SARSA Agent in Frozen Lake Environment

import numpy as np
import matplotlib.pyplot as plt
import gymnasium as gym
from gymnasium.envs.toy_text.frozen_lake import generate_random_map

from rlforge.agents.tabular import SarsaAgent
from rlforge.experiments import ExperimentRunner

env = gym.make("FrozenLake-v1",
               is_slippery=False,
               render_mode="rgb_array",
               desc=generate_random_map(size=4, p=0.9))
agent = SarsaAgent(step_size=0.125,
                   discount=1,
                   num_states=env.observation_space.n,
                   num_actions=env.action_space.n,
                   epsilon=0.1)

runner = ExperimentRunner(env, agent)

results = runner.run_episodic(
    num_runs=100,
    num_episodes=500,
    max_steps_per_episode=10000
)

runner.summary()

============================================================
 Experiment Summary (Episodic)
============================================================
Runs:                    100
Total Runtime:           9.565 seconds
Average Runtime/Run:     0.096 seconds
------------------------------------------------------------
Episodes per run (Max):  500

REWARDS
First Episode:           0.330 ± 0.470
Last Episode:            0.960 ± 0.196
Overall Mean:            0.973 ± 0.163
Last 10 Episodes:        0.974 ± 0.159

STEPS
First Episode:           29.3 ± 21.1
Last Episode:            6.5 ± 1.0
Overall Mean:            7.1 ± 3.8
============================================================

runner.plot_results(window_size=1, max_reward=1)

runner.plot_results(metric='step', window_size=1)