NumThreads = 1;
+                    settings.InitAlgorithm = Trainers.KMeans.KMeansPlusPlusTrainer.InitAlgorithm.KMeansPlusPlus;
                 }));
 
             var model = pipeline.Fit(data);
             var transformedData = model.Transform(data);
 
-            var onnxModel = mlContext.Model.Portability.ConvertToOnnx(model, data);
+            var onnxModel = mlContext.Model.ConvertToOnnx(model, data);
 
             // Compare results produced by ML.NET and ONNX's runtime.
-            if (RuntimeInformation.IsOSPlatform(OSPlatform.Windows))
+            if (RuntimeInformation.IsOSPlatform(OSPlatform.Windows) && Environment.Is64BitProcess)
             {
                 var onnxFileName = "model.onnx";
                 var onnxModelPath = GetOutputPath(onnxFileName);
@@ -134,6 +131,9 @@ public void KmeansOnnxConversionTest()
                 CompareSelectedR4VectorColumns("Score", "Score0", transformedData, onnxResult, 3);
             }
 
+            // Check ONNX model's text format. We save the produced ONNX model as a text file and compare it against
+            // the associated file in ML.NET repo. Such a comparison can be retired if ONNXRuntime ported to ML.NET
+            // can support Linux and Mac.
             var subDir = Path.Combine("..", "..", "BaselineOutput", "Common", "Onnx", "Cluster", "BreastCancer");
             var onnxTextName = "Kmeans.txt";
             var onnxTextPath = GetOutputPath(subDir, onnxTextName);
@@ -142,7 +142,12 @@ public void KmeansOnnxConversionTest()
             Done();
         }
 
-        private void CompareSelectedR4VectorColumns(string leftColumnName, string rightColumnName, IDataView left, IDataView right, int precision=6)
+        private void CreateDummyExamplesToMakeComplierHappy()
+        {
+            var dummyExample = new BreastCancerFeatureVector() { Features = null };
+        }
+
+        private void CompareSelectedR4VectorColumns(string leftColumnName, string rightColumnName, IDataView left, IDataView right, int precision = 6)
         {
             var leftColumnIndex = left.Schema[leftColumnName].Index;
             var rightColumnIndex = right.Schema[rightColumnName].Index;
@@ -166,7 +171,7 @@ private void CompareSelectedR4VectorColumns(string leftColumnName, string rightC
             }
         }
 
-        private void CompareSelectedR4ScalarColumns(string leftColumnName, string rightColumnName, IDataView left, IDataView right, int precision=6)
+        private void CompareSelectedR4ScalarColumns(string leftColumnName, string rightColumnName, IDataView left, IDataView right, int precision = 6)
         {
             var leftColumnIndex = left.Schema[leftColumnName].Index;
             var rightColumnIndex = right.Schema[rightColumnName].Index;