Problem with bsub when I process the training

[Denescor]

Hello

I have an issue when I want to learn the model coming from the bsub lib.

I followed all the steps to install the model and pre-process the data and all works without big problems.
I'm using a miniconda environment when there is python 3.5 (because of an error with python 3.5 during the pre-process) but my default python (i.e. when I write the command python) is 2.7.

First, bsub was not in the requirements.txt, so it wasn't installed when I launch the learning command in the readme. Moreover, for incompatibility reasons with python 3, I can't install it with conda. I succeed with pip and install it for python 2 and python 3.

However, if I launch bsub with python 3 (i.e. python3 -m bsub -n 2 -W [...]), the command crash because of an syntax error in the lib on a print without parenthesis (allowed in python 2 but not in python 3) showing that bsub is definitely incompatible with python 3.
If I launch with python 2 (i.e. python -m bsub -n 2 -W [...]), I only have in return empty job and no training is processed.

Nevertheless, if I launch the training without bsub (i.e. for v in 1 2 3 do python3 -m model.train \ [...]), the training is processing correctly but only on CPU.
Currently my training is running like this.

So, my questions are :

• How install and/or configure bsub (version, requirements, etc...) ?
• Is the initial training script (train.py) planing to use GPU or only CPU ?

Thanks for answering and helping me.

[lbozarth]

I have the same issue. Did you solve it?