Batched bandits

Read more in the user guide.

Protocol

agent.BatchBanditAgent	Base class for batched bandit agents.
environment.BatchMAB	Wrap a stochastic bandit into a batched one.
interaction.BatchBanditInteraction	Interaction loop for the batched bandit setting.

Agents

agents.BIMED.BIMED	Batched IMED with within-batch sequential index updates.
agents.BCB.BCB	BCB with CVaR = Expectation (adapted batch version).
agents.BCB.BCBnaif	BCB without the optimistic within-batch count increment.
agents.BABA.BABA	Batched Anytime Bandit Algorithm
agents._Oracle.Oracle	Baseline filling every batch with the best arm.
agents._Random.Random	Uniform exploration: fill each batch with independent random arms.

Environments

envs.parametric.BatchBernoulliBandit	Batched Bernoulli bandit, built from means and a named batch schedule.
envs.parametric.BatchGaussianBandit	Batched Gaussian bandit, built from means and a named batch schedule.
envs.parametric.BatchTruncatedGaussianBandit	Batched bandit with Gaussian arms truncated to [low, high].

Batch schedules

A schedule maps a round index to the size of that round’s batch. Pass one by name to the environment factories above, or directly to BatchMAB.

envs.parametric.baba_schedule	Batch sizes of the BABA epoch grid, as a plain list.
envs.parametric.exotic_schedule	Batch schedule cycling through linear, exponential, constant, and cubic.