memory = Memory(REPLAY_SIZE)
agent = Agent(env, memory)
initial_observation = env.reset()
if 'cuda' in str(device):
print('The GPU is being used')
else:
print('The CPU is being used')
if option_dict['random']:
play_random(env, UP_ACTION, DOWN_ACTION, seconds=5)
if option_dict['train']:
print("Training")
print("ReplayMemory will require {}gb of GPU RAM".format(round(REPLAY_SIZE * 32 * 84 * 84 / 1e+9, 2)))
agent.reset_environtment()
train(env, net, target_net, epsilon_data, agent, memory, GAMMA, device,
DELAY_LEARNING, TARGET_UPDATE_FREQ, BATCH_SIZE, model)
if option_dict['oldnetwork']:
file_path = './pull/pong_v4_data/DQN/DQN_10_6-1700.dat'
seconds = 120
test_old_network(env, net, file_path, seconds, device)
