introduce IUnsupervisedLearningWithWeights

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

@@ -69,6 +69,16 @@ public abstract class LearnerInputBaseWithWeight : LearnerInputBaseWithLabel
         public Optional<string> WeightColumn = Optional<string>.Implicit(DefaultColumnNames.Weight);
     }
 
+    /// <summary>
+    /// The base class for all unsupervised learner inputs that support a weight column.
+    /// </summary>
+    [TlcModule.EntryPointKind(typeof(CommonInputs.IUnsupervisedTrainerWithWeight))]
+    public abstract class UnsupervisedLearnerInputBaseWithWeight : LearnerInputBase
+    {
+        [Argument(ArgumentType.AtMostOnce, HelpText = "Column to use for example weight", ShortName = "weight", SortOrder = 4, Visibility = ArgumentAttribute.VisibilityType.EntryPointsOnly)]
+        public Optional<string> WeightColumn = Optional<string>.Implicit(DefaultColumnNames.Weight);
+    }
+
     /// <summary>
     /// The base class for all evaluators inputs.
     /// </summary>
@@ -224,6 +234,14 @@ public interface ITrainerInputWithLabel : ITrainerInput
             string LabelColumn { get; }
         }
 
+        /// <summary>
+        /// Interface that all API trainer input classes will implement.
+        /// </summary>
+        public interface IUnsupervisedTrainerWithWeight : ITrainerInput
+        {
+            Optional<string> WeightColumn { get; }
+        }
+
         /// <summary>
         /// Interface that all API trainer input classes will implement.
         /// </summary>

src/Microsoft.ML.KMeansClustering/KMeansPlusPlusTrainer.cs

@@ -44,7 +44,7 @@ public enum InitAlgorithm
             KMeansParallel = 2
         }
 
-        public class Arguments : LearnerInputBase
+        public class Arguments : UnsupervisedLearnerInputBaseWithWeight
         {
             [Argument(ArgumentType.AtMostOnce, HelpText = "The number of clusters", SortOrder = 50)]
             [TGUI(SuggestedSweeps = "5,10,20,40")]
@@ -162,7 +162,7 @@ private void TrainCore(IChannel ch, RoleMappedData data)
             long missingFeatureCount;
             long totalTrainingInstances;
 
-            var cursorFactory = new FeatureFloatVectorCursor.Factory(data, CursOpt.Features | CursOpt.Id);
+            var cursorFactory = new FeatureFloatVectorCursor.Factory(data, CursOpt.Features | CursOpt.Id | CursOpt.Weight);
             // REVIEW: It would be nice to extract these out into subcomponents in the future. We should
             // revisit and even consider breaking these all into individual KMeans-flavored trainers, they
             // all produce a valid set of output centroids with various trade-offs in runtime (with perhaps 
@@ -234,7 +234,8 @@ public static CommonOutputs.ClusteringOutput TrainKMeans(IHostEnvironment env, A
             EntryPointUtils.CheckInputArgs(host, input);
 
             return LearnerEntryPointsUtils.Train<Arguments, CommonOutputs.ClusteringOutput>(host, input,
-                () => new KMeansPlusPlusTrainer(host, input));
+                () => new KMeansPlusPlusTrainer(host, input),
+                getWeight: () => LearnerEntryPointsUtils.FindColumn(host, input.TrainingData.Schema, input.WeightColumn));
         }
     }
 

src/Microsoft.ML.PCA/PcaTrainer.cs

@@ -49,7 +49,7 @@ public sealed class RandomizedPcaTrainer : TrainerBase<RoleMappedData, PcaPredic
         internal const string Summary = "This algorithm trains an approximate PCA using Randomized SVD algorithm. "
             + "This PCA can be made into Kernel PCA by using Random Fourier Features transform.";
 
-        public class Arguments : LearnerInputBase
+        public class Arguments : UnsupervisedLearnerInputBaseWithWeight
         {
             [Argument(ArgumentType.AtMostOnce, HelpText = "The number of components in the PCA", ShortName = "k", SortOrder = 50)]
             [TGUI(SuggestedSweeps = "10,20,40,80")]
@@ -67,9 +67,6 @@ public class Arguments : LearnerInputBase
 
             [Argument(ArgumentType.AtMostOnce, HelpText = "The seed for random number generation", ShortName = "seed")]
             public int? Seed;
-
-            [Argument(ArgumentType.AtMostOnce, HelpText = "Column to use for example weight", ShortName = "weight", SortOrder = 4, Visibility = ArgumentAttribute.VisibilityType.EntryPointsOnly)]
-            public Optional<string> WeightColumn = Optional<string>.Implicit(DefaultColumnNames.Weight);
         }
 
         private int _dimension;

test/BaselineOutput/Common/EntryPoints/core_manifest.json

@@ -9117,6 +9117,18 @@
           "IsNullable": false,
           "Default": "Features"
         },
+        {
+          "Name": "WeightColumn",
+          "Type": "String",
+          "Desc": "Column to use for example weight",
+          "Aliases": [
+            "weight"
+          ],
+          "Required": false,
+          "SortOrder": 4.0,
+          "IsNullable": false,
+          "Default": "Weight"
+        },
         {
           "Name": "NormalizeFeatures",
           "Type": {
@@ -9253,6 +9265,7 @@
         }
       ],
       "InputKind": [
+        "IUnsupervisedTrainerWithWeight",
         "ITrainerInput"
       ],
       "OutputKind": [
@@ -10875,6 +10888,7 @@
         }
       ],
       "InputKind": [
+        "IUnsupervisedTrainerWithWeight",
         "ITrainerInput"
       ],
       "OutputKind": [
@@ -22577,6 +22591,23 @@
           "Type": "TransformModel"
         }
       ]
+    },
+    {
+      "Kind": "IUnsupervisedTrainerWithWeight",
+      "Settings": [
+        {
+          "Name": "WeightColumn",
+          "Type": "String"
+        },
+        {
+          "Name": "TrainingData",
+          "Type": "DataView"
+        },
+        {
+          "Name": "FeatureColumn",
+          "Type": "String"
+        }
+      ]
     }
   ]
 }

test/Microsoft.ML.Core.Tests/UnitTests/TestEntryPoints.cs

@@ -1918,7 +1918,7 @@ public void EntryPointTrainTestMacroNoTransformInput()
         [Fact]
         public void EntryPointKMeans()
         {
-            TestEntryPointRoutine("Train-Tiny-28x28.txt", "Trainers.KMeansPlusPlusClusterer");
+            TestEntryPointRoutine("Train-Tiny-28x28.txt", "Trainers.KMeansPlusPlusClusterer", "col=Weight:R4:0 col=Features:R4:1-784", ",'InitAlgorithm':'KMeansPlusPlus'");
         }
 
         [Fact]