Data file: yelp_labelled.txt<\/p>\n
SentimentData.cs<\/p>\n
\r\nusing Microsoft.ML.Data;\r\n\/\/ <\/SnippetAddUsings>\r\n\r\nnamespace SentimentAnalysis\r\n{\r\n \/\/ <SnippetDeclareTypes>\r\n public class SentimentData\r\n {\r\n [LoadColumn(0)]\r\n public string SentimentText;\r\n\r\n [LoadColumn(1), ColumnName("Label")]\r\n public bool Sentiment;\r\n }\r\n\r\n public class SentimentPrediction : SentimentData\r\n {\r\n\r\n [ColumnName("PredictedLabel")]\r\n public bool Prediction { get; set; }\r\n\r\n public float Probability { get; set; }\r\n\r\n public float Score { get; set; }\r\n }\r\n \/\/ <\/SnippetDeclareTypes>\r\n}\r\n<\/pre>\nProgram.cs<\/p>\n
\r\n\/\/ <SnippetAddUsings>\r\nusing System;\r\nusing System.Collections.Generic;\r\nusing System.IO;\r\nusing System.Linq;\r\nusing Microsoft.ML;\r\nusing Microsoft.ML.Data;\r\nusing static Microsoft.ML.DataOperationsCatalog;\r\nusing Microsoft.ML.Trainers;\r\nusing Microsoft.ML.Transforms.Text;\r\n\/\/ <\/SnippetAddUsings>\r\n\r\nnamespace SentimentAnalysis\r\n{\r\n class Program\r\n {\r\n \/\/ <SnippetDeclareGlobalVariables>\r\n static readonly string _dataPath = Path.Combine(Environment.CurrentDirectory, "Data", "yelp_labelled.txt");\r\n \/\/ <\/SnippetDeclareGlobalVariables>\r\n\r\n static void Main(string[] args)\r\n {\r\n \/\/ Create ML.NET context\/local environment - allows you to add steps in order to keep everything together\r\n \/\/ as you discover the ML.NET trainers and transforms\r\n \/\/ <SnippetCreateMLContext>\r\n MLContext mlContext = new MLContext();\r\n \/\/ <\/SnippetCreateMLContext>\r\n\r\n \/\/ <SnippetCallLoadData>\r\n TrainTestData splitDataView = LoadData(mlContext);\r\n \/\/ <\/SnippetCallLoadData>\r\n\r\n \/\/ <SnippetCallBuildAndTrainModel>\r\n ITransformer model = BuildAndTrainModel(mlContext, splitDataView.TrainSet);\r\n \/\/ <\/SnippetCallBuildAndTrainModel>\r\n\r\n \/\/ <SnippetCallEvaluate>\r\n Evaluate(mlContext, model, splitDataView.TestSet);\r\n \/\/ <\/SnippetCallEvaluate>\r\n\r\n \/\/ <SnippetCallUseModelWithSingleItem>\r\n UseModelWithSingleItem(mlContext, model);\r\n \/\/ <\/SnippetCallUseModelWithSingleItem>\r\n\r\n \/\/ <SnippetCallUseModelWithBatchItems>\r\n UseModelWithBatchItems(mlContext, model);\r\n \/\/ <\/SnippetCallUseModelWithBatchItems>\r\n\r\n Console.WriteLine();\r\n Console.WriteLine("=============== End of process ===============");\r\n }\r\n\r\n public static TrainTestData LoadData(MLContext mlContext)\r\n {\r\n \/\/ Note that this case, loading your training data from a file,\r\n \/\/ is the easiest way to get started, but ML.NET also allows you\r\n \/\/ to load data from databases or in-memory collections.\r\n \/\/ <SnippetLoadData>\r\n IDataView dataView = mlContext.Data.LoadFromTextFile<SentimentData>(_dataPath, hasHeader: false);\r\n \/\/ <\/SnippetLoadData>\r\n\r\n \/\/ You need both a training dataset to train the model and a test dataset to evaluate the model.\r\n \/\/ Split the loaded dataset into train and test datasets\r\n \/\/ Specify test dataset percentage with the `testFraction`parameter\r\n \/\/ <SnippetSplitData>\r\n TrainTestData splitDataView = mlContext.Data.TrainTestSplit(dataView, testFraction: 0.2);\r\n \/\/ <\/SnippetSplitData>\r\n\r\n \/\/ <SnippetReturnSplitData>\r\n return splitDataView;\r\n \/\/ <\/SnippetReturnSplitData>\r\n }\r\n\r\n public static ITransformer BuildAndTrainModel(MLContext mlContext, IDataView splitTrainSet)\r\n {\r\n \/\/ Create a flexible pipeline (composed by a chain of estimators) for creating\/training the model.