Skip to main content

Gameworld 10k

The Gameworld environments are designed to develop sample-efficient learning algorithms, which is enforced in the challenge by limiting the interactions to 10K steps. These environments could be solved by humans within minutes, ensuring that learning does not hinge on brittle exploration or complex credit assignment.

Gameworld also supports controlled perturbations such as changes in object color or shape, testing an agent's ability to generalize across superficial domain shifts.

Environments

The suite includes 10 diverse games generated with the aid of a large language model, drawing inspiration from ALE and classic video games, while maintaining a lightweight and structured design.

AviateBounceCrossDriveExplode
FruitsGoldHuntImpactJump

Installation

To install the gameworld library, either use the pypi registry pip install gameworld, or install from source by cloning the repository and running pip install -e ..

This library has been developed and tested onpython3.11 for both Linux and macOS.

API

The Gameworld environments use the gymnasium.Gym api. To run your own algorithm against our environments, create an environment instance as:

import gameworld.envs # Triggers registering the environments in Gymnasium
import gymnasium

game = "Aviate"
env = gymnasium.make(f"Gameworld-{game}-v0")

obs, info = env.reset()

for t in range(10_000):
    # random actions as example
    action = env.action_space.sample()

    # step env
    obs, reward, done, truncated, info = env.step(action)

    # reset when done
    if done:
        obs, info = env.reset()

Perturbations

The Gameworld environments support controlled perturbations in order. For each environment, you can choose no (None), shape (shape), or a color (color) perturbation.

For example, a shape perturbation after 5000 steps in the Explode environment can be created using the following snippet:

env = gymnasium.make(
    f"Gameworld-Explode-v0", perturb='shape', perturb_step=5000
)

Below we show an example of a shape and color perturbation on Explode and Fruits:

Explode ColorExplode ShapeFruits ColorFruits Shape

Metadata

Release files for gameworld 0.1.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for gameworld 0.1.0
File Size Uploaded
gameworld-0.1.0.tar.gz 2.4 MB Details

Built distribution (wheel)

Table of built distributions (wheels) for gameworld 0.1.0
File Interpreter ABI Platform
gameworld-0.1.0-py3-none-any.whl Python 3 none any Details

Total release size: 2.5 MB

Release files / gameworld-0.1.0.tar.gz

Download URL gameworld-0.1.0.tar.gz
Size 2.4 MB
Tags Source
SHA-256 checksum
How to use checksums
ffda9b8e06f405bc1255888d0c9d192d15e0025bad3140db29ecf36538ca1d65
BLAKE2b-256 checksum
How to use checksums
cee3ff0ac23c10960b88f821d7a3f535921e15c1c502f137252dc391901ebef7
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/6.1.0 CPython/3.11.11

Release files / gameworld-0.1.0-py3-none-any.whl

Download URL gameworld-0.1.0-py3-none-any.whl
Size 35.3 kB
Tags Python 3
SHA-256 checksum
How to use checksums
4153803fa7e3996ccfd42cd667c5af4a6cc700cd2f9af2662d7b2690d23d8836
BLAKE2b-256 checksum
How to use checksums
998cf3523ccc6244e778fd8bd7aa49fd18498df80e6f687cc154bf293c6274eb
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/6.1.0 CPython/3.11.11

Release history Release notifications | RSS feed

This release

0.1.0 This release

2 release files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page