README.md

ARC with Neural Network

Abstraction and Reasoning Challenge

• Kaggle Competition - ARC Prize 2024

• arcprize.org

Output

PixelEachSubstitutor

Accuracy First Search (with recurrsion of PixelEachSubstitutor )

Main Libraries

• Pytorch (Machine Learning Framework)

• Pytorch Lightning (Lightweight PyTorch Wrapper)

• Hydra (Configuration Manager)

• Rich (Formatting)

---

1. Setup

conda env create -f environment.yml -n myenv
conda activate myenv

2. Download ARC Data

To use Kaggle API, place kaggle.json in the location appropriate for your operating system

mkdir data
cd data
kaggle competitions download -c arc-prize-2024
unzip arc-prize-2024.zip -d arc-prize-2024

3. Run

src/data.py 

• Define Dataset, Dataloader, Datamodule (Pytorch Lighting)

• Run: Visualize all data

src/classify.py

• Filter data to train or test

• Run: Visualize filtered data

src/arc/model/*

• Define Model

config/*

• Set Hyperparameters for train, test

src/train.py

• Run: Train model

src/test.py

• Run: Test model trained