# TODO: load DQN agent
# ...
state_dim = 4
num_actions = 2
Q = MLP(state_dim, num_actions)
Q_target = MLP(state_dim, num_actions)
agent = DQNAgent(Q, Q_target, num_actions)
agent.load("C:\\Users\\Monish\\Desktop\\workspace\\exercise3_R\\reinforcement_learning\\models_cartpole\\dqn_agent_199.pt")
n_test_episodes = 15
episode_rewards = []
for i in range(n_test_episodes):
stats = run_episode(env, agent, deterministic=True, do_training=False, rendering=True)
episode_rewards.append(stats.episode_reward)
# save results in a dictionary and write them into a .json file
results = dict()
results["episode_rewards"] = episode_rewards
results["mean"] = np.array(episode_rewards).mean()
results["std"] = np.array(episode_rewards).std()
if not os.path.exists("./results"):
os.mkdir("./results")
fname = "./results/cartpole_results_dqn-%s.json" % datetime.now().strftime("%Y%m%d-%H%M%S")
fh = open(fname, "w")
json.dump(results, fh)
num_actions = 2
Q_network = MLP(state_dim, num_actions)
Q_target_network = MLP(state_dim, num_actions)
agent = DQNAgent(Q=Q_network,
Q_target=Q_target_network,
num_actions=num_actions)
agent.load(args.model)
n_test_episodes = args.episodes
episode_rewards = []
for i in range(n_test_episodes):
stats = run_episode(env,
agent,
deterministic=True,
do_training=False,
rendering=True,
max_timesteps=250)
episode_rewards.append(stats.episode_reward)
print('Episode %d (reward: %d)' % (i, stats.episode_reward))
# save results in a dictionary and write them into a .json file
results = dict()
results["episode_rewards"] = episode_rewards
results["mean"] = np.array(episode_rewards).mean()
results["std"] = np.array(episode_rewards).std()
if not os.path.exists("./results"):
os.mkdir("./results")
fname = "./results/cartpole_results_dqn-%s.json" % datetime.now().strftime(
