Why running SGD with multiple hosts return segfault? about galois HOT 6 CLOSED

Hello,
We have been able to reproduce this error and are working on the fix. We will push the changes as soon as possible and let you know.

from galois.

I've also encountered a similar issue in matrix completion. The program went segmentation fault on both gr and sgr format input even with one thread.
Any idea why I can't run SGD on one thread?

from galois.

Hello.

@WyjAloneSmile May you give me the command you used to convert your edgelist to the gr format as well as the command line you used to run sgd?

@Yang-YiFan Please open a separate issue with more details such as the command line used, graph used, etc, and I can take a look at your problem.

Thank you.

from galois.

Hello.

@WyjAloneSmile May you give me the command you used to convert your edgelist to the gr format as well as the command line you used to run sgd?

@Yang-YiFan Please open a separate issue with more details such as the command line used, graph used, etc, and I can take a look at your problem.

Thank you.

convert edgelist to gr: ./graph-convert input_graph output_graph -edgelist2gr -edgeType=int32
run SGD: yhrun -N 2 -n 2 -c 1 ./sgd input_gr -runs=1 -t=1 -DECAY_RATE=0.5 -LAMBDA=0.001 -LEARNING_RATE=0.001 -maxIterations=1000

from galois.

Hello @WyjAloneSmile.

Please try using "-edgeType=float64" instead of "-edgeType=int32". The distributed SGD graph expects a double for its datatype.

I am not able to reproduce the exact error you are seeing, but using int32 does cause the program to hang on my end. Using float64 causes it to run to completion correctly.

Let me know how it goes.

Thanks,
Loc

from galois.

I'll be closing this issue unless there are still problems. In that case, please reopen it.

from galois.