Deep Deducing on Sudoku

This repository contains codes illustrating how deep deducing solves Sudoku. To reproduce the results in the paper, simply run Deducing_xxx.py

Requirements

Packages requirements:

numpy

scipy

Learning phase content

To train sets of weight matricesin the paper, run this command:

Learning.py            

Deducing phase content

To use sets of trained weight matrices to start to solve each row, column and grid of the Sudoku table, run this command:

Deducing_xxx.py              

Already-trained sets of weight matrices

We have sets of weight matrices already-trained for your convenience. Please see the definition below to understand the content of each set of weight matrices.

self.6x6 _ original _ 1 _ 100x100x100 _ 30 _ 0.000001 _ 200m _ [1] _ weight_list

means a single set of weight matrices.

The definition for each bracket (from left to right) is listed below:

6x6
means this neural network is trained to solve each row, column and grid in 6x6 Sudoku.

original
means the payoff rule in the learning phase for this trained neural network is original. In other situation, for example, odd_even means the neural network must not only contain total different numbers in a row but also have the odd being separated from even to gain bonus.

1
means the sample batch in the learning phase for this trained neural network is 1 per each learning epoch.

100x100x100
means the trained neural network has three hidden layers, each with 100 neurons.

30
means the initial value for the set of slope multipliers to be updated in the learning phase.

0.000001
means the learning rate in the learning phase for this trained neural network is 0.000001.

200m
means the learning epochs in the learning phase for this trained neural network is 200 million or 2*10^8. The learning epochs are usually big in order to force the neural network to over-fit.

[1]
means the label of this trained neural network under the above training condition. For example, if it is [3], then it means this neural network is the third neural network under the training condition original _ 1 _ 100x100x100 _ 30 _ 0.000001 _ 200m.

weight_list
means the set of weight matrices of this trained neural network.

Results

For results, please refer to the supplementary material in the paper.