This is the repo for the paper: CLR-DRNets: Curriculum learning with restarts to solve visual combinatorial games.
This dataset is stored in minimum.npy. And it has aroud 49, 000 Sudoku with unique hints. Based on that, you can create Sudoku with different number of hints.
For pure Sudoku, you can search "entropy_cell" and "entropy_alldiff" in "visual_cGAN_9by9.py".
More coming.