banditpylib.learners

Submodules

• banditpylib.learners.linear_bandit_learner
  • Classes
• banditpylib.learners.mab_collaborative_ftbai_learner
  • Classes
• banditpylib.learners.mab_fbbai_learner
  • Classes
• banditpylib.learners.mab_fcbai_learner
  • Classes
• banditpylib.learners.mab_learner
  • Classes
• banditpylib.learners.mnl_bandit_learner
  • Classes
• banditpylib.learners.thresholding_bandit_learner
  • Classes

Classes

• Goal: Abstract class for the goal of a learner

• IdentifyBestArm: Best arm identification

• MaximizeTotalRewards: Reward maximization

• MaximizeCorrectAnswers: Maximize correct answers

• MakeAllAnswersCorrect: Make all answers correct

• Learner: Abstract class for learners

• SinglePlayerLearner: Abstract class for single player learners

• CollaborativeLearner: Abstract class for collaborative learners

• CollaborativeAgent: Abstract class for collaborative agents

• CollaborativeMaster: Abstract class for collaborative masters that handle arm assignment and

class banditpylib.learners.Goal

Abstract class for the goal of a learner

Inheritance

abstract property name: str

Name of the goal

class banditpylib.learners.IdentifyBestArm(best_arm: data_pb2.Arm)

Best arm identification

Parameters

best_arm (Arm) – best arm identified by the learner

Inheritance

property name: str

Name of the goal

class banditpylib.learners.MaximizeTotalRewards

Reward maximization

Inheritance

property name: str

Name of the goal

class banditpylib.learners.MaximizeCorrectAnswers(answers: List[int])

Maximize correct answers

This is used by thresholding bandit learners.

Parameters

answers (List[int]) – answers obtained by the learner

Inheritance

property name: str

Name of the goal

class banditpylib.learners.MakeAllAnswersCorrect(answers: List[int])

Make all answers correct

This is used by thresholding bandit learners.

Parameters

answers (List[int]) – answers obtained by the learner

Inheritance

property name: str

Name of the goal

class banditpylib.learners.Learner(name: Optional[str])

Abstract class for learners

Parameters

name (Optional[str]) – alias name

Inheritance

abstract property goal: banditpylib.learners.utils.Goal

Goal of the learner

property name: str

Name of the learner

abstract reset()

Reset the learner

Warning

This function should be called before the start of the game.

abstract property running_environment: Union[type, List[type]]

Type of bandit environment the learner plays with

class banditpylib.learners.SinglePlayerLearner(name: Optional[str])

Abstract class for single player learners

Parameters

name (Optional[str]) – alias name

Inheritance

abstract actions(context: data_pb2.Context) → data_pb2.Actions

Actions of the learner

Parameters

context – contextual information about the bandit environment

Returns

actions to take

abstract update(feedback: data_pb2.Feedback)

Update the learner

Parameters

feedback – feedback returned by the bandit environment after actions() is executed

class banditpylib.learners.CollaborativeLearner(agent: banditpylib.learners.utils.CollaborativeAgent, master: banditpylib.learners.utils.CollaborativeMaster, num_agents: int, name: Optional[str] = None)

Abstract class for collaborative learners

Parameters

• agent (CollaborativeAgent) – one instance of a collaborative agent

• master (CollaboratveMaster) – instance of a collaborative master

• num_agents (int) – total number of agents involved

• name (Optional[str]) – alias name

Inheritance

property agents: List[banditpylib.learners.utils.CollaborativeAgent]

Involved agents

property master: banditpylib.learners.utils.CollaborativeMaster

Controlling master

reset()

Reset the learner

Warning

This function should be called before the start of the game.

class banditpylib.learners.CollaborativeAgent(name: Optional[str])

Abstract class for collaborative agents

Parameters

name (Optional[str]) – alias name

Inheritance

abstract actions(context: data_pb2.Context) → data_pb2.Actions

Actions of the agent

Parameters

context – contextual information about the bandit environment

Returns

actions to take

abstract broadcast() → Dict[int, Tuple[float, int]]

Broadcast information learnt in the current round

Returns

arm ids, corresponding average rewards seen, and numbers of pulls used to

deduce average rewards

property name: str

Name of the agent

abstract reset()

Reset the agent

Warning

This function should be called before the start of each game.

abstract set_input_arms(arms: List[int])

Assign a set of arms to the agent

Parameters

arms – arm indices that have been assigned

abstract update(feedback: data_pb2.Feedback)

Update the agent

Parameters

feedback – feedback returned by the bandit environment after actions() is executed

class banditpylib.learners.CollaborativeMaster(name: Optional[str])

Abstract class for collaborative masters that handle arm assignment and elimination

Parameters

name (Optional[str]) – alias name

Inheritance

abstract elimination(messages: Dict[int, Dict[int, Tuple[float, int]]]) → Dict[int, List[int]]

Update the set of active arms based on some criteria and return arm assignment

Parameters

messages – dict of messages broadcasted from agents, where key is agent_id

Returns

arm assignment per agent

abstract initial_arm_assignment() → Dict[int, List[int]]

The arm assignment before the first round

Returns

arm assignment per agent for all agents

property name: str

Name of the master

abstract reset()

Reset the master

Warning

This function should be called before the start of each game.