Welcome to the world of offline reinforcement learning! Introducing Minari, a powerful Python library designed for conducting research in offline reinforcement learning, akin to an offline version of Gymnasium. In this guide, we will walk you through the steps to install Minari, use its command-line API, and dive into basic usage with datasets.

Installation Guide

To get started with Minari, you need to install it from the Python Package Index (PyPI). Here’s how you can do that:

• For the minimum required dependencies, run the following command in your terminal:

pip install minari

• If you want to include all additional dependencies at once, execute this:

pip install minari[all]

• If you’re looking to contribute to Minari or test it from the source, use the following commands:

git clone https://github.com/Farama-Foundation/Minari.git
cd Minari
pip install -e .[all]

Command Line API

Once you have installed Minari, you can use its command line to manage datasets. Here are some useful commands:

• To check available remote datasets:

minari list remote

• To download a specific dataset like D4RL’s doorhuman-v2:

minari download D4RLdoorhuman-v2

• To see available local datasets:

minari list local

• To show the details of a particular dataset:

minari show D4RLdoorhuman-v2

• For a list of all commands available:

minari --help

Basic Usage

Now, let’s look at how you can use Minari to read and write datasets. Think of a dataset as a box of puzzles, where each puzzle piece represents individual observations, actions, rewards, etc. Here’s how you can interact with these dataset box puzzles:

Reading a Dataset

To read a dataset, you first need to load it. Here is how you might do that:

import minari

dataset = minari.load_dataset('D4RLdoorhuman-v2')
for episode_data in dataset.iterate_episodes():
    observations = episode_data.observations
    actions = episode_data.actions
    rewards = episode_data.rewards
    terminations = episode_data.terminations
    truncations = episode_data.truncations
    infos = episode_data.infos
    # Process your episode_data here

Writing a Dataset

To write (or create) your own dataset, think of filling your own puzzle box with pieces that represent an environment’s behavior. Here’s how you can set that up:

import minari
import gymnasium as gym
from minari import DataCollector

env = gym.make('FrozenLake-v1')
env = DataCollector(env)

for _ in range(100):
    env.reset()
    done = False
    while not done:
        action = env.action_space.sample()  # Use your policy here
        obs, rew, terminated, truncated, info = env.step(action)
        done = terminated or truncated
        
dataset = env.create_dataset('frozenlaketest-v0')

For more examples, check the Basic Usage section of the documentation. For an in-depth tutorial on creating new datasets, see our Pointmaze D4RL Dataset tutorial.

Troubleshooting

If you encounter any issues while installing or running Minari, here are some troubleshooting tips:

• Ensure that your Python environment is set up correctly. Sometimes conflicts arise from existing packages.
• If the installation fails, check your internet connection and try again.
• Refer to the Minari documentation for detailed guidance on specific commands.
• If you’re still stuck, feel free to reach out through our public Discord server for help or to coordinate development work.

For more insights, updates, or to collaborate on AI development projects, stay connected with fxis.ai.

Conclusion

At fxis.ai, we believe that such advancements are crucial for the future of AI, as they enable more comprehensive and effective solutions. Our team is continually exploring new methodologies to push the envelope in artificial intelligence, ensuring that our clients benefit from the latest technological innovations.