taigaoma1997 / omlppo

The RL course project code

OMLPPO

This is the EECS 598 - 002 RL course project for the PyTorch implementation for OML-PPO (https://iopscience.iop.org/article/10.1088/2632-2153/abc327) without spinningup

Prerequisites

• python=3.7.4
• torch=1.3.1
• torchvision=0.5.0
• tmm=0.1.7

Make environments

• make environment: conda create --name py37 python=3.7
• make RLMultilayer package: pip install -e .
• Install any necessary packages.

Run experiments

Max length = 6:
python ppo_absorber_visnir.py --cpu 16 --maxlen 6 --exp_name absorber6 --use_rnn --discrete_thick --num_runs 1 --env PerfectAbsorberVisNIR-v0

Max length = 15:
python ppo_absorber_visnir.py --cpu 16 --maxlen 15 --exp_name perfect_absorber15 --use_rnn --discrete_thick --num_runs 1 --env PerfectAbsorberVisNIR-v1

To do list:

• 1. Add perturbation to thickness
• 2. Add extra rewards at middle steps