Merge remote-tracking branch 'origin/master' into singlis/scrub-lightgbm

Author: singlis

.vsts-dotnet-ci.yml

@@ -31,6 +31,15 @@ phases:
   parameters:
     name: Centos
     buildScript: ./build.sh
+    customMatrixes:
+      Build_Debug_Intrinsics:
+        _configuration: Debug-Intrinsics
+        _config_short: DI
+        _includeBenchmarkData: false
+      Build_Release:
+        _configuration: Release
+        _config_short: R
+        _includeBenchmarkData: true
     queue:
       name: Hosted Ubuntu 1604
       container: CentosContainer
@@ -39,6 +48,15 @@ phases:
   parameters:
     name: Ubuntu
     buildScript: ./build.sh
+    customMatrixes:
+      Build_Debug:
+        _configuration: Debug
+        _config_short: D
+        _includeBenchmarkData: false
+      Build_Release_Intrinsics:
+        _configuration: Release-Intrinsics
+        _config_short: RI
+        _includeBenchmarkData: true
     queue:
       name: Hosted Ubuntu 1604
       container: UbuntuContainer

BuildToolsVersion.txt

@@ -1 +1 @@
-3.0.0-preview1-03129-01
+3.0.0-preview1-03721-01

docs/samples/Microsoft.ML.Samples/Dynamic/FeatureContributionCalculationTransform.cs

@@ -44,8 +44,7 @@ public static void Example()
             var outData = featureContributionCalculator.Fit(scoredData).Transform(scoredData);
 
             // Let's extract the weights from the linear model to use as a comparison
-            var weights = new VBuffer<float>();
-            model.Model.GetFeatureWeights(ref weights);
+            var weights = model.Model.Weights;
 
             // Let's now walk through the first ten records and see which feature drove the values the most
             // Get prediction scores and contributions
@@ -63,7 +62,7 @@ public static void Example()
                 var value = row.Features[featureOfInterest];
                 var contribution = row.FeatureContributions[featureOfInterest];
                 var name = data.Schema[featureOfInterest + 1].Name;
-                var weight = weights.GetValues()[featureOfInterest];
+                var weight = weights[featureOfInterest];
 
                 Console.WriteLine("{0:0.00}\t{1:0.00}\t{2}\t{3:0.00}\t{4:0.00}\t{5:0.00}",
                     row.MedianHomeValue,

docs/samples/Microsoft.ML.Samples/Dynamic/Trainers/PriorTrainerSample.cs renamed to docs/samples/Microsoft.ML.Samples/Dynamic/Trainers/BinaryClassification/PriorTrainerSample.cs

@@ -7,22 +7,22 @@ public class PriorTrainer
     {
         public static void Example()
         {
-            // Downloading the dataset from github.com/dotnet/machinelearning.
-            // This will create a sentiment.tsv file in the filesystem.
-            // You can open this file, if you want to see the data. 
-            string dataFile = SamplesUtils.DatasetUtils.DownloadSentimentDataset()[0];
+            // Create a new context for ML.NET operations. It can be used for exception tracking and logging, 
+            // as a catalog of available operations and as the source of randomness.
+            var mlContext = new MLContext();
+
+            // Download and featurize the dataset.
+            var dataFiles = SamplesUtils.DatasetUtils.DownloadSentimentDataset();
+            var trainFile = dataFiles[0];
+            var testFile = dataFiles[1];
 
             // A preview of the data. 
             // Sentiment	SentimentText
             //      0	    " :Erm, thank you. "
             //      1	    ==You're cool==
 
-            // Create a new context for ML.NET operations. It can be used for exception tracking and logging, 
-            // as a catalog of available operations and as the source of randomness.
-            var mlContext = new MLContext();
-
-            // Step 1: Load the data as an IDataView.
-            // First, we define the loader: specify the data columns and where to find them in the text file.
+            // Step 1: Read the data as an IDataView.
+            // First, we define the reader: specify the data columns and where to find them in the text file.
             var loader = mlContext.Data.CreateTextLoader(
                 columns: new[]
                     {
@@ -31,12 +31,9 @@ public static void Example()
                     },
                 hasHeader: true
             );
-            
-            // Load the data
-            var data = loader.Load(dataFile);
 
-            // Split it between training and test data
-            var trainTestData = mlContext.BinaryClassification.TrainTestSplit(data);
+            // Load the data
+            var trainData = loader.Load(trainFile);
 
             // Step 2: Pipeline 
             // Featurize the text column through the FeaturizeText API. 
@@ -47,19 +44,27 @@ public static void Example()
                     .Append(mlContext.BinaryClassification.Trainers.Prior(labelColumnName: "Sentiment"));
 
             // Step 3: Train the pipeline
-            var trainedPipeline = pipeline.Fit(trainTestData.TrainSet);
+            var trainedPipeline = pipeline.Fit(trainData);
 
             // Step 4: Evaluate on the test set
-            var transformedData = trainedPipeline.Transform(trainTestData.TestSet);
+            var transformedData = trainedPipeline.Transform(loader.Load(testFile));
             var evalMetrics = mlContext.BinaryClassification.Evaluate(transformedData, label: "Sentiment");
-
-            // Step 5: Inspect the output
-            Console.WriteLine("Accuracy: " + evalMetrics.Accuracy);
+            SamplesUtils.ConsoleUtils.PrintMetrics(evalMetrics);
 
             // The Prior trainer outputs the proportion of a label in the dataset as the probability of that label.
-            // In this case it means that there is a split of around 64%-36% of positive and negative labels in the dataset.
+            // In this case 'Accuracy: 0.50' means that there is a split of around 50%-50% of positive and negative labels in the test dataset.
             // Expected output:
-            // Accuracy: 0.647058823529412
+
+            //  Accuracy: 0.50
+            //  AUC: 0.50
+            //  F1 Score: 0.67
+            //  Negative Precision: 0.00
+            //  Negative Recall: 0.00
+            //  Positive Precision: 0.50
+            //  Positive Recall: 1.00
+            //  LogLoss: 1.05
+            //  LogLossReduction: -4.89
+            //  Entropy: 1.00
         }
     }
 }

docs/samples/Microsoft.ML.Samples/Dynamic/Trainers/RandomTrainerSample.cs renamed to docs/samples/Microsoft.ML.Samples/Dynamic/Trainers/BinaryClassification/RandomTrainerSample.cs

@@ -7,59 +7,64 @@ public static class RandomTrainer
     {
         public static void Example()
         {
-            // Downloading the dataset from github.com/dotnet/machinelearning.
-            // This will create a sentiment.tsv file in the filesystem.
-            // You can open this file, if you want to see the data. 
-            string dataFile = SamplesUtils.DatasetUtils.DownloadSentimentDataset()[0];
+            // Create a new context for ML.NET operations. It can be used for exception tracking and logging, 
+            // as a catalog of available operations and as the source of randomness.
+            var mlContext = new MLContext(seed: 1);
+
+            // Download and featurize the dataset.
+            var dataFiles = SamplesUtils.DatasetUtils.DownloadSentimentDataset();
+            var trainFile = dataFiles[0];
+            var testFile = dataFiles[1];
 
             // A preview of the data. 
             // Sentiment	SentimentText
             //      0	    " :Erm, thank you. "
             //      1	    ==You're cool==
 
-            // Create a new context for ML.NET operations. It can be used for exception tracking and logging, 
-            // as a catalog of available operations and as the source of randomness.
-            var mlContext = new MLContext(seed: 1);
-
-            // Step 1: Load the data as an IDataView.
-            // First, we define the loader: specify the data columns and where to find them in the text file.
-            var loader = mlContext.Data.CreateTextLoader(
+            // Step 1: Read the data as an IDataView.
+            // First, we define the reader: specify the data columns and where to find them in the text file.
+            var reader = mlContext.Data.CreateTextLoader(
                 columns: new[]
                     {
                         new TextLoader.Column("Sentiment", DataKind.Single, 0),
                         new TextLoader.Column("SentimentText", DataKind.String, 1)
                     },
                 hasHeader: true
             );
-            
-            // Load the data
-            var data = loader.Load(dataFile);
 
-            // Split it between training and test data
-            var trainTestData = mlContext.BinaryClassification.TrainTestSplit(data);
+            // Read the data
+            var trainData = reader.Load(trainFile);
 
             // Step 2: Pipeline 
             // Featurize the text column through the FeaturizeText API. 
             // Then append a binary classifier, setting the "Label" column as the label of the dataset, and 
             // the "Features" column produced by FeaturizeText as the features column.
             var pipeline = mlContext.Transforms.Text.FeaturizeText("Features", "SentimentText")
-                    .AppendCacheCheckpoint(mlContext) // Add a data-cache step within a pipeline.
+                    .AppendCacheCheckpoint(mlContext)
                     .Append(mlContext.BinaryClassification.Trainers.Random());
 
             // Step 3: Train the pipeline
-            var trainedPipeline = pipeline.Fit(trainTestData.TrainSet);
+            var trainedPipeline = pipeline.Fit(trainData);
 
