Skip to content

add bots, change log likelihood, add tests, remove some dependencies

Taylor Curley requested to merge github/fork/itsdfish/master into master

Created by: itsdfish

This looks really interesting. I would like to learn more at some point. I made some suggested changes that might help with maintaining the package and workflow. In addition, I suggested a few changes to the log likelihood function in your example. Here is a summary:

  • no need to specify types with @concrete, but you can opt in

  • pass sim_params to the log likelihood function so you can change them without changing the function

  • i think the likelihood function is a Multinomial

  • I don't think the fit normal was contributing to your log likelihood, you might consider sampling μ and σ through DEMCMC

  • added SafeTestsets with a simple test

  • added CI for unit tests

  • added compat helper and tagbot

  • I'm not sure if 40 trials will give you stable results

  • if your predictions are changing each trial, you might want to compute the log likelihood of each trial to fully leverage the data to estimate decay. For example, you might want to return an n_sim X n_trial matrix and compute your Multinomial log likelihood across n_sim where n_sim is somewhat large

  • I added compatibility constraints to Project.toml

  • I removed Turing DEMCMC, StatsPlots and Plots because the convention is to reduce unnecessary dependencies. In this case it is important because Turing and Plots are large dependencies. Most developers work in a "sandbox" package where you can add Turing and Plots to a temporary environment

Merge request reports