python >= 3.7 pytorch >= 1.7.0 numpy >= 1.19 cython == 0.29.21 gym == 0.18.0 matplotlib == 3.3.2 baselines == 0.1.6 a g++ complier compatible with C++11 standard
env2.pyx,EnvCWrapper.pxd: Cython code; define the interface of MPS with python training module. environment.cpp: The MPS implementation, which receives agents' action as input and returns observations and rewards. config.cpp,config.h: Define the useful mappings, important hyperparameters and the setup with respect to the dataset for MPS. facility.cpp,facility.h: Define the facility object with crucial properties such as capacity and types. individual.cpp,individual.h: Define the individual object with crucial properties such as health state, supply level, affiliation to facility and age. city.cpp,city.h: Define the city object, which serves as a global counter for agents in different health states.
a2c_ppo_acktr/envs: Calls baselines function to wrap the pyx class into a gym-compatible environment. Other files under a2c_ppo_acktr folder are deprecated. DQN.py,Storage.py: The algorithm (and buffer) implementation of SMADQN. config.py: Contains hyperparameters for RL training.
main.py: Main code that integrates the training and simulation pipeline. utils.py: Useful functions for python code. draw.py: Visualization in python using debugging information from the C++ code of MPS. build.py: Script for Cython compilation. mkfl.sh: This script calls for the compilation the C++ code and runs main.py; run this script for experiments.
graph_sorted.txt: The dataset for experiment.
The hyperparameters controlling the current type of experiment are listed at the beginning of environment.cpp; alter these to get different results. After modifying the code, run mkfl.sh to start the execution.