Skip to main content

Datasets for offline deep reinforcement learning

Project description

Minari is the new name of this library. Minari used to be called Kabuki.

Minari is intended to be a Python library for conducting research in offline reinforcement learning, akin to an offline version of Gymnasium or an offline RL version of HuggingFace's datasets library. The goal is to release a fully working beta in late November or early December.

We have a public discord server (which we also use to coordinate development work) that you can join here: https://discord.gg/jfERDCSw.

Installation

pip install numpy cython

pip install git+https://github.com/Farama-Foundation/Minari.git

Downloading datasets

import minari

dataset = minari.download_dataset("LunarLander_v2_test-dataset")

Recreating Gymnasium environments (Coming very soon!)

import gymnasium as gym

env = gym.make(gym.SpecStack(json.loads(dataset.environment_stack)))

Uploading datasets

dataset.save(
    ".datasets/LunarLander-v2-test_dataset.hdf5"
)  # todo: abstract away parent directory and hdf5 extension
dataset = minari.upload_dataset("LunarLander_v2_test-dataset")

Saving to dataset format

It is not the aim of Minari to insist that you use a certain buffer implementation. However, in order to maintain standardisation across the library, we have a standardised format, the MinariDataset class, for saving replay buffers to file.

This converter will have tests to ensure formatting standards

Checking available remote datasets

import minari

minari.list_remote_datasets()

Checking available local datasets

import minari
minari.list_local_datasets()  # todo: implement

Datasets are stored in the .datasets directory in your project directory.


Minari is a shortening of Minarai, the Japanese word for "learning by observation".

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

minari-0.2.0.tar.gz (168.5 kB view hashes)

Uploaded Source

Built Distribution

minari-0.2.0-cp39-cp39-manylinux_2_34_x86_64.whl (861.3 kB view hashes)

Uploaded CPython 3.9 manylinux: glibc 2.34+ x86-64

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page