from dqn import DQN import Environment import numpy as np env_name = 'visual_banana' env = Environment.CollectBanana(env_name, 'Banana.x86') dqn = DQN(env.name, env.state_size, env.action_size, env) env.train_mode = True #scores = dqn.train(n_episodes=2000, target_score=13.0) #np.save(env_name+'_scores.npy', np.array(scores)) env.train_mode = False dqn.play(load=True, steps=2000)
# number of actions action_size = brain.vector_action_space_size # examine the state space state = env_info.vector_observations[0] state_size = len(state) # initialize the Nav Deep Q network agent agent = Agent(state_size=state_size, action_size=action_size, seed=0, pixels=PIXELS) else: # please do not modify the line below root = os.path.dirname(__file__) path = root + "/VisualBanana_Linux/Banana.x86_64" env = Environment.CollectBanana(path) # initialize the Nav Deep Q network agent agent = Agent(state_size=env.state_size, action_size=env.action_size, seed=0, pixels=PIXELS) train = True evaluate = False if train: scores = dqn_train(env, PIXELS, agent, n_episodes=2000, max_t=1000, eps_start=1.0, eps_end=0.01, eps_decay=0.995, train_m=True) if evaluate: