---
title: Sudoku Deeplearning 100% accurate Sudoku solving with deep learning algorithm
emoji: 🖊️🎲🎯
colorFrom: blue
colorTo: purple
sdk: streamlit
sdk_version: "1.41.0"
app_file: app.py
pinned: true
---

# Sudoku Deeplearning 100% accurate Sudoku solving with deep learning algorithm

## Setup

- Install python dependencies: `pip install -r requirments.txt` (venv might me be the best option)
- Download `sudoku_reshaped_3_million.npz` to the root from https://www.kaggle.com/datasets/radcliffe/3-million-sudoku-puzzles-with-ratings
- Execute `3million_sudoku_reshape.ipynb` (it reshape the dataset into nice numpy tensors)
- Execute `experiments/boost_training_8_m3_train_only_once_clean.ipynb` (it process the boost layers model training)
- Execute `experiments/boost_training_9_m3_train_based_on_8_train_trial_error.ipynb` (it trains the trial error model)
- Execute `experiments/boost_training_9_m3_test_3m_sample.ipynb` (it adjusts the thresholds on the big validation dataset)

The final model weights is store in the `model_9_m3_big_validation_th.ckpt` file.

## Paper
currently unpublished, please read `100% accurate Sudoku solving with deep learning algorithm.pdf`