First public release of the streamlined package, extracted from the replication code for Weaver, Kumar, and Jain, “Nonparametric Pricing Bandits Leveraging Informational Externalities to Learn the Demand Curve” (Marketing Science).
PricingBandit(): single entry point running one pricing
experiment with any of six policies (“UCB”, “TS”, “GP-UCB”, “GP-TS”,
“GP-UCB-M”, “GP-TS-M”), a user-supplied willingness-to-pay vector, and
an arbitrary price grid.hetero = TRUE) available
for all Gaussian process variants, and periodic history resets
(reset) for time-varying demand.GetDiagnostics()) counting
how often numerical fallback paths fired; the counters draw no random
numbers, so results are unaffected.num_knots defaults to 11 for the monotonic variants
regardless of the number of arms (larger values can destabilize the
truncated sampler).timeout (default 5 seconds per
attempt) is a user-settable argument of
PricingBandit().