On the github readme for sunode ( GitHub - pymc-devs/sunode: Solve ODEs fast, with support for PyMC) they have an ‘extra’ parameter that is set to np.zeros(1)
. (This is in the call to sunode.wrappers.as_pytensor.solve_ivp
)
This doesn’t seem to be optional, and if you do not include it the sampling fails ValueError: When changing to a smaller dtype...
I don’t understand why this is needed? Is this working as intended?