memorax.algorithms.PQN

class memorax.algorithms.PQN

Bases: object

PQN(cfg: memorax.algorithms.pqn.PQNConfig, env: gymnax.environments.environment.Environment, env_params: gymnax.environments.environment.EnvParams, q_network: flax.linen.module.Module, optimizer: optax._src.base.GradientTransformation, epsilon_schedule: collections.abc.Callable[[typing.Union[jax.jaxlib._jax.Array, numpy.ndarray, numpy.bool, numpy.number, float, int]], typing.Union[jax.jaxlib._jax.Array, numpy.ndarray, numpy.bool, numpy.number, float, int]])

cfg: PQNConfig

env: Environment

env_params: EnvParams

q_network: Module

optimizer: GradientTransformation

epsilon_schedule: Callable[[Array | ndarray | bool | number | float | int], Array | ndarray | bool | number | float | int]

init(key)

Return type:

tuple[Array, PQNState, Array, EnvState]

warmup(key, state, num_steps)

Return type:

tuple[Array, PQNState]

Parameters:

• key (Array)

• state (PQNState)

• num_steps (int)

train(key, state, num_steps)

Return type:

tuple[Array, PQNState, dict]

Parameters:

• key (Array)

• state (PQNState)

• num_steps (int)

evaluate(key, state, num_steps)

Return type:

tuple[Array, dict]

Parameters:

• key (Array)

• state (PQNState)

• num_steps (int)

__init__(cfg, env, env_params, q_network, optimizer, epsilon_schedule)

Parameters:

• cfg (PQNConfig)

• env (Environment)

• env_params (EnvParams)

• q_network (Module)

• optimizer (GradientTransformation)

• epsilon_schedule (Callable[[Array | ndarray | bool | number | float | int], Array | ndarray | bool | number | float | int])

Return type:

None

replace(**updates)

Returns a new object replacing the specified fields with new values.