evolve

Evolutionary operations for Neuroevolution fitting.

The selection operation is implicit in cneuromax, see update_exchange_and_mutate_info() for more details.

cneuromax.fitting.neuroevolution.utils.evolve.mutate(agents_batch, exchange_and_mutate_info_batch, num_pops)

Mutate agents_batch.

Parameters:

• agents_batch (list[list[BaseAgent]]) – See agents_batch.

• exchange_and_mutate_info_batch (ndarray[Shape[Len_agents_batch, Num_pops, [mpi_buffer_size, agent_pair_position, sending, seeds]], dtype[uint32]]) – A sub-array of exchange_and_mutate_info maintained by this process.

• num_pops (int) – See num_pops().

Return type:

None

cneuromax.fitting.neuroevolution.utils.evolve.evaluate_on_cpu(agents_batch, space, curr_gen)

Evaluate agents_batch.

Parameters:

• agents_batch (list[list[BaseAgent]]) – See agents_batch.

• space (BaseSpace) – The BaseSpace instance used throughout the execution.

• curr_gen (int) – See curr_gen.

Return type:

ndarray[Shape[Len_agents_batch, Num_pops, [fitness, num_env_steps]], dtype[float32]]

Returns:

The output of agent evaluation performed by the process calling

this function on the agents it maintains (agents_batch). See evaluate().

cneuromax.fitting.neuroevolution.utils.evolve.evaluate_on_gpu(ith_gpu_comm, agents_batch, space, curr_gen, *, transfer)

Gather agents_batch on process subset & evaluate.

Parameters:

• ith_gpu_comm (Comm) – A mpi4py communicator used by existing CPU processes to exchange agents for GPU work queueing.

• agents_batch (list[list[BaseAgent]]) – See agents_batch.

• space (BaseSpace) – See space.

• curr_gen (int) – See curr_gen.

• transfer (bool) – Whether any of env_transfer, fit_transfer or mem_transfer is True.

Return type:

ndarray[Shape[Len_agents_batch, Num_pops, [fitness, num_env_steps]], dtype[float32]]

Returns:

The output of agent evaluation by this process. See

evaluate().