public ITrainerEstimator CreateInstance(MLContext mlContext, IEnumerable <SweepableParam> sweepParams,
ColumnInformation columnInfo, IDataView validationSet)
{
LightGbmMulticlassTrainer.Options options = TrainerExtensionUtil.CreateLightGbmOptions <LightGbmMulticlassTrainer.Options, VBuffer <float>, MulticlassPredictionTransformer <OneVersusAllModelParameters>, OneVersusAllModelParameters>(sweepParams, columnInfo);
return(mlContext.MulticlassClassification.Trainers.LightGbm(options));
}
LightGbm(this SweepableMultiClassificationTrainers trainer, string labelColumnName = "Label", string featureColumnName = "Features", SweepableOption <LightGbmMulticlassTrainer.Options> optionBuilder = null, LightGbmMulticlassTrainer.Options defaultOption = null)
{
var context = trainer.Context;
if (optionBuilder == null)
{
optionBuilder = LightGbmMulticlassTrainerSweepableOptions.Default;
}
optionBuilder.SetDefaultOption(defaultOption);
return(context.AutoML().CreateSweepableEstimator(
(context, option) =>
{
if (defaultOption != null)
{
Util.CopyFieldsTo(defaultOption, option);
}
option.LabelColumnName = labelColumnName;
option.FeatureColumnName = featureColumnName;
return context.MulticlassClassification.Trainers.LightGbm(option);
},
optionBuilder,
trainerName: nameof(LightGbmMulticlassTrainer),
inputs: new string[] { featureColumnName },
outputs: new string[] { PredictedLabel }));
}
// This example requires installation of additional NuGet package for
// Microsoft.ML.FastTree at
// https://www.nuget.org/packages/Microsoft.ML.FastTree/
public static void Example()
{
// 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. Setting the seed to a fixed number
// in this example to make outputs deterministic.
var mlContext = new MLContext(seed: 0);
// Create a list of training data points.
var dataPoints = GenerateRandomDataPoints(1000);
// Convert the list of data points to an IDataView object, which is
// consumable by ML.NET API.
var trainingData = mlContext.Data.LoadFromEnumerable(dataPoints);
// Define trainer options.
var options = new LightGbmMulticlassTrainer.Options
{
Booster = new DartBooster.Options()
{
TreeDropFraction = 0.15,
XgboostDartMode = false
}
};
// Define the trainer.
var pipeline =
// Convert the string labels into key types.
mlContext.Transforms.Conversion.MapValueToKey("Label")
// Apply LightGbm multiclass trainer.
.Append(mlContext.MulticlassClassification.Trainers
.LightGbm(options));
// Train the model.
var model = pipeline.Fit(trainingData);
// Create testing data. Use different random seed to make it different
// from training data.
var testData = mlContext.Data
.LoadFromEnumerable(GenerateRandomDataPoints(500, seed: 123));
// Run the model on test data set.
var transformedTestData = model.Transform(testData);
// Convert IDataView object to a list.
var predictions = mlContext.Data
.CreateEnumerable <Prediction>(transformedTestData,
reuseRowObject: false).ToList();
// Look at 5 predictions
foreach (var p in predictions.Take(5))
{
Console.WriteLine($"Label: {p.Label}, " +
$"Prediction: {p.PredictedLabel}");
}
// Expected output:
// Label: 1, Prediction: 1
// Label: 2, Prediction: 2
// Label: 3, Prediction: 3
// Label: 2, Prediction: 2
// Label: 3, Prediction: 3
// Evaluate the overall metrics
var metrics = mlContext.MulticlassClassification
.Evaluate(transformedTestData);
PrintMetrics(metrics);
// Expected output:
// Micro Accuracy: 0.98
// Macro Accuracy: 0.98
// Log Loss: 0.07
// Log Loss Reduction: 0.94
// Confusion table
// ||========================
// PREDICTED || 0 | 1 | 2 | Recall
// TRUTH ||========================
// 0 || 156 | 0 | 4 | 0.9750
// 1 || 0 | 171 | 6 | 0.9661
// 2 || 1 | 0 | 162 | 0.9939
// ||========================
// Precision ||0.9936 |1.0000 |0.9419 |
}
private static void Main(string[] args)
{
// Simuliert die Messungenauigkeit. Ist der relative Standardfehler. 0.1 bedeutet also,
// dass bei einem berechneten Wert von 70 der Messwert in 68% der Fälle
// (Normalverteilung) zwischen 70 +/- 7 ist.
