BanditAgent#

class statrl.settings.bandits.stochastic.knownhorizon.agent.BanditAgent(name, seed=1)[source]#

Bases: ABC

Base class for horizon-aware stochastic bandit agents.

Identical to the anytime protocol except that reset() receives the horizon, letting an agent tune its behaviour to the number of rounds it will play.

Parameters:

name (str) – Label used in logfiles and plot legends.
seed (int, default=1) – Seed for the agent’s own randomness.

horizon#

Number of rounds of the current run, set by reset().

Type:: int

np_random#

Agent-local generator, available after the first reset().

Type:: numpy.random.Generator

See also

statrl.settings.bandits.stochastic.anytime.agent.BanditAgent: The anytime counterpart.
statrl.settings.bandits.stochastic.knownhorizon.wrappers.wrapper_anytime_knownhorizon.AnytimeToKnownHorizonAgentWrapper: Runs an anytime agent in this setting.

Methods

`__init__`(name[, seed])
`reset`(horizon)	Start a new independent run of known length.
`select_arm`()	Choose the arm to pull next.
`update`(arm, reward)	Learn from the reward observed for the arm just pulled.

reset(horizon)[source]#

Start a new independent run of known length.

Parameters:: horizon (int) – Number of rounds that will be played. Stored on horizon and free to be used by the selection rule.

abstractmethod select_arm()[source]#

Choose the arm to pull next.

Returns:: Index of the selected arm, in range(env.number_arms).
Return type:: int

update(arm, reward)[source]#

Learn from the reward observed for the arm just pulled.

Parameters:

arm (int) – Index of the arm that was pulled.
reward (float) – Reward observed for that arm.