We should implement an interface, where there is an option to provide a general function to get initial guesses for the truth solvers. For example, one could get initial guesses from another reduced basis or just take a random state. (And of course different, yet unknown truth solvers will probably need such an option.)
So instead of init_from_rb, Greedy could contain a initial_guess::Function field.
This could also help with getting a more general interpolate for MPS initial guesses, where the bond dimension for MPS addition can be passed as a keyword argument instead of the dm.sweeps.maxdim solution we use at the moment. Also we could probably remove the _ in the vector-type interpolate.
We should implement an interface, where there is an option to provide a general function to get initial guesses for the truth solvers. For example, one could get initial guesses from another reduced basis or just take a random state. (And of course different, yet unknown truth solvers will probably need such an option.)
So instead of
init_from_rb
,Greedy
could contain ainitial_guess::Function
field.This could also help with getting a more general
interpolate
for MPS initial guesses, where the bond dimension for MPS addition can be passed as a keyword argument instead of thedm.sweeps.maxdim
solution we use at the moment. Also we could probably remove the_
in the vector-typeinterpolate
.