Make plot and predict consistent

JuliaTrustworthyAI / LaplaceRedux.jl

Effortless Bayesian Deep Learning through Laplace Approximation for Flux.jl neural networks.

https://www.taija.org/LaplaceRedux.jl/

MIT License

39 stars 3 forks source link

Closed pat-alt closed 1 month ago

pat-alt commented 1 month ago

Should we return the observational noise when predicting from regression?

Currently, predict returns glm predictions for regression, not incorporating the observational noise (which I think I chose to do to stay consistent with PyTorch)
In TaijaPlotting.jl, we plot the interval that contains both sources of noise (parameters and observational).

In any case, we should add warnings and/or documentation

cc @pasq-cat

pasq-cat commented 1 month ago

directly from alex immer github docs.

pat-alt commented 1 month ago

directly from alex immer github docs.

That's exactly the formula we use here for plotting. Let's discuss during our meeting.

pat-alt commented 1 month ago

addressed in #118