Karl Ratzsch - 2017-09-30

That is unfortunately a rather tricky feature, and most solutions i have seen in other software do not work very reliably.

The problem is to determine what should happen when a parameter reaches that border. Set it to that value and/or take it out of the fit? Make it "expensive" to get ever closer to the min/max value? How expensive? A good choice will surely be problem-dependent, and a bad one could lead to wrong ar at least different results.

Luckily, this however (often) becomes a non-problem if you determine your starting parameters more accurately (e.g. by fitting repeatedly with reduced parameter set and/or data range). If not, it usually means your mathematical model is bad, or your data is too few and too noisy.

 

Last edit: Karl Ratzsch 2017-10-10