Creating a conda environment

Here's a recipe to create a new conda environment with multi-source Sockeye (NRC).

bash

# we don’t have sufficient /tmp space to install thus point to another local dir
export TMPDIR=~/tmp

# Choose your version based on a sha1 from the git repo
export git_sha1=5123f668c
export git_sha1=1945ebf05
export git_sha1=ceeb7ab03

# Where you want your conda environment to reside.  I suggest using the sha1
export env_prefix=/space/group/nrc_ict/project/sam037/WMT2019/sockeye-multisource.$git_sha1

# Make sure you have conda in your PATH which is not always the case on the gpsc
source /space/group/nrc_ict/project/pkgs/ubuntu18/gcc-7.3.0/miniconda3/activate

# Create a conda environment.  In this example I specify where I want it but you could simply use --name sockeye-multisource.$git_sha1 instead
conda create  --yes  --prefix  $env_prefix  python=3.6 tensorboard tensorflow matplotlib ipython

# Get in the environment
source activate $env_prefix

# Sometimes pip is out-of-date
pip install --upgrade pip

# This is where you install from our repo.  I guess this should actually point to github instead
pip install git+file:///home/sam037/sockeye@$git_sha1
or
pip install git+ https://github.com/nrc-cnrc/sockeye-multisource@$git_sha1

# Add some helper libs
# This line is important for ` mxnet-cu100mkl==1.3.1` or else you won’t go full speed
pip install mxboard pudb subword-nmt sentencepiece mxnet-cu100mkl==1.3.1

# I like to record what was installed which is not entirely capture by conda’s checkpoints
conda list > $CONDA_PREFIX/conda.list.txt
exit