The PyStan version this package used before (2.19.1.1) is almost three years old now. Other projects require a more recent version (typically > 3.0.0) which leads to incompatibilities if baycomp is meant to be used in those projects. Aside from that, PyStan 3.0.0 was a complete rewrite with a much cleaner user experience (e.g. model caching by default). A summary of changes from PyStan 2.19.1.1 to 3.0.0 can be found here.
Summary of changes/implications for baycomp:
Fix deprecation warnings in Stan model code:
<- was replaced by =
increment_log_prob was replaced by target +=
Use the new PyStan interface in place of the old one.
Pickling Stan models manually is not necessary any more: Since version 3, PyStan only interacts with httpstan which already performs model caching.
Windows support might become worse than before (PyStan >= 3.0.0 does not officially support Windows any more whereas PyStan < 3.0.0 had “partial support” according to the PyStan docs—whatever that may mean).
Note that I deliberately did not perform any other changes (e.g. regarding code formatting etc.) so that the changeset can be checked straightforwardly.
Reasons for this change:
The PyStan version this package used before (2.19.1.1) is almost three years old now. Other projects require a more recent version (typically > 3.0.0) which leads to incompatibilities if baycomp is meant to be used in those projects. Aside from that, PyStan 3.0.0 was a complete rewrite with a much cleaner user experience (e.g. model caching by default). A summary of changes from PyStan 2.19.1.1 to 3.0.0 can be found here.
Summary of changes/implications for baycomp:
<-
was replaced by=
increment_log_prob
was replaced bytarget +=
Note that I deliberately did not perform any other changes (e.g. regarding code formatting etc.) so that the changeset can be checked straightforwardly.