Fix column purpose for PipelineSweeperMacro

src/Microsoft.ML.PipelineInference/AutoInference.cs

@@ -208,6 +208,7 @@ public sealed class AutoMlMlState : IMlState
             private TransformInference.SuggestedTransform[] _availableTransforms;
             private RecipeInference.SuggestedRecipe.SuggestedLearner[] _availableLearners;
             private DependencyMap _dependencyMapping;
+            private RoleMappedData _dataRoles;
             public IPipelineOptimizer AutoMlEngine { get; set; }
             public PipelinePattern[] BatchCandidates { get; set; }
             public SupportedMetric Metric { get; }
@@ -313,7 +314,7 @@ private void MainLearningLoop(int batchSize, int numOfTrainingRows)
                     var currentBatchSize = batchSize;
                     if (_terminator is IterationTerminator itr)
                         currentBatchSize = Math.Min(itr.RemainingIterations(_history), batchSize);
-                    var candidates = AutoMlEngine.GetNextCandidates(_sortedSampledElements.Values, currentBatchSize);
+                    var candidates = AutoMlEngine.GetNextCandidates(_sortedSampledElements.Values, currentBatchSize, _dataRoles);
 
                     // Break if no candidates returned, means no valid pipeline available.
                     if (candidates.Length == 0)
@@ -370,19 +371,21 @@ private TransformInference.SuggestedTransform[] InferAndFilter(IDataView data, T
                 TransformInference.SuggestedTransform[] existingTransforms = null)
             {
                 // Infer transforms using experts
-                var levelTransforms = TransformInference.InferTransforms(_env, data, args);
+                var levelTransforms = TransformInference.InferTransforms(_env, data, args, _dataRoles);
 
                 // Retain only those transforms inferred which were also passed in.
                 if (existingTransforms != null)
                     return levelTransforms.Where(t => existingTransforms.Any(t2 => t2.Equals(t))).ToArray();
                 return levelTransforms;
             }
 
-            public void InferSearchSpace(int numTransformLevels)
+            public void InferSearchSpace(int numTransformLevels, RoleMappedData dataRoles = null)
             {
                 var learners = RecipeInference.AllowedLearners(_env, TrainerKind).ToArray();
                 if (_requestedLearners != null && _requestedLearners.Length > 0)
                     learners = learners.Where(l => _requestedLearners.Contains(l.LearnerName)).ToArray();
+
+                _dataRoles = dataRoles;
                 ComputeSearchSpace(numTransformLevels, learners, (b, c) => InferAndFilter(b, c));
             }
 
@@ -536,7 +539,21 @@ public PipelinePattern[] GetNextCandidates(int numberOfCandidates)
                 var currentBatchSize = numberOfCandidates;
                 if (_terminator is IterationTerminator itr)
                     currentBatchSize = Math.Min(itr.RemainingIterations(_history), numberOfCandidates);
-                BatchCandidates = AutoMlEngine.GetNextCandidates(_sortedSampledElements.Select(kvp => kvp.Value), currentBatchSize);
+                BatchCandidates = AutoMlEngine.GetNextCandidates(_sortedSampledElements.Select(kvp => kvp.Value), currentBatchSize, _dataRoles);
+
+                using (var ch = _host.Start("Suggested Pipeline"))
+                {
+                    foreach (var pipeline in BatchCandidates)
+                    {
+                        ch.Info($"AutoInference Pipeline Id : {pipeline.UniqueId}");
+                        foreach (var transform in pipeline.Transforms)
+                        {
+                            ch.Info($"AutoInference Transform : {transform.Transform}");
+                        }
+                        ch.Info($"AutoInference Learner : {pipeline.Learner}");
+                    }
+                }
+
                 return BatchCandidates;
             }
 

src/Microsoft.ML.PipelineInference/AutoMlEngines/DefaultsEngine.cs

@@ -5,6 +5,7 @@
 using System;
 using System.Collections.Generic;
 using System.Linq;
+using Microsoft.ML.Runtime.Data;
 using Microsoft.ML.Runtime.EntryPoints;
 using Microsoft.ML.Runtime.PipelineInference;
 
@@ -33,9 +34,10 @@ public DefaultsEngine(IHostEnvironment env, Arguments args)
             _currentLearnerIndex = 0;
         }
 
-        public override PipelinePattern[] GetNextCandidates(IEnumerable<PipelinePattern> history, int numCandidates)
+        public override PipelinePattern[] GetNextCandidates(IEnumerable<PipelinePattern> history, int numCandidates, RoleMappedData dataRoles)
         {
             var candidates = new List<PipelinePattern>();
+            DataRoles = dataRoles;
 
             while (candidates.Count < numCandidates)
             {
@@ -53,7 +55,8 @@ public override PipelinePattern[] GetNextCandidates(IEnumerable<PipelinePattern>
 
                 do
                 {   // Make sure transforms set is valid. Repeat until passes verifier.
-                    pipeline = new PipelinePattern(SampleTransforms(out var transformsBitMask), learner, "", Env);
+                    pipeline = new PipelinePattern(SampleTransforms(out var transformsBitMask), 
+                        learner, "", Env);
                     valid = PipelineVerifier(pipeline, transformsBitMask);
                     count++;
                 } while (!valid && count <= 1000);
@@ -77,7 +80,7 @@ private TransformInference.SuggestedTransform[] SampleTransforms(out long transf
 
