Handle multi-scale models
Multi scale models tipically involves reactions that differ in their speed by several orders of magnitude. This raises two issues:
- Species with low variations never get picked by
get_skeletons
- The global optimization will put the focus on parameter fits that explain the biggest variations, which is expected
An idea would be:
- To pick the species whose variations should be explained at time t using
s^* = \underset{s\in S}{\argmax}~ V_s(t)/V_s_max
, therefore at time t we pick the species whose variations are the highest w.r.t. its maximum variation level throughout the trace
- Perform the global optimization with ((y-y_pred) / y_max)^2 instead of (y-ypred)^2, which will put the same importance for all species.