search

AdamCobb / hamiltorch

PyTorch-based library for Riemannian Manifold Hamiltonian Monte Carlo (RMHMC) and inference in Bayesian neural networks
BSD 2-Clause "Simplified" License
426 stars 63 forks source link

Slightly more optimized hyperparams for CNN on MNIST #7

Open MECLabTUDA opened 4 years ago

MECLabTUDA commented 4 years ago

This combination led to expected accuracy above 0.8 across multiple runs. I tried it out in Colab.

tau = 1/10 step size = 0.02 num_samples = 5000 burn = 500

AdamCobb commented 4 years ago

Thanks very much! I am in the process of making a few updates to hamiltorch and will try and add these hyperparameters in on the next push!

All the best!

Adam