run_plots.py

# external imports
import argparse

import gym

# internal import
from src.analysis.launch_plots import launch_plots
from src.learning.actor import Actor
from run_experiment import ENV_LIST


##############
# Read inputs

def read_input():
    """
    Read the inputs as an arg parser.
    """

    parser = argparse.ArgumentParser()

    parser.add_argument("--save_path", default="none", type=str, help="Path where to save results")
    parser.add_argument("--results_path", default=".", type=str, help="Path to store analysis results")
    parser.add_argument("--verbose", action="store_true")

    parser.add_argument("--env", default="all", type=str, help="Env names separated with | (default: all)")
    parser.add_argument("--algo", default="all", type=str, help="Algo names separated with | (default: all)")

    parser.add_argument("--max_evals", default=1000000, type=int, help="Nr of evaluations for the archive to plot")
    parser.add_argument("--eval_batch_size", default=100, type=int, help="Nr individuals per generation")
    parser.add_argument("--save_stat_period", default=20000, type=int, help="Period to save metrics")

    parser.add_argument("--min_fitness", default="", type=str, help="Min-fitness for analysis plots (default: none)")
    parser.add_argument("--max_fitness", default="", type=str, help="Max-fitness for analysis plots (default: none)")

    parser.add_argument("--neurons_list", default="128 128", type=str, help="Actor NN: [neurons_list + [action dim]]")

    parser.add_argument("--visualisation", action="store_true", help="Plot the visualisation graphs")
    parser.add_argument("--save_videos", action="store_true", help="Save visualisation videos")

    parser.add_argument("--p_values", action="store_true", help="Compute p-values")

    args = parser.parse_args()
    args.neurons_list = [int(x) for x in args.neurons_list.split()]

    return args


#######
# Main

if __name__ == "__main__":

    args = read_input()

    # Read algos and envs
    algo_list = [] if args.algo == "all" else args.algo.rstrip().split("|")
    env_list = ENV_LIST if args.env == "all" else args.env.rstrip().split("|")

    # Create Actor instance for each environment
    env_actors = {}
    for env_name in env_list:
        assert env_name in ENV_LIST, "!!!ERROR!!! Unvalid environment name" + str(
            env_name
        )
        env = gym.make(env_name)
        state_dim = env.observation_space.shape[0]
        action_dim = env.action_space.shape[0]
        max_action = float(env.action_space.high[0])
        env_actors[env_name] = Actor(
            state_dim,
            action_dim,
            max_action,
            args.neurons_list,
        )
        env.close()

    # Start plots
    launch_plots(
        args.save_path,
        algo_list,
        args.max_evals,
        args.eval_batch_size,
        args.save_stat_period,
        env_list,
        env_actors,
        save_path=args.results_path,
        min_fit=None if args.min_fitness == "" else float(args.min_fitness),
        max_fit=None if args.max_fitness == "" else float(args.max_fitness),
        archive=not(args.visualisation),
        progress=not(args.visualisation),
        variation=not(args.visualisation),
        visualisation=args.visualisation,
        save_videos=args.save_videos,
        p_values=args.p_values,
        verbose=args.verbose,
    )