-
-
Notifications
You must be signed in to change notification settings - Fork 4
Description
Summary:
Not sure what the behavior should be for such an event, but need to figure it out.
Description:
The ambiguity arises when rate != 0 and ii = 0 (constant rate infusion), and amt != 0. First of, in the case of a constant rate infusion, amt will be ignored if addl = 0. Maybe we should produce a warning message if he or she passes a non-zero amt?
When addl != 0, the dosing regime that led to the steady state gets repeated at the end of the steady state. This leads to a prolonged steady state. For example, if amt = 1200, and rate = 150, a constant infusion begins when the steady state ends, and ends after 8 hours. As a result, the steady state is prolonged for 8 hours. The gradient with respect F (biovar) didn't get properly evaluated either (I saw discrepancies of up to 15% between finite and automatic differentiation, not for the steady state event, but the subsequent ones).
Reproducible Steps:
In generalOdeModel_test.cpp, run the second test, genCpt_One_SS_2, and set addl[0] = 10 (any non-zero value works). This should cause the autodiff unit tests to fail.
Current Version:
Stan v2.15.0
Torsten v0.83beta