             // Step 4: Evaluate on the test set
-            var transformedData = trainedPipeline.Transform(trainTestData.TestSet);
+            var transformedData = trainedPipeline.Transform(reader.Load(testFile));
             var evalMetrics = mlContext.BinaryClassification.Evaluate(transformedData, label: "Sentiment");
-
-            // Step 5: Inspect the output
-            Console.WriteLine("Accuracy: " + evalMetrics.Accuracy);
+            SamplesUtils.ConsoleUtils.PrintMetrics(evalMetrics);
 
             // We expect an output probability closet to 0.5 as the Random trainer outputs a random prediction.
             // Regardless of the input features, the trainer will predict either positive or negative label with equal probability.
-            // Expected output (close to 0.5):
-            // Accuracy: 0.588235294117647
+            // Expected output: (close to 0.5):
+
+            //  Accuracy: 0.56
+            //  AUC: 0.57
+            //  F1 Score: 0.60
+            //  Negative Precision: 0.57
+            //  Negative Recall: 0.44
+            //  Positive Precision: 0.55
+            //  Positive Recall: 0.67
+            //  LogLoss: 1.53
+            //  LogLossReduction: -53.37
+            //  Entropy: 1.00
         }
     }
 }

docs/samples/Microsoft.ML.Samples/Dynamic/Trainers/BinaryClassification/SymbolicStochasticGradientDescentWithOptions.cs

@@ -20,7 +20,7 @@ public static void Example()
             var split = mlContext.BinaryClassification.TrainTestSplit(data, testFraction: 0.1);
             // Create data training pipeline
             var pipeline = mlContext.BinaryClassification.Trainers.SymbolicStochasticGradientDescent(
-                    new ML.Trainers.HalLearners.SymSgdClassificationTrainer.Options()
+                    new ML.Trainers.HalLearners.SymbolicStochasticGradientDescentClassificationTrainer.Options()
                     {
                         LearningRate = 0.2f,
                         NumberOfIterations = 10,

docs/samples/Microsoft.ML.Samples/Dynamic/Trainers/Regression/OrdinaryLeastSquaresWithOptions.cs

@@ -44,7 +44,7 @@ public static void Example()
 
             // Create the estimator, here we only need OrdinaryLeastSquares trainer 
             // as data is already processed in a form consumable by the trainer
-            var pipeline = mlContext.Regression.Trainers.OrdinaryLeastSquares(new OlsLinearRegressionTrainer.Options()
+            var pipeline = mlContext.Regression.Trainers.OrdinaryLeastSquares(new OrdinaryLeastSquaresRegressionTrainer.Options()
             {
                 L2Weight = 0.1f,
                 PerParameterSignificance = false

docs/samples/Microsoft.ML.Samples/Static/SDCARegression.cs

@@ -46,12 +46,10 @@ public static void SdcaRegression()
             var model = learningPipeline.Fit(trainData);
 
             // Check the weights that the model learned
-            VBuffer<float> weights = default;
-            pred.GetFeatureWeights(ref weights);
+            var weights = pred.Weights;
 
-            var weightsValues = weights.GetValues();
-            Console.WriteLine($"weight 0 - {weightsValues[0]}");
-            Console.WriteLine($"weight 1 - {weightsValues[1]}");
+            Console.WriteLine($"weight 0 - {weights[0]}");
+            Console.WriteLine($"weight 1 - {weights[1]}");
 
             // Evaluate how the model is doing on the test data
             var dataWithPredictions = model.Transform(testData);

src/Microsoft.ML.Data/Dirty/PredictorInterfaces.cs

@@ -146,7 +146,8 @@ internal interface ICanSaveInSourceCode
     /// <summary>
     /// Interface implemented by components that can assign weights to features.
     /// </summary>
-    public interface IHaveFeatureWeights
+    [BestFriend]
+    internal interface IHaveFeatureWeights
     {
         /// <summary>
         /// Returns the weights for the features.

src/Microsoft.ML.DnnImageFeaturizer.AlexNet/AlexNetExtension.cs

@@ -2,6 +2,7 @@
 // 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;
 using System.IO;
 using Microsoft.ML.Data;
 
@@ -21,7 +22,7 @@ public static class AlexNetExtension
         /// </summary>
         public static EstimatorChain<ColumnCopyingTransformer> AlexNet(this DnnImageModelSelector dnnModelContext, IHostEnvironment env, string outputColumnName, string inputColumnName)
         {
-            return AlexNet(dnnModelContext, env, outputColumnName, inputColumnName, Path.Combine(AssemblyPathHelpers.GetExecutingAssemblyLocation(), "DnnImageModels"));
+            return AlexNet(dnnModelContext, env, outputColumnName, inputColumnName, Path.Combine(AppDomain.CurrentDomain.BaseDirectory, "DnnImageModels"));
         }
 
         /// <summary>