SkillAgentSearch skills...

DeepCube

Implementation of DeepCube paper that solves the Rubik's Cube using Deep Reinforcement Learning

Install / Use

/learn @azaharyan/DeepCube
About this skill

Quality Score

0/100

Supported Platforms

Universal

README

DeepCube

Tensorflow/Keras implementation of DeepCube paper by McAleer et al. that solves the Rubik's Cube using Deep Reinforcement Learning. We used a feed-forward network and a RNN network to compare the result. Both models are augmented with Monte Carlo Tree Search for solving the cube itself.

Prerequisites

Running

To run the solver all you must do is to run the main.py file. It will created scrambled cube and will try to solve it. The created cube will scramble number_of_turns(a parameter of the environment) times. The function keras.models.load_model has parameter a name of a neural network name. You can create a model instead of using saved one by uncommenting the commented part of the main.py file and removing the line for loading a model.

To train a neural network executing adi.py is enough. However one my want to tweak the hyperparameters there(iterations,l,k).

View on GitHub
GitHub Stars7
CategoryEducation
Updated13d ago
Forks2

Languages

Jupyter Notebook

Security Score

85/100

Audited on Mar 20, 2026

No findings