PR feedback.

Author: codemzs

src/Microsoft.ML.TimeSeries/AdaptiveSingularSpectrumSequenceModeler.cs

@@ -1609,13 +1609,22 @@ internal static void ComputeForecastIntervals(ref SsaForecastResult forecast, Si
 
             public void Train(IDataView dataView, string inputColumnName) => Train(new RoleMappedData(dataView, null, inputColumnName));
 
-            public IEnumerable<float> Forecast(int horizon)
+            public float[] Forecast(int horizon)
             {
                 ForecastResultBase<float> result = null;
                 Forecast(ref result, horizon);
-                var values = result.PointForecast.GetValues().ToArray();
-                foreach(var value in values)
-                    yield return value;
+                return result.PointForecast.GetValues().ToArray();
+            }
+
+            public void ForecastWithConfidenceIntervals(int horizon, out float[] forecast, out float[] confidenceIntervalLowerBounds, out float[] confidenceIntervalUpperBounds, float confidenceLevel = 0.95f)
+            {
+                ForecastResultBase<float> result = null;
+                Forecast(ref result, horizon);
+                SsaForecastResult ssaResult = (SsaForecastResult)result;
+                ComputeForecastIntervals(ref ssaResult, confidenceLevel);
+                forecast = result.PointForecast.GetValues().ToArray();
+                confidenceIntervalLowerBounds = ssaResult.LowerBound.GetValues().ToArray();
+                confidenceIntervalUpperBounds = ssaResult.UpperBound.GetValues().ToArray();
             }
 
             public void Update(IDataView dataView, string inputColumnName)
@@ -1660,7 +1669,7 @@ public void Update(IDataView dataView, string inputColumnName)
         /// </summary>
         /// <param name="horizon">Number of values to forecast.</param>
         /// <returns>Forecasted values.</returns>
-        public IEnumerable<float> Forecast(int horizon) => _modeler.Forecast(horizon);
+        public float[] Forecast(int horizon) => _modeler.Forecast(horizon);
 
         /// <summary>
         /// For saving a model into a repository.
@@ -1673,5 +1682,8 @@ public void Save(ModelSaveContext ctx)
             ctx.Writer.Write(_inputColumnName);
             ctx.SaveModel(_modeler, "ForecastWrapper");
         }
+
+        public void ForecastWithConfidenceIntervals(int horizon, out float[] forecast, out float[] confidenceIntervalLowerBounds, out float[] confidenceIntervalUpperBounds, float confidenceLevel = 0.95f) =>
+            _modeler.ForecastWithConfidenceIntervals(horizon, out forecast, out confidenceIntervalLowerBounds, out confidenceIntervalUpperBounds, confidenceLevel);
     }
 }

src/Microsoft.ML.TimeSeries/Forecast.cs

@@ -2,7 +2,6 @@
 // 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 System.IO;
 using Microsoft.ML.Data;
 using static Microsoft.ML.Transforms.TimeSeries.AdaptiveSingularSpectrumSequenceForecastingModeler;
@@ -13,7 +12,7 @@ namespace Microsoft.ML.TimeSeries
     /// Interface for forecasting models.
     /// </summary>
     /// <typeparam name="T">The type of values that are forecasted.</typeparam>
-    public interface ICanForecast<out T> : ICanSaveModel
+    public interface ICanForecast<T> : ICanSaveModel
     {
         /// <summary>
         /// Train a forecasting model from an <see cref="IDataView"/>.
@@ -33,7 +32,19 @@ public interface ICanForecast<out T> : ICanSaveModel
         /// </summary>
         /// <param name="horizon">Number of values to forecast.</param>
         /// <returns>Forecasted values.</returns>
-        IEnumerable<T> Forecast(int horizon);
+        T[] Forecast(int horizon);
+
+        /// <summary>
+        /// Perform forecasting until a particular <paramref name="horizon"/> and also computes confidence intervals.
+        /// For confidence intervals to be computed the model must be trained with <see cref="AdaptiveSingularSpectrumSequenceModeler.ShouldComputeForecastIntervals"/>
+        /// set to true.
+        /// </summary>
+        /// <param name="horizon">Number of values to forecast.</param>
+        /// <param name="forecast">Forecasted values</param>
+        /// <param name="confidenceIntervalLowerBounds">Lower bound confidence intervals of forecasted values.</param>
+        /// <param name="confidenceIntervalUpperBounds">Upper bound confidence intervals of forecasted values.</param>
+        /// <param name="confidenceLevel">Confidence level.</param>
+        void ForecastWithConfidenceIntervals(int horizon, out T[] forecast, out float[] confidenceIntervalLowerBounds, out float[] confidenceIntervalUpperBounds, float confidenceLevel = 0.95f);
     }
 
