/// <summary>
/// Creates an in-memory EF Core Database and loads it with airline data
/// </summary>
/// <param name="options"></param>
private static void LoadAirlinesData(DbContextOptions <AirlinesContext> options)
{
// Load data into the DB
using (var airlinesModel = new AirlinesContext(options))
{
// Key for fake Ids
int key = 0;
// Load the FlightCode Data from csv
using (TextReader reader = new StreamReader(@"TrainingData\ManyFlightCodes.csv"))
{
var csvReader = new CsvReader(reader);
var flightCodes = csvReader.GetRecords <FlightCodeFeatures>();
airlinesModel.FlightCodes.AddRange(flightCodes.Select(f =>
{
var fc = new FlightCodes();
fc.Id = ++key;
fc.FlightCode = f.FlightCode;
fc.Iatacode = f.IATACode;
return(fc);
}));
airlinesModel.SaveChanges();
}
}
}
public static ITransformer TrainModel(DbContextOptions <AirlinesContext> dbOptions,
bool cacheData = false, int concurrency = 0, int nth = 1)
{
ITransformer trainedModel = null;
// Create an ML.NET environment
var mlContext = new MLContext(seed: 0);
// Train from EF DBContext
using (var airlinesModel = new AirlinesContext(dbOptions))
{
// Create an enumerable view of the DB training data
var flightCodeTrainingData = airlinesModel.FlightCodes.Where(fc => fc.Id % nth == 0).AsEnumerable()
.Select(f => new FlightCodeFeatures()
{
FlightCode = f.FlightCode,
IATACode = f.Iatacode
});
var trainingDataView = mlContext.Data.LoadFromEnumerable(flightCodeTrainingData);
// Set the key column (IATACode), featurize the text FlightCode column (to a long) and add it to the features collection
var dataProcessPipeline = mlContext.Transforms.Conversion.MapValueToKey(outputColumnName: "Label", inputColumnName: nameof(FlightCodeFeatures.IATACode))
.Append(mlContext.Transforms.Text.FeaturizeText(outputColumnName: "FlightCodeFeaturized", inputColumnName: nameof(FlightCodeFeatures.FlightCode)))
.Append(mlContext.Transforms.Concatenate(outputColumnName: "Features", "FlightCodeFeaturized"));
if (cacheData)
{
// Optionally cache the input (used if multiple passes required)
dataProcessPipeline.AppendCacheCheckpoint(mlContext);
}
// Define the trainer to be used
IEstimator <ITransformer> trainer = null;
trainer = mlContext.MulticlassClassification.Trainers.SdcaMaximumEntropy();
// Create a training pipeline that adds the trainer to the data pipeline and maps prediction to a string in the output (default name)
var trainingPipeline = dataProcessPipeline.Append(trainer)
.Append(mlContext.Transforms.Conversion.MapKeyToValue("PredictedLabel"));
// Do the actual training, reads the features and builds the model
Console.WriteLine($"Starting training");
var watch = System.Diagnostics.Stopwatch.StartNew();
trainedModel = trainingPipeline.Fit(trainingDataView);
watch.Stop();
long elapsedMs = watch.ElapsedMilliseconds;
Console.WriteLine($"Training took {elapsedMs / 1000f} secs");
Console.WriteLine();
}
return(trainedModel);
}