Multi-arm bandit for continuous traffic allocation in a/b testing having continuous rewards

Abstract

Certain aspects of the disclosure provide for dynamically routing traffic in A/B testing environments. In examples, a method may include receiving reward data associated with each respective variant of a set of variants corresponding to continuous values representing an outcome measure for the respective variant; generating a posterior probability density function for each respective variant in the set of variants based on the reward data for the respective variant and a prior probability density function for the respective variant; performing a sampling operation to generate an expected value of reward associated with each variant in the set of variants; assigning new traffic proportionally across each respective variant in the set of variants based on the expected value of reward associated with the respective variant; and based on the assigned new traffic, causing content associated with a variant of the set of variants to be provided to a user interface.