base

BaseSpace and its config.

class cneuromax.fitting.neuroevolution.space.base.BaseSpaceConfig(eval_num_steps='${config.eval_num_steps}', logging='${config.logging}')

Bases: object

Holds BaseSpace config values.

Parameters:

• eval_num_steps (int, default: '${config.eval_num_steps}') – See eval_num_steps.

• logging (bool, default: '${config.logging}') – See logging.

class cneuromax.fitting.neuroevolution.space.base.BaseSpace(config, num_pops, *, evaluates_on_gpu)

Bases: ABC

Space Base class.

A Space is a torchrl environment wrapper with which agents produce behaviour and receive fitness scores.

Parameters:

• config (BaseSpaceConfig)

• num_pops (int) – Number of agents interacting with each other in a given space.

• evaluates_on_gpu (bool) – Whether GPU devices are used to evaluate agents.

abstract evaluate(agents, curr_gen)

.

Method called once per iteration (every generation) in order to evaluate and attribute fitnesses to agents.

Parameters:

• agents (list[list[BaseAgent]]) – Agent(s) to evaluate.

• curr_gen (int) – The current generation number/index.

Return type:

ndarray[float32, Any]

Returns:

The fitnesses and number of steps ran.