public float Train(TrainingSet trainingSet, float learningRate)
{
if (shuffle)
{
trainingSet.Shuffle();
}
var lastLayer = Layers[Layers.Length - 1];
var outputs = lastLayer.Outputs;
var errors = new Tensor(outputs.Dimensions);
float sumCost = 0f;
for (int i = 0, count = trainingSet.Count; i < count; i++)
{
errors.Clear();
(var inputs, var expected) = trainingSet[i];
var actuals = CalculateTrainingOutputs(inputs);
sumCost += Loss.CalculateTotal(actuals, expected);
for (int o = 0; o < actuals.Length; o++)
{
errors[o] = Loss.Derivative(actuals[o], expected[o]);
}
BackPropagate(errors, learningRate, momentumRate);
}
return(sumCost / trainingSet.Count);
}
public float Test(TrainingSet validationSet)
{
float sumCost = 0f;
for (int i = 0, count = validationSet.Count; i < count; i++)
{
var inputs = validationSet[i].Item1;
var expected = validationSet[i].Item2;
var actuals = CalculateOutputs(inputs);
sumCost += Loss.CalculateTotal(actuals, expected);
}
return(sumCost / validationSet.Count);
}