Sentiment analysis using ML.NET and F#

Program.fs

open Shared
open System
open System.IO
open Microsoft.ML
open Shared.Models.MLModel

// Define variables
let baseDirectory = __SOURCE_DIRECTORY__
let baseDirectory' = Directory.GetParent(baseDirectory)

let appPath = Path.GetDirectoryName(Environment.GetCommandLineArgs().[0])

let baseDatasetsLocation = Path.Combine(baseDirectory'.FullName, "MLData")
let dataPath = sprintf @"%s/yelp_labelled.csv" baseDatasetsLocation

let baseModelsPath = Path.Combine(baseDirectory'.FullName, "MLData")
let modelPath = sprintf @"%s/SentimentModel.zip" baseModelsPath

let absolutePath relativePath = 
    let dataRoot = FileInfo(Reflection.Assembly.GetExecutingAssembly().Location)
    Path.Combine(dataRoot.Directory.FullName, relativePath)

let buildTrainEvaluateAndSaveModel (mlContext : MLContext) =
   
    // STEP 1: Common data loading configuration
    let dataView = mlContext.Data.LoadFromTextFile<SentimentIssue>(dataPath, hasHeader = true)
    
    let trainTestSplit = mlContext.Data.TrainTestSplit(dataView, testFraction=0.2)
    let trainingDataView = trainTestSplit.TrainSet
    let testDataView = trainTestSplit.TestSet

    // STEP 2: Common data process configuration with pipeline data transformations          
    let dataProcessPipeline = mlContext.Transforms.Text.FeaturizeText("Features", "Text")

    // STEP 3: Set the training algorithm, then create and config the modelBuilder                            
    let trainer = mlContext.BinaryClassification.Trainers.FastTree(labelColumnName = "Label", featureColumnName = "Features")
    let trainingPipeline = dataProcessPipeline.Append(trainer)
    
    // STEP 4: Train the model fitting to the DataSet
    printfn "=============== Training the model ==============="
    let trainedModel = trainingPipeline.Fit(trainingDataView)
    
    // STEP 5: Evaluate the model and show accuracy stats
    printfn "===== Evaluating Model's accuracy with Test data ====="
    let predictions = trainedModel.Transform testDataView
    let metrics = mlContext.BinaryClassification.Evaluate(predictions, "Label", "Score")

    Common.ConsoleHelper.printBinaryClassificationMetrics (trainer.ToString()) metrics

    // STEP 6: Save/persist the trained model to a .ZIP file
    use fs = new FileStream(modelPath, FileMode.Create, FileAccess.Write, FileShare.Write)
    mlContext.Model.Save(trainedModel, trainingDataView.Schema, fs)

    printfn "The model is saved to %s" (absolutePath modelPath)


// Test a single prediction by loding the model from the file, first.
let testSinglePrediction (mlContext : MLContext) =
    let sampleStatement = { Label = true; Text = "Sentiment analysis" }
    
    use stream = new FileStream(modelPath, FileMode.Open, FileAccess.Read, FileShare.Read)
    let trainedModel,inputSchema = mlContext.Model.Load(stream)
    
    // Create prediction engine related to the loaded trained model
    let predEngine= mlContext.Model.CreatePredictionEngine<SentimentIssue, SentimentPrediction>(trainedModel)

    // Score
    let resultprediction = predEngine.Predict(sampleStatement)

    printfn "=============== Single Prediction  ==============="
    printfn 
        "Text: %s | Prediction: %s sentiment | Probability: %f"
        sampleStatement.Text
        (if not resultprediction.Prediction then "Negative" else "Positive")
        resultprediction.Probability
    printfn "=================================================="

[<EntryPoint>]
let main argv =
    
    //Set a random seed for repeatable/deterministic results across multiple trainings.
    let mlContext = MLContext(seed = Nullable 1)

    // Create, Train, Evaluate and Save a model
    buildTrainEvaluateAndSaveModel mlContext
    Common.ConsoleHelper.consoleWriteHeader "=============== End of training process ==============="

    // Make a single test prediction loding the model from .ZIP file
    testSinglePrediction mlContext

    Common.ConsoleHelper.consoleWriteHeader "=============== End of process ==============="
 
    0 

Just create a console app with F# in .NET 6 or higher, and inside create a directory called MLData that contains yelp_labelled it can be used in .txt, .csv and .tsv. The choice is up to your preference.