mabman

This library is set to serve various implementations of multi armed bandit problem. The implementation includes some intuitive additions to it, that the users will need when putting the MAB into practice. The package includes callbacks, which allows users to log and checkpoint their agent. Also, it allows users to run continuous / multiple experiments without any code changes.

simulation

from bandit.agents.adversarial import EXP3
from bandit.arms import Arm
from bandit.callbacks import HistoryLogger, CheckPoint

callbacks = [
    HistoryLogger(),
    CheckPoint(in_every=50)
]
arms = [Arm('a', p=0.6), Arm('b', p=0.3), Arm('c', p=0.4)]
agent = EXP3(arms=arms, episodes=1000, callbacks=callbacks)
while not agent.stop:
    if name := agent.choose():
        r = agent.arm(name).draw()
        agent.reward(name, r)