/// <summary>
/// Learns a model that can map the given inputs to the given outputs.
/// </summary>
/// <param name="x">The model inputs.</param>
/// <param name="y">The desired outputs associated with each <paramref name="x">inputs</paramref>.</param>
/// <param name="weights">The weight of importance for each input-output pair (if supported by the learning algorithm).</param>
/// <returns>A model that has learned how to produce <paramref name="y" /> given <paramref name="x" />.</returns>
/// <exception cref="System.InvalidOperationException">
/// Please set the Learner property before calling the Learn(x, y) method.
/// or
/// Please set the Learner property before calling the Learn(x, y) method.
/// </exception>
public override SplitResult <TModel, TInput, TOutput> Learn(TInput[] x, TOutput[] y, double[] weights = null)
{
if (Learner == null)
{
throw new InvalidOperationException("Please set the Learner property before calling the Learn(x, y) method.");
}
if (Loss == null)
{
throw new InvalidOperationException("Please set the Learner property before calling the Learn(x, y) method.");
}
int n = x.Length;
if (this.Indices == null || this.IndicesValidationSet == null || this.IndicesTrainingSet == null)
{
this.Indices = CreateValidationSplits(x, y);
this.IndicesValidationSet = Indices.Find(i => i == 1);
this.IndicesTrainingSet = Indices.Find(i => i == 0);
}
else
{
this.ValidationSetProportion = this.IndicesValidationSet.Length / (double)this.Indices.Length;
}
var split = new TrainValDataSplit <TInput, TOutput>(0, x, y, weights, IndicesTrainingSet, IndicesValidationSet);
SplitResult <TModel, TInput, TOutput> result = LearnSubset(split);
return(result);
}
/// <summary>
/// Learns and evaluates a model in a single subset of the data.
/// </summary>
///
/// <param name="subset">The subset of the data containing the training and testing subsets where
/// a model should be trained and evaluated, respectively.</param>
/// <param name="index">The index of this subset, if applicable.</param>
///
/// <returns>A <see cref="SplitResult{TModel, TInput, TOutput}"/> object containing the created model
/// and its performance on the training and validation sets.</returns>
///
protected SplitResult <TModel, TInput, TOutput> LearnSubset(TrainValDataSplit <TInput, TOutput> subset, int index = 0)
{
// Create the learning algorithm
TLearner teacher = Learner(subset.Training);
TInput[] trainInputs = subset.Training.Inputs;
TOutput[] trainOutputs = subset.Training.Outputs;
double[] trainWeights = subset.Training.Weights;
// Use the learning algorithm to learn a new model
try
{
TModel model = Fit(teacher, trainInputs, trainOutputs, trainWeights);
// Evaluate metrics using the created model
TInput[] valInputs = subset.Validation.Inputs;
TOutput[] valOutputs = subset.Validation.Outputs;
double[] valWeights = subset.Validation.Weights;
var trainResult = new SetResult <TModel>(model, subset.Training.Indices, "Training", subset.Training.Proportion);
var valResult = new SetResult <TModel>(model, subset.Validation.Indices, "Validation", subset.Validation.Proportion);
trainResult.Value = Loss(trainOutputs, model.Transform(trainInputs), trainResult);
valResult.Value = Loss(valOutputs, model.Transform(valInputs), valResult);
return(new SplitResult <TModel, TInput, TOutput>(model, index: index)
{
Training = trainResult,
Validation = valResult
});
}
catch
{
if (!DefaultValue.HasValue)
{
throw; // no default value has been set; we cannot do anything in this case
}
return(new SplitResult <TModel, TInput, TOutput>(null, index: index)
{
Training = new SetResult <TModel>(null, subset.Training.Indices, "Training", subset.Training.Proportion)
{
Value = DefaultValue.Value
},
Validation = new SetResult <TModel>(null, subset.Validation.Indices, "Validation", subset.Validation.Proportion)
{
Value = DefaultValue.Value
}
});
}
}
