Closed TorkelE closed 8 months ago
Two things:
I think the reason for the bad performance in the parameter estimation is the bounds, by default they are [1e-3, 1e3]
which you true parameter values exceed (typically for most sysbio models we stay withing these bounds hence they are the default), bounds can be set in PEtabParameter
Your second problem is because by default a PEtabParameter
is assumed to be on the log10
-scale. The following should work:
petab_problem.compute_cost(log10.(last.(p_true)))
Hopefully these two things fix the problem!
You are correct, on both. Thanks a lot! :D
I have a relatively simple case of parameter fitting where I try to use PEtab, but unfortunately, it fails. It might be that I have not used a good noise formula or similar. However, there are some warnings when I try to evaluate the cost function for the true parameter set which makes me suspicious.
Code
The model is from a catalysis chain network used for e.g. chemical synthesis. Parameter values and initial conditions are at least somewhat realistic.
Generate data.
Fit model
Weird stuff
Here, this code
generates these warnings:
Which seems weird. Surely the original parameter set should at least be simulatable? It seems PEtab is trying to determine an initial step size for this parameter set, but fails, resulting in a failed simulation, and infinite cost.