Fix for ColumnConcatenatingTransformer (#4861)

* using concat for columnconcatenation

* resolving comments

Author: Lynx1820

src/Microsoft.ML.Data/Transforms/ColumnConcatenatingTransformer.cs

@@ -896,7 +896,7 @@ public void SaveAsOnnx(OnnxContext ctx)
                 Host.CheckValue(ctx, nameof(ctx));
                 Contracts.Assert(CanSaveOnnx(ctx));
 
-                string opType = "FeatureVectorizer";
+                string opType = "Concat";
                 for (int iinfo = 0; iinfo < _columns.Length; ++iinfo)
                 {
                     var colInfo = _parent._columns[iinfo];
@@ -926,9 +926,9 @@ public void SaveAsOnnx(OnnxContext ctx)
                     }
 
                     var node = ctx.CreateNode(opType, inputList.Select(t => t.Key),
-                        new[] { ctx.AddIntermediateVariable(outColType, outName) }, ctx.GetNodeName(opType));
+                        new[] { ctx.AddIntermediateVariable(outColType, outName) }, ctx.GetNodeName(opType), "");
 
-                    node.AddAttribute("inputdimensions", inputList.Select(x => x.Value));
+                    node.AddAttribute("axis", 1);
                 }
             }
         }

test/BaselineOutput/Common/Onnx/BinaryClassification/BreastCancer/ExcludeVariablesInOnnxConversion.txt

@@ -179,19 +179,15 @@
         "output": [
           "Features"
         ],
-        "name": "FeatureVectorizer",
-        "opType": "FeatureVectorizer",
+        "name": "Concat",
+        "opType": "Concat",
         "attribute": [
           {
-            "name": "inputdimensions",
-            "ints": [
-              "1",
-              "10"
-            ],
-            "type": "INTS"
+            "name": "axis",
+            "i": "1",
+            "type": "INT"
           }
-        ],
-        "domain": "ai.onnx.ml"
+        ]
       },
       {
         "input": [

test/BaselineOutput/Common/Onnx/BinaryClassification/BreastCancer/ModelWithLessIO.txt

@@ -128,19 +128,15 @@
         "output": [
           "Features"
         ],
-        "name": "FeatureVectorizer",
-        "opType": "FeatureVectorizer",
+        "name": "Concat",
+        "opType": "Concat",
         "attribute": [
           {
-            "name": "inputdimensions",
-            "ints": [
-              "8",
-              "9"
-            ],
-            "type": "INTS"
+            "name": "axis",
+            "i": "1",
+            "type": "INT"
           }
-        ],
-        "domain": "ai.onnx.ml"
+        ]
       },
       {
         "input": [
@@ -677,27 +673,27 @@
             "floats": [
               -0.9850374,
               -1,
-              -0.42857143,
+              -0.428571433,
               0.05882353,
               0.9655172,
-              0.47826087,
-              7E-45,
+              0.478260875,
+              7.006492E-45,
               0.9354839,
               -0.837172,
-              -0.89662564,
+              -0.896625638,
               -0.3455931,
-              0.22312601,
+              0.223126009,
               0.8040303,
               0.60825175,
               -0.06932944,
-              -0.40204307,
+              -0.402043074,
               -0.7417274,
-              -0.40843493,
+              -0.408434927,
               0.7105746,
               0.1875386,
               0.7631735,
-              0.70617324,
-              0.62590647,
+              0.706173241,
+              0.625906467,
               -0.35968104
             ],
             "type": "FLOATS"

test/BaselineOutput/Common/Onnx/BinaryClassification/BreastCancer/OneHotBagPipeline.txt

