This code implements the block coordinate descent optimizer in [1]. newton.py implements the newton and gradient-based optimizer. mnist.ipynb is an example script applying Newton-GD to the MNIST classification benchmark.
Python >= 3.5
numpy
scipy
matplotlib
toolz
tensorflow >= 2.2
tensorflow-datasets
SAND No: SAND2021-2384 O
[1] R. G. Patel, N. A. Trask, M. A. Gulian, and E. C. Cyr. A block coordinate descent optimizer for classification problems exploiting convexity. arXiv preprint arXiv:2006.10123, 2020.