README.md

metadata

tags:
  - FrozenLake-v2-4x4
  - q-learning
  - reinforcement-learning
  - custom-implementation
model-index:
  - name: q-FrozenLake-v2-4x4-Slippery
    results:
      - task:
          type: reinforcement-learning
          name: reinforcement-learning
        dataset:
          name: FrozenLake-v2-4x4
          type: FrozenLake-v2-4x4
        metrics:
          - type: mean_reward
            value: 0.51 +/- 0.50
            name: mean_reward
            verified: false

Q-Learning Agent playing FrozenLake-v2

• This is a trained model of a Q-Learning agent playing FrozenLake-v2. The goal of the agent is to navigate from starting state to the goal state by walking only on Frozen tiles and avoid holes.

• The environment support two different sizes: 4x4 and 8x8grids

• The environment has two modes:

  • is_slippery=True agent may not move in the desired direction due to the slippery nature of the tails.
  • is_slippery=False agent moves in the desired direction due to the non-slippery nature of the tails.

• How to create the FrozenLake environment?

  • gym.make("FrozenLake-v1", desc=None, map_name="4x4", is_slippery=True, render_mode="rgb_array")
  • desc=None specify non-predetermined maps, map_name will be used. You can also specify custom map like desc = ["SFFH", "HFHF", "FFHF", "HFFG"]. If both are None, random 8x8 map with 80% of locations frozen will be generated.

Usage

model = load_from_hub(repo_id="HugeFighter/q-FrozenLake-v2-4x4-Slippery", filename="q-learning.pkl")

# Don't forget to check if you need to add additional attributes (is_slippery=False etc)
env = gym.make(model["env_id"])