Skip to content

JakeColtman/SurPyval

Repository files navigation

WIP - API subject to sweeping changes

SurPyval

SurPyval is a Bayesian survival analysis library

Philosophy:

* Models should be transparent about their assumptions and workings * Models should allow tweaks and modifications 

Implementing this philosophy has a number of positive effects on the library:

* The log-likihood and plate diagrams of models are exposed * Models are created through composition of simple units * SurPyval objects thinly wrap and expose well-know libraries (esp. scipy) * There are no hand-offs to non-python objects * Models allow for substitution of any of their composite blocks 

The trade-off to get these goods is performance. Models provide in the library are designed to be tweakable, which limits performance optimizations. This manifests itself in a number of ways:

* Straight up crunching speed * Memory useage * Models often don't exploit conjugacy where it exists 

For very large data sets or very complicated models, you might be better off using something like Stan.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages