This is the codebase for the AMLD Introduction to Reinforcement Learning with applications to satellite scheduling problems.
Start by cloning the repository by running:
https://github.com/TheoLeFur/AMLD.git
Enter the project directory and create your virtual environment:
conda create -n your_env_name
Install the dependencies using:
pip3 install -r requirements.txt
Setup the project using
pip3 install -e .
We will give a basic implementation of a neural network architecture, trained with Policy Gradients, to solve combinatorial optimization problems. In our case, we will show how the model performs on the Travelling Salesman Problem, an NP hard problem
Run the training script from the scripts
directory:
/bin/bash run_crbl.sh
One can see here a sample solution, for a model trained on 500000 random graphs, each having 16 nodes:
Neural Combinatorial Optimization with Reinforcement Learning
Policy Gradient Methods for Reinforcement Learning with Function Approximation