float bias = 0.1f;
Randomizer.Seed = new Random(82541);
Faker fkr = new Faker();
Accesspoint[] accesspoints = new Accesspoint[]
{
new Accesspoint("a1", new Point(0, 30)),
new Accesspoint("a2", new Point(10, 20)),
new Accesspoint("a3", new Point(0, 0))
};
// WIr positionieren 30 virtuelle Smartphones im Stockwerk.
var devices = (from i in Enumerable.Range(0, 30)
select new
{
Id = Guid.NewGuid().ToString("N"),
Point = new Point(fkr.Random.Float(0, 10), fkr.Random.Float(0, 30))
}).ToArray();
List <Measurement> measurements = new List <Measurement>();
// Nun führen wir jede Minute eine Messung durch.
for (DateTime date = new DateTime(2020, 1, 1, 12, 0, 0); date < new DateTime(2020, 1, 1, 13, 0, 0); date = date.AddMinutes(1))
{
// Jedes Gerät liefert einen Messwert.
foreach (var device in devices)
{
// Wir bekommen zwischen 1 und (Anz AP) pro Messung gesendet. Da nicht immer alle
// APs empfangen werden, ist das ein guter Test, ob unser Programm damit umgehen
// kann.
var measuredAccesspoints = fkr.Random.ListItems(accesspoints, fkr.Random.Int(1, accesspoints.Length));
var signals = from ap in measuredAccesspoints
let signalStrength = device.Point.SignalStrength(ap, val => fkr.Random.GaussianFloat(val, val * bias))
select new Signal(ap, signalStrength);
Measurement m = new Measurement(device.Id, date, device.Point, signals);
measurements.Add(m);
}
}
var avgAp = measurements.Select(m => m.Signals.Count()).Average();
Console.WriteLine($"{measurements.Count} Messungen, {avgAp:0.00} APs pro Messung im Mittel.");
var table = from m in measurements
select new
{
m.Device,
m.Date,
m.Location.Room,
// Floatarray mit den gemessenen Feldstärken der Accesspoint. Ist genau in
// der Reihenfolge und größe wie unser accesspoints Array.
// Details siehe in der Funktion Spread in der Datei ArrayExtensions.cs.
Signals = m.Signals.Spread(accesspoints, s => s.Accesspoint).Select(s => s?.Value ?? 0).ToArray()
};
using (var outStream = new StreamWriter("measurements.txt", false, Encoding.UTF8))
{
outStream.WriteLine($"DEVICE\tROOM\tDATE\t{string.Join('\t', accesspoints.Select(s => s.Mac))}");
foreach (var row in table)
{
outStream.WriteLine($"{row.Device}\t{row.Room}\t{row.Date}\t{string.Join('\t', row.Signals)}");
}
}
// https://docs.microsoft.com/en-us/dotnet/api/microsoft.ml.lightgbmextensions.lightgbm?view=ml-dotnet
// Define trainer options.
var options = new LightGbmMulticlassTrainer.Options
{
LabelColumnName = nameof(RoomMeasurement.Label),
FeatureColumnName = nameof(RoomMeasurement.Values),
Booster = new DartBooster.Options()
{
TreeDropFraction = 0.15,
XgboostDartMode = false
}
};
var mlContext = new MLContext(seed: 0);
var trainData = mlContext.Data.LoadFromEnumerable(from m in measurements
select new RoomMeasurement
{
Label = m.Location.Room,
Values = m.Signals.Spread(accesspoints, s => s.Accesspoint).Select(s => s?.Value ?? 0).ToArray()
});
// Define the trainer.
var pipeline = mlContext.Transforms.Conversion.MapValueToKey(nameof(RoomMeasurement.Label))
.Append(mlContext.MulticlassClassification.Trainers.LightGbm(options));
var model = pipeline.Fit(trainData);
// Create testing data. Use different random seed to make it different
// from training data.
var testData = trainData;
// Wandelt die Label in UInt32 Werte für die interne Verarbeitung.
var transformedTestData = model.Transform(testData);
// Convert IDataView object to a list.
var predictions = mlContext.Data
.CreateEnumerable <RoomPrediction>(transformedTestData, reuseRowObject: false).ToList();
// Look at 5 predictions
foreach (var p in predictions.Take(5))
{
Console.WriteLine($"Label: {p.Label}, Prediction: {p.PredictedLabel}");
}
var metrics = mlContext.MulticlassClassification.Evaluate(transformedTestData);
PrintMetrics(metrics);
}
