/// <summary>
/// Predicts the related items, given the item index and the type (either References or Entities).
/// </summary>
/// <param name="itemId">The item index in the corresponding feature map column.</param>
/// <param name="count">count</param>
/// <param name="itemType">
/// Type of item to return related items (i.e. References = user ratings OR Entities = books or
/// movies)
/// </param>
/// <returns>Vector of predictions.</returns>
public Vector PredictRelated(int itemId, int count = 5, ItemType itemType = ItemType.References)
{
var predictions = (ThetaX * ThetaY.T).Each((v, r, c) => v + Mu[r]);
var feature = itemType == ItemType.Entities
? predictions[EntityFeatureMap.IndexOf(itemId), VectorType.Col]
: predictions[ReferenceFeatureMap.IndexOf(itemId), VectorType.Row];
var result = Vector.Zeros(count);
switch (itemType)
{
case ItemType.Entities:
{
result = predictions.GetCols()
.Select((s, i) => new { Col = s, Idx = i })
.OrderBy(v => RelatedDistanceFunction.Compute(feature, v.Col))
.Take(count)
.Select(s => (double)s.Idx).ToVector();
}
break;
case ItemType.References:
{
result = predictions.GetRows()
.Select((s, i) => new { Row = s, Idx = i })
.OrderBy(v => RelatedDistanceFunction.Compute(feature, v.Row))
.Take(count)
.Select(s => (double)s.Idx).ToVector();
}
break;
}
return(result);
}
/// <summary>
/// Predicts all the recommendations of the Items for the supplied reference, i.e. a user.
/// </summary>
/// <param name="referenceId">Reference index to use for generating predictions.</param>
/// <returns>Vector of predictions.</returns>
public Vector Predict(int referenceId)
{
// [entities x features] * [references * features]
var predictions = (ThetaX * ThetaY.T).Each((v, r, c) => v + Mu[r]);
int[] indices;
predictions[ReferenceFeatureMap.IndexOf(referenceId), VectorType.Col].Sort(false, out indices);
return(Y.Slice(indices, true));
}
