This is a toolbox implementing the algorithm introduced in [1]. Documentation, examples, and user guide are found here.
A Python version using Diffrax and Jax is here. It's less tested and necessarily uses a slightly worse curve evolution algorithm due to more limited callback functionalities in the Python ODE solvers.
The package is a model analysis tool. It finds functional relationships between model parameters that best preserve model behaviour.
-
You provide a differentiable cost function that maps parameters to 'how bad the model behaviour is'. You also provide a locally optimal set of parameters θ*.
-
The package will generate curves in parameter space, emanating from θ*. Each point on the curve corresponds to a set of model parameters. These curves are 'minimally disruptive' with respect to the cost function (i.e. model behaviour).
-
These curves can be used to better understand interdependencies between model parameters, as detailed in the documentation.
[1] Raman, Dhruva V., James Anderson, and Antonis Papachristodoulou. "Delineating parameter unidentifiabilities in complex models." Physical Review E 95.3 (2017): 032314.