public ModelTraining(IOrchestrator proxy)
{
var sample = new Queue <Temperature>();
Twin.Subscribe(async twin =>
{
Logger.LogInformation("Twin update");
lock (_sync)
{
_aggregationSize = twin.AggregationSize;
_tumblingWindowPercentage = twin.TumblingWindowPercentage;
}
await Twin.ReportAsync(twin);
return(TwinResult.Ok);
});
proxy.Training.Subscribe(this, async signal =>
{
Model model = null;
lock (sample)
{
sample.Enqueue(signal);
if (sample.Count >= _aggregationSize)
{
var kMeansTraining = new KMeansTraining(_numClusters);
kMeansTraining.TrainModel(sample.Select(e => new[] { e.TimeStamp, e.Value }).ToArray());
model = new Model
{
Algorithm = Algorithm.kMeans,
DataJson = kMeansTraining.SerializeModel()
};
for (var i = 0; i < _tumblingWindowPercentage * _aggregationSize / 100; i++)
{
sample.Dequeue();
}
}
}
if (model != null)
{
await Model.PublishAsync(model).ConfigureAwait(false);
}
return(MessageResult.Ok);
});
}
public ModelTraining(IOrchestrator proxy)
{
_sample = new Queue <Temperature>();
Twin.Subscribe(async twin =>
{
Console.WriteLine($"{typeof(ModelTraining).Name}::Twin update");
lock (_sync)
{
_aggregationSize = twin.AggregationSize;
_tumblingWindowPercentage = twin.TumblingWindowPercentage;
}
await Twin.ReportAsync(twin);
return(TwinResult.Ok);
});
proxy.Training.Subscribe(this, async signal =>
{
Model model = null;
lock (_sample)
{
_sample.Enqueue(signal);
if (_sample.Count >= _aggregationSize)
{
_kMeansTraining = new KMeansTraining(_numClusters);
_kMeansTraining.TrainModel(_sample.Select(e => new double[] { e.TimeStamp, e.Value }).ToArray());
model = new Model()
{
Algorithm = ThermostatApplication.Algorithm.kMeans,
DataJson = _kMeansTraining.SerializeModel()
};
for (int i = 0; i < _tumblingWindowPercentage * _aggregationSize / 100; i++)
{
_sample.Dequeue();
}
}
}
if (model != null)
{
await Model.PublishAsync(model).ConfigureAwait(false);
}
return(MessageResult.Ok);
});
}
