Cluster

[bartvm]

Includes stuff from both #39 and #38 (sorry to stack PR on PR, I'll merge everything once it's all tested a bit more thoroughly).

There's a submission script (nmt.pbs) to start jobs on the cluster, with the following changes:

• Make logging ports configurable so that multiple jobs can run on the same device.
• Random ports for logging, batches and control; this way you can start multiple jobs without having to set different ports for each manually (else they'll get in each other's way on the same node)
• Use relative file locations in config.json e.g. train/wmt16.de-en.tok.true.clean.shuf.en instead of the absolute path. The submission script then replaces this with the correct path to use on the cluster. A similar script can be used to submit things on Kepler, so that the same configuration files can be used in both cases.
• environment.yml is a Conda environment that describes the minimum requirements to run the experiment.
• I added jq to the cluster in $RAP/nmt/bin
• For some reason the default encoding for jobs is something very weird, so be sure to set the LANG=en_US.UTF-8 in the environment (done in nmt.pbs).

[bartvm]

Wrong target branch