-
Notifications
You must be signed in to change notification settings - Fork 27
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Segmentation fault while solving basis pursuit problem with convex.jl #24
Comments
For reference, I'm running with Julia 0.3.7 with Convex.jl 0.0.4 and SCS.jl 0.0.3 (which is based on SCS 1.0.7). I'm on 64 bit Ubuntu 12.04. I did a bit of debugging with gdb. At the time of the crash,
It looks like Debugging further is beyond my interest (and time availability) at the moment, but I hope this is useful in fixing the bug. |
It'd be good to file this on the SCS repository too - I'm not sure the maintainer of SCS looks/is subscribed to the issues for the Julia wrapper |
I'd also recommend trying a different LP solver like Clp. |
I am solving my problem using ECOS. Interestingly, I had a similar bug with ECOS that happened twice but then went away (even with a fixed random seed at the beginning of the program, it failed once and then didn't fail again). |
Does this still happen with the latest version of SCS.jl? |
In fact, Convex.jl can call an LP solver when the 1-norm is used; we can clarify that in the documentation. You can take a look at the norm atom to see exactly what we do. |
I'm unsure if the error is fixed (since I need to run the code for a long time to find it). After a decent number of trials, I haven't seen a crash yet. However, I do get relatively frequent WARNING: Problem status UnknownError; solution may be inaccurate. error messages. I'm doing some more testing now and will report back if I get a segfault or any other kind of failure. |
This is still happening (julia v0.4, SCS 1.1.5, latest SCS.jl). I ran it for a long time to get:
Does anybody know what exactly the I think this is caused by a memory leak. While running this script the memory used by Julia creeps up, in the screenshot below it's taking up a whopping 2.4Gb of memory, just solving these tiny problems. My best guess is that the leak is actually in Convex.jl (jump-dev/Convex.jl#83), since the exact same thing happens using ECOS (at least the memory usage increase), and I can't reproduce the problem by calling SCS directly. |
You are trying to solve the problem with equality constraints with arbitrary precision try to reformulate the part of the problem by minimizing the same objective function with a precision epsilon |
This was possibly caused by #90. |
I have the following code for a basis pursuit simulation:
Running this works for a while (several thousand calls to
solve!
) but eventually seg-faults:I'm not sure if the bug lies in Convex.jl, SCS.jl, or SCS. I'm debugging now with gdb, will update issue as I learn more.
The text was updated successfully, but these errors were encountered: