myrlkit 0.1.2

A tiny, teaching-focused reinforcement learning kit (Q-learning, SARSA, Bandits) with toy environments.

myrlkit

A minimal, lightweight reinforcement learning toolkit — built for quick experimentation, small projects, and clear, dependency-light code.
Includes core RL algorithms and simple toy environments with ready-to-run examples and matplotlib visualizations.

✨ Features

• Algorithms: Q-learning, SARSA, and Epsilon-Greedy Bandit.
• Environments: GridWorld and K-armed Bandit.
• Visualization: Built-in learning curves and policy renderings.
• No heavy frameworks — just Python + NumPy + Matplotlib.
• Readable & well-documented — easy to extend and adapt.

Install

pip install myrlkit

If installing from source:

pip install -e .

Quickstart

Q-learning on GridWorld

import numpy as np
from myrlkit.agents import QLearningAgent
from myrlkit.envs import GridWorld

env = GridWorld(width=5, height=5, start=(0,0), goal=(4,4), obstacles=[(1,1), (1,2), (2,1)])
agent = QLearningAgent(state_size=env.n_states, action_size=env.n_actions, alpha=0.5, gamma=0.99, epsilon=0.1)

episodes = 300
rewards = []
for _ in range(episodes):
    s = env.reset()
    done = False
    total = 0.0
    while not done:
        a = agent.choose_action(s)
        ns, r, done, _ = env.step(a)
        agent.update(s, a, r, ns, done)
        s = ns
        total += r
    rewards.append(total)

policy = env.render_policy(agent.q_table)
print(policy)

SARSA on GridWorld

from myrlkit.agents import SARSAAgent
from myrlkit.envs import GridWorld

env = GridWorld(width=4, height=4, start=(0,0), goal=(3,3))
agent = SARSAAgent(state_size=env.n_states, action_size=env.n_actions, alpha=0.5, gamma=0.99, epsilon=0.1)

Epsilon-Greedy Bandit

from myrlkit.agents import EpsilonGreedyBandit
from myrlkit.envs import KArmedBandit

env = KArmedBandit(k=10, means=[0.0]*10, std=1.0, seed=42)
agent = EpsilonGreedyBandit(k=env.k, epsilon=0.1)

rewards = []
for t in range(1000):
    a = agent.select_action()
    r = env.pull(a)
    agent.update(a, r)
    rewards.append(r)

API

• myrlkit.agents.QLearningAgent
• myrlkit.agents.SARSAAgent
• myrlkit.agents.EpsilonGreedyBandit
• myrlkit.envs.GridWorld
• myrlkit.envs.KArmedBandit

Examples

See examples/ for runnable scripts (learning curves, policy printouts).

License

MIT

---

Author

Nihar Lohar
LinkedIn: https://www.linkedin.com/in/nihar-lohar-20234a287/ github: https://github.com/Nihar3453/passport_ocr