6 Replies Latest reply on Nov 14, 2017 5:07 AM by Kprasad_Intel

    Keras with Theano backend on DevCloud


      Hi, I really appreciate the access to DevCloud

      According to

      Intel® Nervana™ DevCloud - Intel Nervana

      Keras should have been installed already, however, I don't see it installed on the cluster.

      I tried to pip install --user to my user directory and it works to certain degrees. When I used it with tensor flow, I got "nan" errors, which I never see it on my local machines. When I used it with theano, I got an error with api version mismatch between my user Keras and global Theano.


      I would like to try Theano first, can you help me with it?


      Another question is: is there is a limit on how long the process is allowed? I got this in my error file :


      =>> PBS: job killed: walltime 21619 exceeded limit 21600


      (c009 is the cluster I am using)