print("Average time steps per episode: {total_epochs / episodes}")
# Load a Windy GridWorld environment
env_name = "LunarLander-v2"
env = gym.make(env_name)
random_agent = RandomDQNAgent(env_name, env, 1000, is_state_box=True, memory_capacity=100000)
# random_agent.train()
agent = DQNAgent(env_name, env, 5000, learning_rate=0.00025, start_epsilon=1.0, discount_factor=0.99, decay_rate=0.0001,
make_checkpoint=True, is_state_box=True, batch_size=64, memory_capacity=100000)
# agent.memory = random_agent.memory
# agent.train()
weights, rewards, episode_len = agent.load("/home/dsalwala/NUIG/Thesis/rl-algos/data/LunarLander-v2_100.npy")
stats = plotting.EpisodeStats(
episode_lengths=episode_len,
episode_rewards=rewards)
# Search for a Q values
# nn, stats = agent.nn.get_weights(), agent.stats
nn = ANN(8, 4, 0.00025)
nn.set_weights(weights)
play_episode(env, nn, 1)
env.close()
plotting.plot_episode_stats(stats)
agent = DQNAgent(env_name,
env,
5000,
learning_rate=0.00025,
start_epsilon=1.0,
discount_factor=0.99,
decay_rate=0.0001,
make_checkpoint=True,
is_state_box=True,
batch_size=64,
memory_capacity=100000)
# agent.memory = random_agent.memory
# agent.train()
weights, rewards, episode_len = agent.load(
"/home/dsalwala/NUIG/Thesis/rl-algos/data/CartPole-v0_3000.npy")
stats = plotting.EpisodeStats(episode_lengths=episode_len,
episode_rewards=rewards)
# Search for a Q values
# nn, stats = agent.nn, agent.stats
nn = ANN(4, 2, 0.00025)
nn.set_weights(weights)
play_episode(env, nn, 1)
env.close()
plotting.plot_episode_stats(stats)