What does the "maxiters" solver option do?

[Volker]

I also get the warning

┌ Warning: Interrupted. Larger maxiters is needed.

when I use RK4 solver with a fixed step size of 0.01 s.

The code which I used is the nearly the same as in the post

Fitting a dynamic system with an exogenous input (nonhomogenous neural ode) via DiffEqFlux

I wouldn’t use ReverseDiff or inplace here. Another mistake that you made was that, notice that the solution is a row vector for each output component, so you were broadcasting a row vector against a column vector in the loss function to generate a matrix. That’s not what you wanted. Here’s a code that trains: using DifferentialEquations, Flux, Optim, DiffEqFlux, DiffEqSensitivity, Plots tspan = (0.1f0, Float32(10.0)) tsteps = range(tspan[1], tspan[2], length = 100) t_vec = collect(tsteps) ex…

This also the code example for exogenous inputs on the documenation page of DiffEqFlux.

The just changed the solver to RK4, a slightly different excitation and updated the code to GalacticOptim.

I’m a little bit confused, because I thought, that maxiters would define the maximum iterations of the solver. Therefore, I would expect, that I would need 100 iteration when I simulated from 0.0 s to 1.0 s. Or is it the overall maximum iteration? In other words, for all iterations of optimization.