ADRL

Folder Structure

ADRL
├── train (Directory where the training files are)
│   ├── train.py
├── test (Directory where the testing files are)
│   ├── test.py
├── environment (Directory where the environment files are)
│   ├── environment.py
├── benchmarks (Directory where the binomial and binomial smoothing filters are tested for AD)
│   ├── binomial_filters.py
├── models (Directory where the trained policy networks are stored)
│   ├── policy_network.pth
├── plotting (Directory where the plotting files for the plots in the paper are stored)
│   ├── plot_actions.py
│   ├── plot_cumulative_rewards.py
│   ├── plot_energy_spectrum_filters.py
│   ├── plot_energy_spectrum_initial.py
│   ├── plot_rewards.py
│   ├── plot_training_reward.py
│   ├── plot_transfer_function.py
├── figures (Directory where the figures are stored)
├── arrays (Directory where the arrays from the simulations are stored)

Commands for Training and Testing on Local Computer

Training

In train.py the training parameters such as the entropy coefficent, epochs, learning rate, number of parallel environments, etc. can be adjusted in the argparse function.

cd train/
python3 train.py

Testing

cd test/
python3 test.py

Commands for Training RL Model on Euler

Loading Modules in Euler

env2lmod
module load gcc/8.2.0 python/3.10.4 hdf5/1.10.1 eth_proxy

Training RL Model with N CPU's (running the simulations in parallel) and M GPU (training the policy network)

cd train/
sbatch --time=04:00:00 --gpus=M --ntasks=N --mem-per-cpu=1024 --wrap="python3 train.py"