Thanks for your great package! It's a marvellous piece of engineering. I've tried using ~8 python quadratic solvers (through the common qpsolvers interface), and I find ecos the best of all of them in terms of precision. In my setting, I value the correctness of the results and favour reporting a failure rather than returning inaccurate results.
I can caused it to crash on my datasets when increasing the sparseness of the random patterns on which it is trained. This occurs non-deterministically when using patterns with f<1/16, i.e., only this fraction of the inputs is non-zero. Repeating the calculation several times, I consistently get a non-deterministic crash with a message asking me to email you.
The a minimal code reproducing the problem is attached.
standalone_ecos_problem_py.txt
Thanks for your great package! It's a marvellous piece of engineering. I've tried using ~8 python quadratic solvers (through the common qpsolvers interface), and I find ecos the best of all of them in terms of precision. In my setting, I value the correctness of the results and favour reporting a failure rather than returning inaccurate results.
I can caused it to crash on my datasets when increasing the sparseness of the random patterns on which it is trained. This occurs non-deterministically when using patterns with f<1/16, i.e., only this fraction of the inputs is non-zero. Repeating the calculation several times, I consistently get a non-deterministic crash with a message asking me to email you.
The a minimal code reproducing the problem is attached.
standalone_ecos_problem_py.txt