Geometric Deep Learning for Rubik's Cube Group

A repository for the project on using symmetry-aware deep learning architectures to solve the Rubik's cube.

To reproduce the results, download data from Google Drive and save it to a data folder. The data for the sampled dataset collected by Herbert Kociemba are available at http://kociemba.org/math/optman/100000optcubes.zip

The generated plots can be found at Google Drive.

Installation

Create a Python environment, e.g., with venv, and activate it:

python -m venv venv
source venv/bin/activate

Install required libraries:

pip install -r requirements.txt

The project is structured the following way.

cube data structures are implemented in folder classes/
data generation and symmetry-equivalence compression is implemented in folder generate/, Python script utils/compressions.py, and Jupyter notebooks generate.ipynb and weisfeiler-lehman_compressions.ipynb
PyTorch models and training runners are defined in pytorch_classes/
all scripts run on the cluster are in folder scripts/
analysis of the training and search results can be found in analyzeResults.ipynb, summarize_accuracies.ipynb, and summarize_search.ipynb Jupyter notebooks
explainability experiments can be found in Jupyter notebook explainer.ipynb

Name		Name	Last commit message	Last commit date
Latest commit History 60 Commits
ablated_ntbs		ablated_ntbs
classes		classes
explain		explain
generate		generate
optuna_pytorch_classes		optuna_pytorch_classes
pytorch_classes		pytorch_classes
scripts		scripts
search		search
utils		utils
.gitignore		.gitignore
README.md		README.md
analyzeResults.ipynb		analyzeResults.ipynb
explainer.ipynb		explainer.ipynb
gen_equal_param_calculation.txt		gen_equal_param_calculation.txt
generate.ipynb		generate.ipynb
kociemba_training.ipynb		kociemba_training.ipynb
one_color_patterns_compression.ipynb		one_color_patterns_compression.ipynb
optuna_training.ipynb		optuna_training.ipynb
requirements.txt		requirements.txt
summarize_accuracies.ipynb		summarize_accuracies.ipynb
summarize_searches.ipynb		summarize_searches.ipynb
training.ipynb		training.ipynb
weisfeiler-lehman_compressions.ipynb		weisfeiler-lehman_compressions.ipynb