NaRLA

For installation instructions and API documentation please refer to the docs

Run main

narla main.py \
  --trial_settings.results_directory Results \
  --environment_settings.environment CART_POLE \
  --multi_agent_network_settings.layer_settings.neuron_settings.neuron_type POLICY_GRADIENT

Execute runner

Run main.py with a product of all the settings

narla scripts/run_jobs.py \
  --settings.trial_settings.results_directory RunnerResults \
  --environments CART_POLE \
  --gpus 0 1 2 3 \
  --jobs_per_gpu 5 \
  --learning_rates 0.01 0.001 0.0001 \
  --neuron_types DEEP_Q ACTOR_CRITIC POLICY_GRADIENT \
  --number_of_layers 1 2 3 4 5 6 7 8 9 10 \
  --number_of_neurons_per_layer 5 10 15

Citation

This repository accompanies the paper Giving Up Control: Neurons as Reinforcement Learning Agents

@misc{ott2020giving,
    title={Giving Up Control: Neurons as Reinforcement Learning Agents},
    author={Jordan Ott},
    year={2020},
    eprint={2003.11642},
    archivePrefix={arXiv},
    primaryClass={cs.NE}
}