poprox_recommender.components.diversifiers.calibration

Functions

compute_kl_divergence(interacted_distr, ...)

KL (p || q), the lower the better.

Classes

Calibrator([theta, num_slots])

class poprox_recommender.components.diversifiers.calibration.Calibrator(theta=0.1, num_slots=10)

Bases: Component

Parameters:

theta (float)

poprox_recommender.components.diversifiers.calibration.compute_kl_divergence(interacted_distr, reco_distr, kl_div=0.0, alpha=0.01)

KL (p || q), the lower the better.

alpha is not really a tuning parameter, it’s just there to make the computation more numerically stable.