aind_behavior_gym.dynamic_foraging.agent package

Submodules

aind_behavior_gym.dynamic_foraging.agent.base module

Generic agent class for dynamic foraging

class aind_behavior_gym.dynamic_foraging.agent.base.DynamicForagingAgentBase(seed=None)

Bases: object

Generic agent class for dynamic foraging

act(observation)

Chooses an action based on the current observation.

Parameters:

observation – The current observation from the environment.

Returns:

The action chosen by the agent.

Return type:

action

fit(data)

Fit the parameters of the agent to data.

Parameters:

data – Either from animal data (model fitting) or simulated data (model recovery)

learn(observation, action, reward, next_observation, done)

Updates the agent’s knowledge or policy based on the last action and its outcome.

This is the core method that should be implemented by all non-trivial agents. It could be Q-learning, policy gradients, neural networks, etc.

Parameters:

• observation – The observation before the action was taken.

• action – The action taken by the agent.

• reward – The reward received after taking the action.

• next_observation – The next observation after the action.

• done – Whether the episode has ended.

load(filepath)

Loads the agent’s state or learned parameters from a file.

Parameters:

filepath (str) – The path to the file from which the agent’s state will be loaded.

perform(task: DynamicForagingTaskBase)

Perform one session (eposide) of the dynamic foraging task while learning.

reset()

Resets the agent’s internal state. Override this if your agent has an internal state.

save(filepath)

Saves the agent’s current state or learned parameters to a file.

Parameters:

filepath (str) – The path to the file where the agent’s state will be saved.

aind_behavior_gym.dynamic_foraging.agent.random_agent module

A Random agent

class aind_behavior_gym.dynamic_foraging.agent.random_agent.RandomAgent(seed=None)

Bases: DynamicForagingAgentBase

A Random agent

act(observation)

Simply random choose in the action space

learn(observation, action, reward, next_observation, done)

No learning for a random agent

class aind_behavior_gym.dynamic_foraging.agent.random_agent.RandomAgentBiasedIgnore(seed=None)

Bases: RandomAgent

A biased agent with ignores

act(*args)

Random choose from three actions

Module contents

Agent module