Overview

A convolutional neural network written in python using tensorflow, that is designed to solve sudoku puzzles. Note: Sudoku solution needs a renforced learning approach, so that the existing numbers are not modified.

Training Datsource :

1 million Sudoku games by Bryan Park A tarball for the above data is included and should work just fine. Otherwise download the above data source and place the csv file in the data/ subdirectory of current repo. Downloading the data source requies sign in with a kaggle account.

Dependencies :

• numpy
• scikit-learn
• tensorflow

Usage :

from a docker container or local installation run, python sudoku_nn.py