@@ -179,19 +179,15 @@
         "output": [
           "Features"
         ],
-        "name": "FeatureVectorizer",
-        "opType": "FeatureVectorizer",
+        "name": "Concat",
+        "opType": "Concat",
         "attribute": [
           {
-            "name": "inputdimensions",
-            "ints": [
-              "1",
-              "10"
-            ],
-            "type": "INTS"
+            "name": "axis",
+            "i": "1",
+            "type": "INT"
           }
-        ],
-        "domain": "ai.onnx.ml"
+        ]
       },
       {
         "input": [

test/Microsoft.ML.Tests/OnnxConversionTest.cs

@@ -563,6 +563,7 @@ public void KeyToVectorWithBagOnnxConversionTest()
             .Append(mlContext.BinaryClassification.Trainers.FastTree(labelColumnName: "Label", featureColumnName: "Features", numberOfLeaves: 2, numberOfTrees: 1, minimumExampleCountPerLeaf: 2));
 
             var model = pipeline.Fit(data);
+            var transformedData = model.Transform(data);
             var onnxModel = mlContext.Model.ConvertToOnnxProtobuf(model, data);
 
             // Check ONNX model's text format. We save the produced ONNX model as a text file and compare it against
@@ -572,8 +573,19 @@ public void KeyToVectorWithBagOnnxConversionTest()
             var onnxTextName = "OneHotBagPipeline.txt";
             var onnxFileName = "OneHotBagPipeline.onnx";
             var onnxTextPath = GetOutputPath(subDir, onnxTextName);
-            var onnxFilePath = GetOutputPath(subDir, onnxFileName);
-            SaveOnnxModel(onnxModel, onnxFilePath, onnxTextPath);
+            var onnxModelPath = GetOutputPath(subDir, onnxFileName);
+            SaveOnnxModel(onnxModel, onnxModelPath, onnxTextPath);
+            // Compare results produced by ML.NET and ONNX's runtime.
+            if (IsOnnxRuntimeSupported())
+            {
+                // Evaluate the saved ONNX model using the data used to train the ML.NET pipeline.
+                var onnxEstimator = mlContext.Transforms.ApplyOnnxModel(onnxModelPath);
+                var onnxTransformer = onnxEstimator.Fit(data);
+                var onnxResult = onnxTransformer.Transform(data);
+                CompareSelectedColumns<float>("Score", "Score", transformedData, onnxResult);
+                CompareSelectedColumns<float>("Probability", "Probability", transformedData, onnxResult);
+                CompareSelectedColumns<bool>("PredictedLabel", "PredictedLabel", transformedData, onnxResult);
+            }
             CheckEquality(subDir, onnxTextName);
             Done();
         }
@@ -905,8 +917,18 @@ public void RemoveVariablesInPipelineTest()
                 var onnxTextName = "ExcludeVariablesInOnnxConversion.txt";
                 var onnxFileName = "ExcludeVariablesInOnnxConversion.onnx";
                 var onnxTextPath = GetOutputPath(subDir, onnxTextName);
-                var onnxFilePath = GetOutputPath(subDir, onnxFileName);
-                SaveOnnxModel(onnxModel, onnxFilePath, onnxTextPath);
+                var onnxModelPath = GetOutputPath(subDir, onnxFileName);
+                SaveOnnxModel(onnxModel, onnxModelPath, onnxTextPath);
+                if (IsOnnxRuntimeSupported())
+                {
+                    // Evaluate the saved ONNX model using the data used to train the ML.NET pipeline.
+                    var onnxEstimator = mlContext.Transforms.ApplyOnnxModel(onnxModelPath);
+                    var onnxTransformer = onnxEstimator.Fit(data);
+                    var onnxResult = onnxTransformer.Transform(data);
+                    CompareSelectedColumns<float>("Score", "Score", transformedData, onnxResult);
+                    CompareSelectedColumns<float>("Probability", "Probability", transformedData, onnxResult);
+                    CompareSelectedColumns<bool>("PredictedLabel", "PredictedLabel", transformedData, onnxResult);
+                }
                 CheckEquality(subDir, onnxTextName, digitsOfPrecision: 3);
             }
             Done();