AMICI
AMICI copied to clipboard
Exploit structures for models with particular datasets
I just had the idea, that models with only 1 data point per simulation condition allows for many, many simplifications... (e.g., comouting the FIM from the gradient). Does it make sense to add some functionality in this direction into Amici? What do you think?
I just had the idea, that models with only 1 data point per simulation condition allows for many, many simplifications... (e.g., comouting the FIM from the gradient). Does it make sense to add some functionality in this direction into Amici? What do you think?
Always depends on how much these simplifications actually help you. If you have an application where this could be relevant, I would say go for it. Computing the FIM as sum of dyadic products of an adjoint gradient over experimental conditions (independent of whether only one observable is available or not) sounds like something pretty useful to me.
I see can be done with one observable only. But I doubt it to work for multiple ones, or at least I don't see it right away... But yes then, I'll do some of these things when I give the Newton solver the next makeover