train.py

"""
For this document, we will setup a basic RL pipeline using our SinglePolicySumoEnv environment.

Refer to this recent and similar SumoRL tool that has an example for MARL using RlLib:
https://github.com/LucasAlegre/sumo-rl/blob/master/experiments/a3c_4x4grid.py

Ray RlLib agent training example.
https://github.com/ray-project/ray/blob/master/rllib/examples/custom_train_fn.py
"""
from seal.trainer.fed_agent import FedPolicyTrainer
from seal.trainer.multi_agent import MultiPolicyTrainer
from seal.trainer.single_agent import SinglePolicyTrainer
from os.path import join


if __name__ == "__main__":
    n_episodes = 1  # 100
    fed_step = 5
    net_files = [
        join("configs", "complex_inter", "complex_inter.net.xml"),
        join("configs", "single_inter", "single_inter.net.xml"),
        join("configs", "two_inter", "two_inter.net.xml")
    ]

    for net_file in net_files:
        for ranked in [True, False]:
            print(">> Training with `FedPolicyTrainer`!")
            FedPolicyTrainer(fed_step=fed_step, net_file=net_file, ranked=ranked).\
                train(n_episodes)
            print(">> Training with `MultiPolicyTrainer`!")
            MultiPolicyTrainer(net_file=net_file, ranked=ranked).\
                train(n_episodes)
            print(">> Training with `SinglePolicyTrainer`!")
            SinglePolicyTrainer(net_file=net_file, ranked=ranked).\
                train(n_episodes)