On Mean-Field inference and machine learning: the example of Sudoku.
Work done during the Summer 2017 internship at the CVLab - EPFL.
- create.py: create a sudoku dataset, you can specify the size of the board, the number of examples and the difficulty of the dataset.
- discr.py: this script tests the discriminative power of a given CRF (not updated to version 2.0)
- experience.py: launches a bunch of experiments (not updated)
- judge.py: evaluates the quality of a learned CRF against a dataset, specifying the number of modes (in case of multi-modal mean-field).
- learning.py: some MF samples (not updated)
- mf.py: the core MF/MMMF inference library
- sudoku_learn.py: uses the MMMF inference to learn the game of sudoku using a multi-modal gradient method.
- sudoku.py: some utility library in order to handle sudoku dataset.