execute_mlde.py getting only to 23%

[chelseashuu]

Hello! I have been trying to run execute_mlde but kept running into conflicts between theano, libcublas, pygpu, cuDNN, and tensorflow. The closest I've gotten to a successful run is on GPUs and reached 23% in default training. However, that is the farthest the program will go. Even after 48+ hours the program doesn't run any further and doesn't generate the specified outputs. Is this a common occurrence? Do I need to allocate more GPUs?

Thank you

[brucejwittmann]

@chelseashuu are you using the mlde or mlde2 conda environments? The versions of the packages you mention should be compatible in there.

You certainly shouldn't need more GPUs. All models within MLDE should work just fine on a CPU unless you are working with a truly massive combinatorial space (much more than the 4 sites in the original paper).

If you are working within the mlde or mlde2 environments, can you copy your exact inputs and the console logs? That would help with troubleshooting.

Also, I don't typically check this repo anymore. If you need help, please tag me, otherwise I might take a while to get back to you.