\r\n \/\/ This is used to format and clean the data.\r\n \/\/ Convert the text column to numeric vectors (Features column)\r\n \/\/ <SnippetFeaturizeText>\r\n var estimator = mlContext.Transforms.Text.FeaturizeText(outputColumnName: "Features", inputColumnName: nameof(SentimentData.SentimentText))\r\n \/\/<\/SnippetFeaturizeText>\r\n \/\/ append the machine learning task to the estimator\r\n \/\/ <SnippetAddTrainer>\r\n .Append(mlContext.BinaryClassification.Trainers.SdcaLogisticRegression(labelColumnName: "Label", featureColumnName: "Features"));\r\n \/\/ <\/SnippetAddTrainer>\r\n\r\n \/\/ Create and train the model based on the dataset that has been loaded, transformed.\r\n \/\/ <SnippetTrainModel>\r\n Console.WriteLine("=============== Create and Train the Model ===============");\r\n var model = estimator.Fit(splitTrainSet);\r\n Console.WriteLine("=============== End of training ===============");\r\n Console.WriteLine();\r\n \/\/ <\/SnippetTrainModel>\r\n\r\n \/\/ Returns the model we trained to use for evaluation.\r\n \/\/ <SnippetReturnModel>\r\n return model;\r\n \/\/ <\/SnippetReturnModel>\r\n }\r\n\r\n public static void Evaluate(MLContext mlContext, ITransformer model, IDataView splitTestSet)\r\n {\r\n \/\/ Evaluate the model and show accuracy stats\r\n\r\n \/\/Take the data in, make transformations, output the data.\r\n \/\/ <SnippetTransformData>\r\n Console.WriteLine("=============== Evaluating Model accuracy with Test data===============");\r\n IDataView predictions = model.Transform(splitTestSet);\r\n \/\/ <\/SnippetTransformData>\r\n\r\n \/\/ BinaryClassificationContext.Evaluate returns a BinaryClassificationEvaluator.CalibratedResult\r\n \/\/ that contains the computed overall metrics.\r\n \/\/ <SnippetEvaluate>\r\n CalibratedBinaryClassificationMetrics metrics = mlContext.BinaryClassification.Evaluate(predictions, "Label");\r\n \/\/ <\/SnippetEvaluate>\r\n\r\n \/\/ The Accuracy metric gets the accuracy of a model, which is the proportion\r\n \/\/ of correct predictions in the test set.\r\n\r\n \/\/ The AreaUnderROCCurve metric is equal to the probability that the algorithm ranks\r\n \/\/ a randomly chosen positive instance higher than a randomly chosen negative one\r\n \/\/ (assuming 'positive' ranks higher than 'negative').\r\n\r\n \/\/ The F1Score metric gets the model's F1 score.\r\n \/\/ The F1 score is the harmonic mean of precision and recall:\r\n \/\/ 2 * precision * recall \/ (precision + recall).\r\n\r\n \/\/ <SnippetDisplayMetrics>\r\n Console.WriteLine();\r\n Console.WriteLine("Model quality metrics evaluation");\r\n Console.WriteLine("--------------------------------");\r\n Console.WriteLine($"Accuracy: {metrics.Accuracy:P2}");\r\n Console.WriteLine($"Auc: {metrics.AreaUnderRocCurve:P2}");\r\n Console.WriteLine($"F1Score: {metrics.F1Score:P2}");\r\n Console.WriteLine("=============== End of model evaluation ===============");\r\n \/\/<\/SnippetDisplayMetrics>\r\n }\r\n\r\n private static void UseModelWithSingleItem(MLContext mlContext, ITransformer model)\r\n {\r\n \/\/ <SnippetCreatePredictionEngine1>\r\n PredictionEngine<SentimentData, SentimentPrediction> predictionFunction = mlContext.Model.CreatePredictionEngine<SentimentData, SentimentPrediction>(model);\r\n \/\/ <\/SnippetCreatePredictionEngine1>\r\n\r\n \/\/ <SnippetCreateTestIssue1>\r\n SentimentData sampleStatement = new SentimentData\r\n {\r\n SentimentText = "This was a very bad steak"\r\n };\r\n \/\/ <\/SnippetCreateTestIssue1>\r\n\r\n \/\/ <SnippetPredict>\r\n var resultPrediction = predictionFunction.Predict(sampleStatement);\r\n \/\/ <\/SnippetPredict>\r\n \/\/ <SnippetOutputPrediction>\r\n Console.WriteLine();\r\n Console.WriteLine("=============== Prediction Test of model with a single sample and test dataset ===============");\r\n\r\n Console.WriteLine();\r\n Console.WriteLine($"Sentiment: {resultPrediction.SentimentText} | Prediction: {(Convert.ToBoolean(resultPrediction.Prediction) ? "Positive" : "Negative")} | Probability: {resultPrediction.Probability} ");\r\n\r\n Console.WriteLine("=============== End of Predictions ===============");\r\n Console.WriteLine();\r\n \/\/ <\/SnippetOutputPrediction>\r\n }\r\n\r\n public static void UseModelWithBatchItems(MLContext mlContext, ITransformer model)\r\n {\r\n \/\/ Adds some comments to test the trained model's data points.\r\n \/\/ <SnippetCreateTestIssues>\r\n IEnumerable<SentimentData> sentiments = new[]\r\n {\r\n new SentimentData\r\n {\r\n SentimentText = "This was a horrible meal"\r\n },\r\n new SentimentData\r\n {\r\n SentimentText = "I love this spaghetti."\r\n }\r\n };\r\n \/\/ <\/SnippetCreateTestIssues>\r\n\r\n \/\/ Load batch comments just created\r\n \/\/ <SnippetPrediction>\r\n IDataView batchComments = mlContext.Data.LoadFromEnumerable(sentiments);\r\n\r\n IDataView predictions = model.Transform(batchComments);\r\n\r\n \/\/ Use model to predict whether comment data is Positive (1) or Negative (0).\r\n IEnumerable<SentimentPrediction> predictedResults = mlContext.Data.CreateEnumerable<SentimentPrediction>(predictions, reuseRowObject: false);\r\n \/\/ <\/SnippetPrediction>\r\n\r\n \/\/ <SnippetAddInfoMessage>\r\n Console.WriteLine();\r\n\r\n Console.WriteLine("=============== Prediction Test of loaded model with multiple samples ===============");\r\n \/\/ <\/SnippetAddInfoMessage>\r\n\r\n Console.WriteLine();\r\n\r\n \/\/ <SnippetDisplayResults>\r\n foreach (SentimentPrediction prediction in predictedResults)\r\n {\r\n Console.WriteLine($"Sentiment: {prediction.SentimentText} | Prediction: {(Convert.ToBoolean(prediction.Prediction) ? "Positive" : "Negative")} | Probability: {prediction.Probability} ");\r\n }\r\n Console.WriteLine("=============== End of predictions ===============");\r\n \/\/ <\/SnippetDisplayResults>\r\n }\r\n }\r\n}\r\n<\/pre>\nOutput:<\/p>\n
\r\nModel quality metrics evaluation\r\n--------------------------------\r\nAccuracy: 83.96%\r\nAuc: 89.88%\r\nF1Score: 84.38%\r\n=============== End of model evaluation ===============\r\n\r\n=============== Prediction Test of model with a single sample and test dataset ===============\r\n\r\nSentiment: This was a very bad steak | Prediction: Negative | Probability: 0.026840538\r\n=============== End of Predictions ===============\r\n\r\n\r\n=============== Prediction Test of loaded model with multiple samples ===============\r\n\r\nSentiment: This was a horrible meal | Prediction: Negative | Probability: 0.041398816\r\nSentiment: I love this spaghetti. | Prediction: Positive | Probability: 0.99719304\r\n<\/pre>\n