Open callistosp opened 1 year ago
[ +cc @kylebaron ]
No, I don't recall any discussion related to this. Thanks for starting one.
I don't know enough about what a NONMEM to mrgsolve conversion looks like to know what sort of info would be needed from the control stream. Could you give an example of a simple conversion?
In terms of getting the info parsed with nmrec, the main blocker would be whether nmrec supports within-record parsing for all the record types that are needed for the conversion. For example, if that conversion would require parsing of $PK
or $ERROR
records, that's a bit more ambitious than anything I had in mind with nmrec.
That's not to say those record types are off the table because I don't a great handle on everything that would involve, though I'm sure it would would look pretty different than the records types that are currently implemented. (Part of the idea with nmrec is that we can add support for different record types as needing rather than having to take on everything. Our current set is sufficient for the initial bbr and bbr.bayes features that prompted nmrec, although some high-level helpers are still needed to handle $THETA
, $SIGMA
, and $OMEGA
.)
(This is putting the cart very much before the horse, but, if nmrec did support all the required record types, my view is that this converter itself should probably live outside of nmrec.)
IMO nmrec
would definitely be useful to support this sort of translation, but, as @kyleam mentioned, you'd have to do some or lots of additional introspection that I'm thinking is way beyond the scope of nmrec
. But this is something we've had on our radar for a while and will continue to look at it.
My dream is to have a tool that automagically converts a NONMEM model to an mrgsolve model. With this parser and some of the plugins for mrgsolve that seems to be within the realm of possibility. Is there any discussion of including this on the roadmap for this package?