Can I extract mathematical formulation?

[xkhainguyen]

Hi,

I wanted to use CUTEst as a standardized benchmark for my machine learning-based approaches using Python. However, calling Fortran routines sequentially on CPUs is inefficient.

To address this, I believe I need to know the underlying mathematical formulation (symbolic expressions of the objective and constraints) for each problem so that I can implement them natively and perform batched evaluations and gradient backpropagation on GPUs.

Do you know of any ways to do this? Also feel free to suggest any alternatives.

Many thanks!

[nimgould]

I am very sorry but that is not the purpose of CUTEst, the problems are formulated to isolate (partially seprarable) structure, and therefore not written as explicit expressions. You might look at the AMPL version of CUTE which may be more what you need.

We would be interested to find out where the inefficiency is, is it in the Python interface layers to fortran via C? We haven't heard of other instances, so this perhaps a concern. I will flag our Python expert @jfowkes who might be able to shed more light when he returns next week

[jfowkes]

See also jfowkes/pycutest#107