             // Add final features concat transform.
             sampledTransforms.AddRange(AutoMlUtils.GetFinalFeatureConcat(Env, FullyTransformedData,
-                DependencyMapping, sampledTransforms.ToArray(), AvailableTransforms));
+                DependencyMapping, sampledTransforms.ToArray(), AvailableTransforms, DataRoles));
 
             return sampledTransforms.ToArray();
         }

src/Microsoft.ML.PipelineInference/AutoMlEngines/RocketEngine.cs

@@ -187,7 +187,7 @@ private TransformInference.SuggestedTransform[] SampleTransforms(RecipeInference
             // cause an error in verification, since it isn't included in the original 
             // dependency mapping (i.e., its level isn't in the dictionary).
             sampledTransforms.AddRange(AutoMlUtils.GetFinalFeatureConcat(Env, FullyTransformedData,
-                DependencyMapping, sampledTransforms.ToArray(), AvailableTransforms));
+                DependencyMapping, sampledTransforms.ToArray(), AvailableTransforms, DataRoles));
             transformsBitMask = mask;
 
             return sampledTransforms.ToArray();
@@ -202,9 +202,10 @@ private RecipeInference.SuggestedRecipe.SuggestedLearner[] GetTopLearners(IEnume
                 .Select(t=>AvailableLearners[t.Index]).ToArray();
         }
 
-        public override PipelinePattern[] GetNextCandidates(IEnumerable<PipelinePattern> history, int numCandidates)
+        public override PipelinePattern[] GetNextCandidates(IEnumerable<PipelinePattern> history, int numCandidates, RoleMappedData dataRoles)
         {
             var prevCandidates = history.ToArray();
+            DataRoles = dataRoles;
 
             switch (_currentStage)
             {
@@ -220,7 +221,7 @@ public override PipelinePattern[] GetNextCandidates(IEnumerable<PipelinePattern>
                         // number of candidates, using second stage logic.
                         UpdateLearners(GetTopLearners(prevCandidates));
                         _currentStage++;
-                        return GetNextCandidates(prevCandidates, numCandidates);
+                        return GetNextCandidates(prevCandidates, numCandidates, DataRoles);
                     }
                     else
                         return GetInitialPipelines(prevCandidates, remainingNum);
@@ -252,9 +253,11 @@ public override PipelinePattern[] GetNextCandidates(IEnumerable<PipelinePattern>
             }
         }
 
-        private PipelinePattern[] GetInitialPipelines(IEnumerable<PipelinePattern> history, int numCandidates) =>
-            _secondaryEngines[_randomInit ? nameof(UniformRandomEngine) : nameof(DefaultsEngine)]
-                .GetNextCandidates(history, numCandidates);
+        private PipelinePattern[] GetInitialPipelines(IEnumerable<PipelinePattern> history, int numCandidates)
+        {
+            var engine = _secondaryEngines[_randomInit ? nameof(UniformRandomEngine) : nameof(DefaultsEngine)];
+            return engine.GetNextCandidates(history, numCandidates, DataRoles);
+        }
 
         private PipelinePattern[] NextCandidates(PipelinePattern[] history, int numCandidates,
             bool defaultHyperParams = false, bool uniformRandomTransforms = false)
@@ -294,8 +297,9 @@ private PipelinePattern[] NextCandidates(PipelinePattern[] history, int numCandi
                 do
                 {   // Make sure transforms set is valid and have not seen pipeline before. 
                     // Repeat until passes or runs out of chances.
-                    pipeline = new PipelinePattern(SampleTransforms(learner, history,
-                        out var transformsBitMask, uniformRandomTransforms), learner, "", Env);
+                    pipeline = new PipelinePattern(
+                        SampleTransforms(learner, history, out var transformsBitMask, uniformRandomTransforms), 
+                        learner, "", Env);
                     hashKey = GetHashKey(transformsBitMask, learner);
                     valid = PipelineVerifier(pipeline, transformsBitMask) && !VisitedPipelines.Contains(hashKey);
                     count++;

src/Microsoft.ML.PipelineInference/AutoMlEngines/UniformRandomEngine.cs

@@ -5,6 +5,7 @@
 using System;
 using System.Collections.Generic;
 using System.Linq;
+using Microsoft.ML.Runtime.Data;
 using Microsoft.ML.Runtime.EntryPoints;
 using Microsoft.ML.Runtime.PipelineInference;
 
@@ -30,8 +31,9 @@ public UniformRandomEngine(IHostEnvironment env)
             : base(env, env.Register("UniformRandomEngine(AutoML)"))
         {}
 
-        public override PipelinePattern[] GetNextCandidates(IEnumerable<PipelinePattern> history, int numberOfCandidates)
+        public override PipelinePattern[] GetNextCandidates(IEnumerable<PipelinePattern> history, int numberOfCandidates, RoleMappedData dataRoles)
         {
+            DataRoles = dataRoles;
             return GetRandomPipelines(numberOfCandidates);
         }
 
