/// <summary>
/// Infers parameters for a given cold item.
/// </summary>
/// <param name="itemFeatures">The item features.</param>
/// <returns>A distribution over the item parameters.</returns>
public ItemParameterDistribution InferItemParameters(SparseFeatureVector itemFeatures)
{
ItemParameterDistribution result;
if (this.itemFeatureParameterPosteriors.FeatureCount == 0)
{
Debug.Assert(
itemFeatures.FeatureCount == 0,
"The number of item features passed must be equal to the number of item features learned.");
result = this.itemParameterDistributionAverage;
}
else
{
GaussianArray traits;
Gaussian bias;
AlgorithmUtils.AddFeatureContribution(
this.itemParameterDistributionAverage,
this.itemFeatureParameterPosteriors,
itemFeatures,
out traits,
out bias);
result = new ItemParameterDistribution(traits, bias);
}
return(result);
}
/// <summary>
/// Adds the feature contribution for a given user/item to its traits and bias.
/// </summary>
/// <param name="parameters">The user/item parameters.</param>
/// <param name="featureWeights">The feature weights.</param>
/// <param name="featureValues">The feature values.</param>
/// <param name="adjustedTraits">The resulting traits.</param>
/// <param name="adjustedBias">The resulting bias.</param>
public static void AddFeatureContribution(
EntityParameterDistribution parameters,
FeatureParameterDistribution featureWeights,
SparseFeatureVector featureValues,
out GaussianArray adjustedTraits,
out Gaussian adjustedBias)
{
adjustedTraits = new GaussianArray(
parameters.Traits.Count,
i => DoublePlusOp.SumAverageConditional(
parameters.Traits[i],
ComputeFeatureContribution(featureWeights.TraitWeights[i], featureValues.NonZeroFeatureValues, featureValues.NonZeroFeatureIndices)));
adjustedBias = DoublePlusOp.SumAverageConditional(
parameters.Bias,
ComputeFeatureContribution(featureWeights.BiasWeights, featureValues.NonZeroFeatureValues, featureValues.NonZeroFeatureIndices));
}
