Inequality transitivity constraints

alexblaessle / constrNMPy

A Python package for constrained Nelder-Mead optimization.

GNU General Public License v3.0

11 stars 2 forks source link

Inequality transitivity constraints #5

Closed Goschjann closed 5 years ago

Goschjann commented 5 years ago

Hi,

I am tuning thresholds for a ML Algorithm to backmap regression scores to ordinal classes.

Constrained optimization is necessary as I want to enforce an ordinal structure amongst them (cut1 < cut2 ... < cut8)

Thus, I would like to use your constrained NM implementation, but:

as far as I get it, I can only set fixed upper/ lower bounds for the single thresholds (variables)
unftly, they depend on each other

Did I miss sth here?

Thx a lot in advance guys!

alexblaessle commented 5 years ago

Hi,

glad you found my toolbox. About your questions:

You can set for each variable a tuple of lower/upper bounds.
My toolbox cannot do bounds that depend on each other yet. Can you transform your problem such that you have fixed boundaries and then apply the NM algorithm?

Goschjann commented 5 years ago

Hi Alex,

thanks for your good work and your answer! Unfortunately I cannot transform the problem that way but I figured out a (quite hacky but still working) solution with scikits fmin() function!

Cheers!

alexblaessle commented 5 years ago

Issue closed since this is more a request for a feature that is somewhat unfeasible (at least for now).