@@ -66,7 +68,7 @@ private PipelinePattern[] GetRandomPipelines(int numOfPipelines)
 
                 // Always include features concat transform
                 selectedTransforms.AddRange(AutoMlUtils.GetFinalFeatureConcat(Env, FullyTransformedData,
-                    DependencyMapping, selectedTransforms.ToArray(), AvailableTransforms));
+                    DependencyMapping, selectedTransforms.ToArray(), AvailableTransforms, DataRoles));
 
                 // Compute hash key for checking if we've already seen this pipeline.
                 // However, if we keep missing, don't want to get stuck in infinite loop.

src/Microsoft.ML.PipelineInference/AutoMlUtils.cs

@@ -257,7 +257,7 @@ public static long TransformsToBitmask(TransformInference.SuggestedTransform[] t
         /// (In other words, if there would be nothing for that concatenate transform to do.)
         /// </summary>
         private static TransformInference.SuggestedTransform[] GetFinalFeatureConcat(IHostEnvironment env,
-            IDataView dataSample, int[] excludedColumnIndices, int level, int atomicIdOffset)
+            IDataView dataSample, int[] excludedColumnIndices, int level, int atomicIdOffset, RoleMappedData dataRoles)
         {
             var finalArgs = new TransformInference.Arguments
             {
@@ -266,7 +266,7 @@ private static TransformInference.SuggestedTransform[] GetFinalFeatureConcat(IHo
                 ExcludedColumnIndices = excludedColumnIndices
             };
 
-            var featuresConcatTransforms = TransformInference.InferConcatNumericFeatures(env, dataSample, finalArgs);
+            var featuresConcatTransforms = TransformInference.InferConcatNumericFeatures(env, dataSample, finalArgs, dataRoles);
 
             for (int i = 0; i < featuresConcatTransforms.Length; i++)
             {
@@ -282,7 +282,7 @@ private static TransformInference.SuggestedTransform[] GetFinalFeatureConcat(IHo
         /// </summary>
         public static TransformInference.SuggestedTransform[] GetFinalFeatureConcat(IHostEnvironment env, IDataView data,
             AutoInference.DependencyMap dependencyMapping, TransformInference.SuggestedTransform[] selectedTransforms,
-            TransformInference.SuggestedTransform[] allTransforms)
+            TransformInference.SuggestedTransform[] allTransforms, RoleMappedData dataRoles)
         {
             int level = 1;
             int atomicGroupLimit = 0;
@@ -292,7 +292,7 @@ public static TransformInference.SuggestedTransform[] GetFinalFeatureConcat(IHos
                 atomicGroupLimit = allTransforms.Max(t => t.AtomicGroupId) + 1;
             }
             var excludedColumnIndices = GetExcludedColumnIndices(selectedTransforms, data, dependencyMapping);
-            return GetFinalFeatureConcat(env, data, excludedColumnIndices, level, atomicGroupLimit);
+            return GetFinalFeatureConcat(env, data, excludedColumnIndices, level, atomicGroupLimit, dataRoles);
         }
 
         public static IDataView ApplyTransformSet(IHostEnvironment env, IDataView data, TransformInference.SuggestedTransform[] transforms)

src/Microsoft.ML.PipelineInference/Interfaces/IPipelineOptimizer.cs

@@ -21,7 +21,7 @@ namespace Microsoft.ML.Runtime.PipelineInference
     /// </summary>
     public interface IPipelineOptimizer
     {
-        PipelinePattern[] GetNextCandidates(IEnumerable<PipelinePattern> history, int numberOfCandidates);
+        PipelinePattern[] GetNextCandidates(IEnumerable<PipelinePattern> history, int numberOfCandidates, RoleMappedData dataRoles);
 
         void SetSpace(TransformInference.SuggestedTransform[] availableTransforms,
             RecipeInference.SuggestedRecipe.SuggestedLearner[] availableLearners,
@@ -44,6 +44,7 @@ public abstract class PipelineOptimizerBase : IPipelineOptimizer
         protected IDataView OriginalData;
         protected IDataView FullyTransformedData;
         protected AutoInference.DependencyMap DependencyMapping;
+        protected RoleMappedData DataRoles;
         protected readonly IHostEnvironment Env;
         protected readonly IHost Host;
         protected readonly Dictionary<long, bool> TransformsMaskValidity;
@@ -60,7 +61,7 @@ protected PipelineOptimizerBase(IHostEnvironment env, IHost host)
             ProbUtils = new SweeperProbabilityUtils(host);
         }
 
-        public abstract PipelinePattern[] GetNextCandidates(IEnumerable<PipelinePattern> history, int numberOfCandidates);
+        public abstract PipelinePattern[] GetNextCandidates(IEnumerable<PipelinePattern> history, int numberOfCandidates, RoleMappedData dataRoles);
 
         public virtual void SetSpace(TransformInference.SuggestedTransform[] availableTransforms,
             RecipeInference.SuggestedRecipe.SuggestedLearner[] availableLearners,