arckit

1.0.1 verified Sat May 09 auth: no python

Tools for working with the Abstraction & Reasoning Corpus (ARC-AGI), including loading datasets (ARC-AGI-1, ARC-AGI-2), task manipulation, evaluation, and visualization. Current version: 1.0.1. Maintained regularly with releases.

pip install arckit

Common errors

error ModuleNotFoundError: No module named 'arckit' ↓

cause Package not installed or installed in wrong environment.

fix

Run 'pip install arckit' in your active Python environment.

error AttributeError: module 'arckit' has no attribute 'solve' ↓

cause Incorrect import of solve_one_task from a submodule that does not exist.

fix

Use 'from arckit import solve_one_task'.

error ValueError: Unknown dataset: 'arc-agi-2' ↓

cause Using an older version of arckit (<1.0.1) that does not support ARC-AGI-2.

fix

Upgrade arckit with 'pip install --upgrade arckit' to v1.0.1+.

Warnings

breaking Default dataset changed from ARC-AGI-1 to ARC-AGI-2 in v1.0.1. Code relying on ARC-AGI-1 tasks must specify version string. ↓

fix Use load_tasks('arc-agi-1') for original ARC-AGI-1 tasks.

breaking Default evaluation topn changed from 1 to 2 in v1.0.1, affecting metrics. ↓

fix Explicitly set topn=1 in evaluation functions if you need the old behavior.

gotcha solve_one_task expects a function that takes a dict task and returns a 2D list or numpy array. Returning incorrect shape silently may produce wrong metrics. ↓

fix Ensure solver returns a grid of same height/width as expected output for the test pair.

Imports

load_tasks
```
from arckit import load_tasks
```
Common mistake: importing from arckit.data or arckit.util incorrectly; the main API is top-level.

solve_one_task

wrong

from arckit.solve import solve_one_task

correct

from arckit import solve_one_task

Old import path; solve_one_task is now at top-level.

print_grid

wrong

import arckit.print_grid

correct

from arckit.vis import print_grid

print_grid is in the vis submodule.

Quickstart

Loads ARC-AGI tasks and runs a dummy solver. Default dataset is ARC-AGI-2.

from arckit import load_tasks, solve_one_task

# Load tasks (defaults to ARC-AGI-2)
tasks = load_tasks()
# Or load ARC-AGI-1:
# tasks = load_tasks('arc-agi-1')

# Pick a training task
task = tasks[0]

# Solve the first training pair with a dummy prediction
def dummy_solve(task):
    # Return the second grid of the first pair
    return task.train[0][1]

result = solve_one_task(task, dummy_solve)
print('Solved:', result)