Forecasting with Spectral Mixture kernels

This is a companion repo for my blog post which takes a look at the Spectral Mixture kernel as presented in Gaussian Process Kernels for Pattern Discovery and Extrapolation.

How to run

Make sure you run in a virtual environment, then pip install -r requirements.txt and you should be good to go. To fit the models and produce plots run the scripts in the scripts folder and provide the command line arguments you like. For instance python scripts/fit_mauna_loa.py --num-restarts 3 --lr 0.1 --num-iters 500 --kernel smk will fit a model using only the SM kernel using three random restarts and 500 iterations per restart.