backgammon/main.py

import argparse
import sys
import time

def print_train_outcome(outcome, trained_eps = 0):
    format_vars = { 'trained_eps': trained_eps,
                    'count': len(train_outcome),
                    'sum': sum(train_outcome),
                    'mean': sum(train_outcome) / len(train_outcome),
                    'time': int(time.time())
    }
    print("train;{time};{trained_eps};{count};{sum};{mean}".format(**format_vars))

def print_eval_outcomes(outcomes, trained_eps = 0):
    for outcome in outcomes:
            scores = outcome[1]
            format_vars = { 'trained_eps': trained_eps,
                            'method': outcome[0],
                            'count': len(scores),
                            'sum': sum(scores),
                            'mean': sum(scores) / len(scores),
                            'time': int(time.time())
            }
            print("eval;{time};{method};{trained_eps};{count};{sum};{mean}".format(**format_vars))

parser = argparse.ArgumentParser(description="Backgammon games")
parser.add_argument('--episodes', action='store', dest='episode_count',
                    type=int, default=1000,
                    help='number of episodes to train')
parser.add_argument('--model-path', action='store', dest='model_path',
                    default='./model',
                    help='path to Tensorflow model')
parser.add_argument('--eval-methods', action='store',
                    default=['random'], nargs='*',
                    help='specifies evaluation methods')
parser.add_argument('--eval', action='store_true',
                    help='whether to evaluate the neural network with a random choice bot')
parser.add_argument('--train', action='store_true',
                    help='whether to train the neural network')
parser.add_argument('--play', action='store_true',
                    help='whether to play with the neural network')

args = parser.parse_args()

config = {
    'model_path': args.model_path,
    'episode_count': args.episode_count,
    'eval_methods': args.eval_methods,
    'train': args.train,
    'play': args.play,
    'eval': args.eval
}

#print("-"*30)
#print(type(args.eval_methods))
#print(args.eval_methods)
#print("-"*30)

import game
g = game.Game(config = config)
g.set_up_bots()

episode_count = args.episode_count

if args.train:
    eps = 0
    while True:
        train_outcome = g.train_model(episodes = episode_count, trained_eps = eps)
        eps += episode_count
        print_train_outcome(train_outcome, trained_eps = eps)
        if args.eval:
            eval_outcomes = g.eval(trained_eps = eps)
            print_eval_outcomes(eval_outcomes, trained_eps = eps)
        sys.stdout.flush()
elif args.eval:
    outcomes = g.eval()
    print_eval_outcomes(outcomes, trained_eps = 0)
#elif args.play:
#    g.play(episodes = episode_count)