Troubleshooting

Compiled here are some common troubleshooting approaches we would recommend if you are encountering difficulties with different functionalities of CKBIT.

Model Loading Troubleshooting

CKBIT is designed to store compiled models so they do not need to be recompiled each time they are run. While this saves significant time, sometimes the model storage can encounter errors if other operations on the computer conflict with it. For those cases, we recommend opening the folder where the script is being run and deleting the stored model. This will cause the model to be compiled again and should solve the problem.

MAP Troubleshooting

Due to sensitivity of the initial starting conditions and complex models being optimized, the MAP functionality can encounter challenges. We recommend specifying the input variable seed with an integer and running the function. If this run fails, specify seed with a new integer. In our experience, the MAP functionality usually works within 5 different seed specifications.

MCMC Troubleshooting

MCMC runs have convergence checks detailed in the Appendix of the publication. When these convergence checks are showing that the sampled values from the distribution are not well converged, we recommend increasing the control criteria like so: control={‘adapt_delta’:0.99999999, ‘max_treedepth’:100}. Please reference Figure 8 of the publication to see how to properly implement this control variable. In our experience, this will solve most issues experienced with these chemical models.

VI Troubleshooting

VI runs also have convergence checks detailed in the Appendix of the publication. When these convergence checks are showing that the distributions did not converge well, please try specifying more informative priors and running the model again. We have found that for some datasets, the VI estimations will not converge without more informed prior distributions being implemented.