jlindbloom / fmax Goto Github PK

@Jsevillamol I am going to do an overhaul of this library this weekend. My goal is to convert to PyMC v4 and have a docs page up and running with some examples.

My biggest question is, what do you think about changing our handling of the attempt distribution to allow for a general choice by the user? Currently, we limit to just a handful of distributions, but in principle I don't see why we can't construct the Max/Min distributions as a function of an input attempt distribution.

I would also like to figure out a way to expose the PyMC model so that the user can specify the prior directly, without needing to be constrained by our interface of passing in a dictionary of parameters of some pre-defined prior that we have chosen.

Any thoughts?

Right now it is unclear for new users how to employ the package.
Adding an example will make the library much more usable.

For instance, see how I did it in a previous library.

E.g., forecast-max/min. Note that Github redirects old links when repository names are changed

In the models.py file there is a legacy class for Weibull forecasting. This should be deleted.

Example NBs are the primary entry point for new users. So more text as to what the example shows would be great, for example here: https://github.com/jlindbloom/fmax/blob/main/notebooks/bayes_tryfos_predictions.ipynb