Open junwei88 opened 4 years ago
Were you able to figure out why? I'm having similar difficulty solving linear problems when the number of parameters gets to around 10,000. I can solve them with a manual gradient descent method, but CVXPY (specifically ECOS) breaks down. Any tips?
The ECOS solver failed to converge on a fairly simple problem. As far as I can tell, the problem is well-conditioned and well-scaled, with coefficients, variables and objective function value between -10 and 10. The issue may be reproduced below:
I understand that the -1 <= w <= 1 constraints are redundant in this case, but these box constraints are part of my more general problem. In other cases that I have tested with some of these binding, I have also occasionally encountered similar failures.
I am using CVXPY 1.1.1 with ECOS solver 2.0.7 and Numpy 1.16.5, on Python 3.7.4.
Shown below is the verbose output from the first and second problems: