Open thodson-usgs opened 1 year ago
Added mini-batch capabilities to the smoothly broken power law. It seems to work fine, but it is oddly not as consistent as the fit without the mini-batch. Feel free to give it a look and let me know what you think.
Using mini-batches with ADVI might be an efficient solution to local minima. Test it on the smoothly broken power law. If it works, add it to the demo in your existing branch. We might make it a general feature.
https://www.pymc.io/projects/examples/en/latest/variational_inference/bayesian_neural_network_advi.html#mini-batch-advi