# Start the environment NUM_CREEPS = 30 game = WaterWorld(width=500, height=500, num_creeps=NUM_CREEPS) p = PLE(game, fps=30, display_screen=True, add_noop_action=False, force_fps=False) game.ple = p p.init() # Start the agent action_set = p.getActionSet() agent = QLearnerEvolver(len(action_set)) p.state_preprocessor = agent.process_state #agent.load("model.h5") #agent.epsilon = 0.05 fail, catch, j = 0, 0, 0 best_score = -np.inf nb_games = 1 while 1: j += 1 # On réinitialise de temps en temps if p.game_over() or j == 50000: fail, catch, j = 0, 0, 0 best_score = max(best_score, p.score())