Sudoku is a popular number puzzle that requires you to fill blanks in a 9X9 grid with digits so that each column, each row, and each of the nine 3×3 subgrids contains all of the digits from 1 to 9. There have been various approaches to solving that, including computational ones. In this project, we show that simple convolutional neural networks have the potential to crack Sudoku without any rule-based postprocessing.

• NumPy >= 1.11.1
• TensorFlow == 1.1

• To see what Sudoku is, check the wikipedia
• To investigate this task comprehensively, read through McGuire et al. 2013.

• 1M games were generated using generate_sudoku.py for training. I've uploaded them on the Kaggle dataset storage. They are available here.
• 30 authentic games were collected from http://1sudoku.com.

• 10 blocks of convolution layers of kernel size 3.

• generate_sudoku.py create sudoku games. You don't have to run this. Instead, download pre-generated games.
• hyperparams.py includes all adjustable hyper parameters.
• data_load.py loads data and put them in queues so multiple mini-bach data are generated in parallel.
• modules.py contains some wrapper functions.
• train.py is for training.
• test.py is for test.

• STEP 1. Download and extract training data.
• STEP 2. Run python train.py. Or download the pretrained file.

• Run python test.py.

Accuracy is defined as: Number of blanks where the prediction matched the solution / Number of blanks.

// Improve CNN

• epoch nums
• filter size
• regularization
• dropout
• more tainining data (medium, hard, expert, evil)

// Search Problem

• Use prob distributions in heuristic search

// RNN (Bidirectional with LSTM cells)

• list of tuples (blank_cell, num_hints) where num_hints = sum of hints in row,col,square
• sort list by num_hints
• feed in row, col, sqaure of cell with highest num_hints to RNN
• take joint prob for cell as solution
• update board with prediction
• recompute list with newly predicted cell
• choose new largest num_hints cell and continue
• (build structure)

// Deep RL

• very difficult
• we have structure for it though

Visualization expected board --> our board

// Input should now be probability distribution

• lets make it all one-hot
• back to 9 labels
• 0 has equal probability for all labels

bi - directional lstm go across the row, column, square as individual inputs to the lstm each pos in row x batch x

only do softmax at end

feed forward layer to translate the outputs of row, col, sqaure (features) to softmax probability feed forward to softmax layer

becomes probabilities after loss combine 3 outputs then feed forward then loss