agent¶

GymAgent & its config.

class cneuromax.projects.neuroevorl_control.agent.GymAgentConfig(env_transfer='${config.env_transfer}', fit_transfer='${config.fit_transfer}', mem_transfer='${config.mem_transfer}', env_name='${space.config.env_name}', hidden_size=50, mutation_std=0.01)[source]¶

Bases: BaseAgentConfig

Holds GymAgent config values.

Parameters:

env_name (str, default: '${space.config.env_name}') – See env_name.
hidden_size (int, default: 50) – Size of the RNN hidden state.
mutation_std (float, default: 0.01) – Standard deviation of the mutation noise.

class cneuromax.projects.neuroevorl_control.agent.GymAgent(config, pop_idx, *, pops_are_merged)[source]¶

Bases: BaseAgent

project BaseAgent.

Parameters:

config (GymAgentConfig)
pop_idx (int) – See pop_idx.
pops_are_merged (bool) – See pops_are_merged.

mutate()[source]¶

Mutates the agent.

Return type:: None

reset()[source]¶

Resets the agent’s memory state.

Return type:: None

env_to_net(x)[source]¶

Processes the observation before feeding it to the network.

Parameters:: x (Float32[Tensor, 'obs_size']) – See x.
Return type:: Float32[Tensor, 'out_size']
Returns:: The observation processed for the network.

net_to_env(x)[source]¶

Processes the network output before feeding it to the env.

Parameters:: x (Float32[Tensor, 'act_size']) – The network output.
Return type:: Float32[Tensor, 'act_size'] | Int64[Tensor, 'act_size']
Returns:: The network output processed for the env.