Non Linear expressions

[ycouble]

Hi,

First of all, thanks for the interface, coming from gurobipy, the transition was seamless.

However, I am looking to use non-linear expression, namely a log in the objective function. Looking inside the code and documentation, I couldn't find the correct way to do it.

Is it implemented yet?

Best,
Yoann

[fserra]

Hey Yoann

I am sorry that currently the only nonlinearities that we can handle are
polynomials.

We want to extend it of course, but I think we first have to decide on the
data structure for general expressions.

On Oct 11, 2016 10:03, "ycouble" notifications@github.com wrote:

Hi,

First of all, thanks for the interface, coming from gurobipy, the
transition was seamless.

However, I am looking to use non-linear expression, namely a log in the
objective function. Looking inside the code and documentation, I couldn't
find the correct way to do it.

Is it implemented yet?

Best,
Yoann

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#41, or mute the
thread
https://github.com/notifications/unsubscribe-auth/AFNG9v_beY_nDp9gSCfnrlX2i44lt9Nnks5qy0LMgaJpZM4KTWLA
.

[ycouble]

Ok, thanks for the update.

[rschwarz]

Besides the internal datastructure, it's also not clear what the usage will look like.
For polynomial expressions, we can use operator overloading on * and + to hijack the expression parsing in Python. But I wouldn't know how catch a call to log(x) where x is a Expr (there's no log function in the global namespace in Python, only math.log). I guess we would have to provide functions (and types) that user would need to import, as in

from pyscipopt import log
cons = (log(x) == 3)

[ycouble]

For the objective function, would it be possible to simply pass a user-defined function as argument of the setObjective call ? Something like:

def eval_f(X):
    return sum(math.log(x) for x in X)

and then in the model definition

model.setObjective(eval_f, X, 'maximize')

[rschwarz]

I don't think we can support general functions. For your example, I could imagine the following:

model.setObjective(sum(pyscipopt.log(x) for x in X))

[tarun018]

Hello !

Is there any update on this ? I need to use a log objective as part of my model.

Thanks.

[fserra]

Hi @tarun018

I am sorry to say that here has been 0 activity in this issue.
I hope I have some time to start working on this, but not before the next couple of weeks

[fserra]

general nonlinear functions are implemented now!