Closed rtimms closed 8 months ago
Hey I would like to work on this.
Can you please give examples of how to use step() function to get solutions which I can use for debugging? I'm quite new here and I cant find it in project doc.
Thanks 😄
Maybe if we change step function to something like:
def step(
self,
old_solution,
model,
dt=0,
non_linear_time=None,
npts=2,
inputs=None,
save=True,
)
Here non_linear_time
is an array of time (example: [0,10,15,17,20]). We can assign dt = non_linear_time[i] - non_linear_time[i-1]
inside a loop and make changes to the function accordingly. Is this a correct solution?
Here
non_linear_time
is an array of time (example: [0,10,15,17,20]).
Hi @prady0t, theoretically it looks promising, though it would be nicer to open a PR to keep things moving & also would be easier for us to suggest changes there 🙂
Sure will do!
Here
non_linear_time
is an array of time (example: [0,10,15,17,20]).Hi @prady0t, theoretically it looks promising, though it would be nicer to open a PR to keep things moving & also would be easier for us to suggest changes there 🙂
Made a PR. Please have a look.
When you call
solver.step
you pass indt
, the time time increment you want to step forward in. You can also passnpts
to get the solution back at a linearly spaced array of times between the initial time,t0
, andt0+dt
. It would be good to instead be able to pass an array of times,t_eval
, like you pass to solve, so you can get step solutions returned at non-linearly spaced times.The motivation for this is that right now you can't pass a
starting_solution
toSimualtion.solve
if you aren't using an experiment, so if you want to chain solutions together you have to callSimulation.step
instead.An alternative is that we allow starting solutions without experiments in
Simulation.solve
.The whole logic for
step
andsolve
is pretty clunky both inSimulation
andBaseSolver
, but that's a separate issue...