Q-Learning Agent playing1 FrozenLake-v1
This is a trained model of a Q-Learning agent playing FrozenLake-v1 .
Usage
model = load_from_hub(repo_id="kinkpunk/q-FrozenLake-v1-custom-map-Slippery-edition",
filename="q-learning.pkl")
# Don't forget to change additional attributes
# when you create environment using 4x4 map
env = gym.make('FrozenLake-v1',
desc=["SFFF", "FHHF", "FFHF", "HFFG"],
is_slippery=True)
Training parameters
# Training parameters
n_training_episodes = 105000 # Total training episodes
learning_rate = 0.8 # Learning rate
# Evaluation parameters
n_eval_episodes = 100 # Total number of test episodes
# Environment parameters
env_id = "FrozenLake-v1" # Name of the environment
max_steps = 99 # Max steps per episode
gamma = 0.98 # Discounting rate
eval_seed = [] # The evaluation seed of the environment
# Exploration parameters
max_epsilon = 0.99 # Exploration probability at start
min_epsilon = 0.02 # Minimum exploration probability
decay_rate = 0.009 # Exponential decay rate for exploration prob
Evaluation results
- mean_reward on FrozenLake-v1-4x4self-reported0.89 +/- 0.31