public static IServiceCollection AddSentimentModelCreationService <TModelLoader>(
this IServiceCollection services,
string modelName = "SentimentModel",
double testSlipFraction = 0.1,
Action <ModelLoderFileOptions>?configure = null)
where TModelLoader : ModelLoader
{
var builder = services.AddModelCreationService <SentimentIssue, BinaryClassificationMetricsResult>(modelName);
builder.AddSourceLoader <SentimentIssue, BinaryClassificationMetricsResult, EmbeddedSourceLoader <SentimentIssue> >(options =>
{
options.Sources.Add(new SourceLoaderFile <SentimentIssue>
{
// overrides default loading mechanism
CustomAction = () =>
{
var inputs = EmbeddedResourceHelper
.GetRecords <InputSentimentIssueRow>("Content.wikiDetoxAnnotated40kRows.tsv", delimiter: "\t", hasHeaderRecord: true);
// convert int to boolean values
var result = new List <SentimentIssue>();
foreach (var item in inputs)
{
var newItem = new SentimentIssue
{
Label = item.Label != 0,
Text = item.comment
};
result.Add(newItem);
}
return(result);
}
});
});
builder.AddModelLoader <SentimentIssue, BinaryClassificationMetricsResult, TModelLoader>(configure);
builder.ConfigureModelDefinition <SentimentIssue, BinaryClassificationMetricsResult>(
testSlipFraction,
options =>
{
options.TrainingPipelineConfigurator = (mlContext) =>
{
// STEP 2: Common data process configuration with pipeline data transformations
var dataProcessPipeline = mlContext.Transforms.Text.FeaturizeText(outputColumnName: "Features", inputColumnName: nameof(SentimentIssue.Text));
// STEP 3: Set the training algorithm, then create and config the modelBuilder
var trainer = mlContext.BinaryClassification.Trainers.SdcaLogisticRegression(labelColumnName: "Label", featureColumnName: "Features");
var trainingPipeline = dataProcessPipeline.Append(trainer);
return(new TrainingPipelineResult(trainingPipeline, trainer.ToString()));
};
options.EvaluateConfigurator = (mlContext, model, trainerName, dataView, _) =>
{
// STEP 5: Evaluate the model and show accuracy stats
var predictions = model.Transform(dataView);
var metrics = mlContext.BinaryClassification.Evaluate(data: predictions, labelColumnName: "Label", scoreColumnName: "Score");
return(new BinaryClassificationMetricsResult(trainerName, metrics));
};
});
builder.ConfigureService <SentimentIssue, BinaryClassificationMetricsResult>(
options =>
{
options.DataLoader = async(loader, cancellationToken) =>
{
cancellationToken.ThrowIfCancellationRequested();
var data = loader.LoadData();
return(await Task.FromResult(data));
};
options.TrainModelConfigurator = (modelBuilder, data, logger) =>
{
// 1. load ML data set
modelBuilder.LoadData(data);
// 1. load default ML data set
modelBuilder.BuildDataView();
// 2. build training pipeline
var buildTrainingPipelineResult = modelBuilder.BuildTrainingPipeline();
// 3. train the model
var trainModelResult = modelBuilder.TrainModel();
// 4. evaluate quality of the pipeline
var evaluateResult = modelBuilder.Evaluate();
logger.LogInformation(evaluateResult.ToString());
return(evaluateResult);
};
options.ClassifyTestConfigurator = async(modelBuilder, logger, cancellationToken) =>
{
// 5. predict on sample data
var sw = ValueStopwatch.StartNew();
var tasks = new List <Task>
{
ClassifyAsync(modelBuilder, "This is a very rude movie", false, logger, cancellationToken),
ClassifyAsync(modelBuilder, "Hate All Of You're Work", true, logger, cancellationToken)
};
await Task.WhenAll(tasks);
};
});
return(services);
}