search

metrumresearchgroup / Torsten

library of C++ functions that support applications of Stan in Pharmacometrics
BSD 3-Clause "New" or "Revised" License
52 stars 11 forks source link

example-models: format and inconsistency with User Manual #4

Closed billgillespie closed 3 years ago

billgillespie commented 5 years ago

Currently the examples in Torsten/example-models vary in format and the examples intended to match with the User Manual are not fully consistent with the manual text.

I propose that the examples be more self-contained within their respective folders rather than spread between the example folder and the R folder (with the exception of shared function scripts like stanTools.R). The User Manual text and the corresponding examples should also be realigned to match.

yizhang-yiz commented 5 years ago

The examples in the latest user guide at https://github.com/metrumresearchgroup/Torsten are actually generated by include, so they should be identical to the corresponding examples in the example-models directory in the same repo. If we update the upstream repo https://github.com/metrumresearchgroup/example-models, then we must update the same folder in the container repo.

charlesm93 commented 5 years ago

@billgillespie I like the idea of having self-contained scripts. I'm however concerned about having duplicate code, since the R script can vary little (if not at all) between examples.

I think it would make sense to have a two compartment model folder, with one R script and three Stan implementations (to illustrate use of analytical, matrix exp, and numerical methods). Similarly the FK folder could have one R script with two models (full integration and mixed solving).

I'll add to this issue what I mentioned in issue #6, namely that the examples can be written more concisely using new language features.

charlesm93 commented 5 years ago

A longer term goal: it would be nice to have a case study on fitting the population FK model. This includes a discussion on the mixed solver, diagnosing divergent transitions and centered / non-centered parameterizations, scaling the parameters for a full covariance analysis, and using within-chain parallelization. I think such a model could be fitted in ~ 2 hours if properly implemented. But yeah, there are a lot of computational techniques to deploy to get there.

yizhang-yiz commented 5 years ago

It’s worth to have duplication in this case. If you look at CVODES examples they’re all duplications. Here being self-contained takes priority IMO.

Yi

On Apr 1, 2019, at 3:18 PM, Charles Margossian notifications@github.com wrote:

@billgillespie https://github.com/billgillespie I like the idea of having self-contained scripts. I'm however concerned about having duplicate code, since the R script can vary little (if not at all) between examples.

I think it would make sense to have a two compartment model folder, with one R script and three Stan implementations (to illustrate use of analytical, matrix exp, and numerical methods). Similarly the FK folder could have one R script with two models (full integration and mixed solving).

I'll add to this issue what I mentioned in issue #6 https://github.com/metrumresearchgroup/Torsten/issues/6, namely that the examples can be written more concisely using new language features.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/metrumresearchgroup/Torsten/issues/4#issuecomment-478768001, or mute the thread https://github.com/notifications/unsubscribe-auth/AfkhsNx-H80FpYXK5iRBFPhmrkAuhQGTks5vcoXBgaJpZM4Y9F8F.

yizhang-yiz commented 3 years ago

All examples are now reside in corresponding directory. Closing.