Test NaN replacement and fix build for core30 and x86

Author: wschin

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

@@ -103,9 +103,34 @@
       },
       {
         "input": [
-          "F1",
           "F21"
         ],
+        "output": [
+          "F22"
+        ],
+        "name": "Imputer",
+        "opType": "Imputer",
+        "attribute": [
+          {
+            "name": "replaced_value_float",
+            "f": "NaN",
+            "type": "FLOAT"
+          },
+          {
+            "name": "imputed_value_floats",
+            "floats": [
+              0
+            ],
+            "type": "FLOATS"
+          }
+        ],
+        "domain": "ai.onnx.ml"
+      },
+      {
+        "input": [
+          "F1",
+          "F22"
+        ],
         "output": [
           "Features"
         ],
@@ -334,10 +359,10 @@
       },
       {
         "input": [
-          "F21"
+          "F22"
         ],
         "output": [
-          "F22"
+          "F23"
         ],
         "name": "Identity1",
         "opType": "Identity"
@@ -478,7 +503,7 @@
         }
       },
       {
-        "name": "F22",
+        "name": "F23",
         "type": {
           "tensorType": {
             "elemType": "FLOAT",
@@ -605,6 +630,24 @@
           }
         }
       },
+      {
+        "name": "F22",
+        "type": {
+          "tensorType": {
+            "elemType": "FLOAT",
+            "shape": {
+              "dim": [
+                {
+                  "dimValue": "1"
+                },
+                {
+                  "dimValue": "10"
+                }
+              ]
+            }
+          }
+        }
+      },
       {
         "name": "Features",
         "type": {

test/Microsoft.ML.Tests/OnnxConversionTest.cs

@@ -74,7 +74,7 @@ public void SimpleEndToEndOnnxConversionTest()
                 var onnxResult = onnxTransformer.Transform(data);
 
                 // Step 4: Compare ONNX and ML.NET results.
-                CompareSelectedR4ScalarColumns("Score", "Score0", transformedData, onnxResult, 2);
+                CompareSelectedR4ScalarColumns("Score", "Score0", transformedData, onnxResult, 1);
             }
 
             // 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
@@ -180,6 +180,7 @@ public void KeyToVectorWithBagOnnxConversionTest()
                 separatorChar: '\t');
 
             var pipeline = mlContext.Transforms.Categorical.OneHotEncoding("F2", "F2", Transforms.Categorical.OneHotEncodingTransformer.OutputKind.Bag)
+            .Append(mlContext.Transforms.ReplaceMissingValues(new MissingValueReplacingTransformer.ColumnInfo("F2")))
             .Append(mlContext.Transforms.Concatenate("Features", "F1", "F2"))
             .Append(mlContext.BinaryClassification.Trainers.FastTree(labelColumn: "Label", featureColumn: "Features", numLeaves: 2, numTrees: 1, minDatapointsInLeaves: 2));
 
@@ -299,7 +300,7 @@ public void LogisticRegressionSaveModelToOnnxTest()
             Done();
         }
 
-        [Fact]
+        [ConditionalFact(typeof(Environment), nameof(Environment.Is64BitProcess))] // LightGBM is 64-bit only
         public void LightGbmBinaryClassificationOnnxConversionTest()
         {
             // Step 1: Create and train a ML.NET pipeline.