evorl.algorithms.a2c

Module Contents

Classes

A2CAgent
A2CNetworkParams	Contains training state for the learner.
A2CWorkflow

Functions

make_mlp_a2c_agent

API

class evorl.algorithms.a2c.A2CAgent

Bases: evorl.agent.Agent

compute_actions(agent_state: evorl.agent.AgentState, sample_batch: evorl.sample_batch.SampleBatch, key: chex.PRNGKey) → tuple[evorl.types.Action, evorl.types.PolicyExtraInfo]

compute_values(agent_state: evorl.agent.AgentState, sample_batch: evorl.sample_batch.SampleBatch) → chex.Array

continuous_action: bool

None

evaluate_actions(agent_state: evorl.agent.AgentState, sample_batch: evorl.sample_batch.SampleBatch, key: chex.PRNGKey) → tuple[evorl.types.Action, evorl.types.PolicyExtraInfo]

init(obs_space: evorl.envs.Space, action_space: evorl.envs.Space, key: chex.PRNGKey) → evorl.agent.AgentState

loss(agent_state: evorl.agent.AgentState, sample_batch: evorl.sample_batch.SampleBatch, key: chex.PRNGKey) → evorl.types.LossDict

property normalize_obs

obs_preprocessor: Any

‘pytree_field(…)’

policy_network: flax.linen.Module

None

value_network: flax.linen.Module

None

class evorl.algorithms.a2c.A2CNetworkParams

Bases: evorl.types.PyTreeData

Contains training state for the learner.

policy_params: evorl.types.Params

None

value_params: evorl.types.Params

None

class evorl.algorithms.a2c.A2CWorkflow(env: evorl.envs.Env, agent: evorl.agent.Agent, optimizer: optax.GradientTransformation, evaluator: evorl.evaluators.Evaluator, config: omegaconf.DictConfig)

Bases: evorl.workflows.OnPolicyWorkflow

learn(state: evorl.types.State) → evorl.types.State

classmethod name()

step(state: evorl.types.State) → tuple[evorl.metrics.MetricBase, evorl.types.State]

evorl.algorithms.a2c.make_mlp_a2c_agent(action_space: evorl.envs.Space, actor_hidden_layer_sizes: tuple[int] = (256, 256), critic_hidden_layer_sizes: tuple[int] = (256, 256), normalize_obs: bool = False, policy_obs_key: str = '', value_obs_key: str = '') → evorl.algorithms.a2c.A2CAgent