     public static class ForecastExtensions

test/Microsoft.ML.TimeSeries.Tests/TimeSeriesDirectApi.cs

@@ -336,7 +336,7 @@ public void AnomalyDetectionWithSrCnn()
             }
         }
 
-        [ConditionalFact(typeof(BaseTestBaseline), nameof(BaseTestBaseline.LessThanNetCore30OrNotNetCore))]
+        [Fact]
         public void Forecasting()
         {
             const int SeasonalitySize = 10;
@@ -362,7 +362,7 @@ public void Forecasting()
             var forecast = model.Forecast(5);
 
             // Update with new observations.
-            model.Update(dataView, "Value");
+            model.Update(dataView);
 
             // Checkpoint.
             ml.Model.SaveForecastingModel(model, "model.zip");
@@ -380,5 +380,58 @@ public void Forecasting()
             // already in memory model should be the same.
             Assert.Equal(forecast, forecastCopy);
         }
+
+        [Fact]
+        public void ForecastingWithConfidenceInterval()
+        {
+            const int SeasonalitySize = 10;
+            const int NumberOfSeasonsInTraining = 5;
+
+            List<Data> data = new List<Data>();
+
+            var ml = new MLContext(seed: 1);
+            var dataView = ml.Data.LoadFromEnumerable<Data>(data);
+
+            for (int j = 0; j < NumberOfSeasonsInTraining; j++)
+                for (int i = 0; i < SeasonalitySize; i++)
+                    data.Add(new Data(i));
+
+            // Create forecasting model.
+            var model = new AdaptiveSingularSpectrumSequenceForecastingModeler(ml, "Value", data.Count, SeasonalitySize + 1, SeasonalitySize,
+                1, AdaptiveSingularSpectrumSequenceForecastingModeler.RankSelectionMethod.Exact, null, SeasonalitySize / 2, shouldComputeForecastIntervals: true, false);
+
+            // Train.
+            model.Train(dataView);
+
+            // Forecast.
+            float[] forecast;
+            float[] lowConfInterval;
+            float[] upperConfInterval;
+            model.ForecastWithConfidenceIntervals(5, out forecast, out lowConfInterval, out upperConfInterval);
+
+            // Update with new observations.
+            model.Update(dataView);
+
+            // Checkpoint.
+            ml.Model.SaveForecastingModel(model, "model.zip");
+
+            // Load the checkpointed model from disk.
+            var modelCopy = ml.Model.LoadForecastingModel<float>("model.zip");
+
+            // Forecast with the checkpointed model loaded from disk.
+            float[] forecastCopy;
+            float[] lowConfIntervalCopy;
+            float[] upperConfIntervalCopy;
+            modelCopy.ForecastWithConfidenceIntervals(5, out forecastCopy, out lowConfIntervalCopy, out upperConfIntervalCopy);
+
+            // Forecast with the original model(that was checkpointed to disk).
+            model.ForecastWithConfidenceIntervals(5, out forecast, out lowConfInterval, out upperConfInterval);
+
+            // Both the forecasted values from model loaded from disk and 
+            // already in memory model should be the same.
+            Assert.Equal(forecast, forecastCopy);
+            Assert.Equal(lowConfInterval, lowConfIntervalCopy);
+            Assert.Equal(upperConfInterval, upperConfIntervalCopy);
+        }
     }
 }