Options

Any recommendations for non-linear optimizer that can handle discontinuous derivatives?

Hi, I'm just checking to see if anyone here has an recommendations for software (either a library or possably an actual system) that can handle doing nonlinear optimization that can use derivative information but can handle when they're discontinuous without getting horribly lost. (I'm looking at eventually doing sparse regression where the regularization weight function is $|v|$ or $\sqrt {|v|}$, both of whose derivatives change as $v$ crosses $0$ and have undefined "bidirectional derivatives" at $0$. I'm not actually planning on starting with sparse regression, but as it's something I'd like to try it makes sense to try and start using an optimizer that can eventually handle that.)

If no-one happens to know this I'll look around and record any findings I make here on the forum.

Comments

  • 1.

    Differential Evolution

    This MIGHT be what you need, might not. Gradient vector (list of derivatives) used to give a hint at the CLIMBING direction, derivatives might not even exist at all points. Many variations could be coded.

    However this is not for faint at heart and it requires large machines.

    Dara

    Comment Source:[Differential Evolution](http://www1.icsi.berkeley.edu/~storn/code.html) This MIGHT be what you need, might not. Gradient vector (list of derivatives) used to give a hint at the CLIMBING direction, derivatives might not even exist at all points. Many variations could be coded. However this is not for faint at heart and it requires large machines. Dara
Sign In or Register to comment.