Simon ARC Env

January 12, 2024 ยท View on GitHub

simon arc env

Action Space Discrete(10)
Observation Shape (32, 32)
Observation High 255
Observation Low 0
Import import simon_arc_env
gymnasium.make("SimonARC-v0")

Description

Abstraction and Reasoning Corpus as a Farama Gymnasium environment.

Related project: ARCLE - ARC Learning Environment.

Installation

pip install simon-arc-env

Usage

  1. Play it by running
python examples/play.py

Press space to draw a pixel.

  1. Import it to train your RL model
import simon_arc_env
env = gymnasium.make("SimonARC-v0")

The package relies on import side-effects to register the environment name so, even though the package is never explicitly used, its import is necessary to access the environment.

Action Space

SimonARC has the action space Discrete(10).

ValueKeyMeaning
0TABNext page
1ReturnSubmit drawing
2SpaceSet pixel
3Arrow upMove cursor_y up
4Arrow downMove cursor_y down
5Arrow leftMove cursor_x left
6Arrow rightMove cursor_x right
7mIncrement color
8nDecrement color
9sAdjust canvas size
N/AEscapeExit game
N/AiPrint info to console

Observation Space

Observation space Box(low=0, high=255, shape=(32, 32), dtype=np.uint8).

Rewards

If you solve the task you get 10000 in reward.

If you predict the width correct you get 100 in reward.

If you predict the height correct you get 100 in reward.

If your prediction have the correct histogram you get 100 in reward.

Run tests

make test

Version History

  • v0: initial version release