You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently, SciMLBase doesn't make use of the batched observed API at all, e.g. sol[[x, y]] should call observed([x, y], ...). This makes observed generation a lot slower than necessary. Also, when generated observed is present, it should always be safe to put actual states, parameters, and independent variables, so there's no need to split them.
Another optimization opportunity that SciMLBase misses is the constant use of the point-wise API even for arrays. Since observed is inherently type unstable, it's much more efficient to create a function barrier and call it like
obs =observed(syms)
obs.(us, ps, ts)
, and we should avoid calling it like
observed.((syms,), us, ps, ts)
The text was updated successfully, but these errors were encountered:
Currently, SciMLBase doesn't make use of the batched observed API at all, e.g.
sol[[x, y]]
should callobserved([x, y], ...)
. This makes observed generation a lot slower than necessary. Also, when generatedobserved
is present, it should always be safe to put actual states, parameters, and independent variables, so there's no need to split them.Another optimization opportunity that SciMLBase misses is the constant use of the point-wise API even for arrays. Since
observed
is inherently type unstable, it's much more efficient to create a function barrier and call it like, and we should avoid calling it like
observed.((syms,), us, ps, ts)
The text was updated successfully, but these errors were encountered: