Dockerfile for NUS-MIT Datathon 2018 server setup @ nscc

Run Jupyter

docker run \
    --runtime nvidia \
    -u $UID:$GID \
    -v /home:/home \
    -e USER \
    -e HOME \
    -w $HOME \
    -p 8888:8888 \
    --rm -it \
    datathon:cuda9.1-ubuntu16.04 \
    /bin/bash -c \
    "jupyter lab \
        --LabApp.open_browser=False \
        --LabApp.ip='0.0.0.0'"

Run Jupyter @ nscc

nvidia-docker-run --net=host kiend/datathon-nscc:cuda9.0-ubuntu16.04 << EOF
    export HOME=$HOME
    export USER=$USER
    cd $HOME
    jupyter lab --LabApp.ip='0.0.0.0'
EOF

Python

Default Python (/opt/bin/python) uses Python 3.6.5.

Pre-installed frameworks:

• tensorflow
• pytorch
• mxnet
• caffe*
• caffe2*
• theano
• keras (keras-mxnet)
• xgboost
• pymc3
• Packages in the Python data science stack: numpy, scipy, pandas, sklearn, matplotlib, ... Run pip list for the full list.

To install other packages: pip install --user <package_name>

* To use caffe or caffe2 in Python, first run import caffe_path or import caffe2_path, i.e.

import caffe_path
import caffe

or

import caffe2_path
import caffe2

caffe CLI (caffe.bin) is available by default without any extra configuration.

R

version 3.5.0

Pre-installed packages:

• tidyverse
  • dplyr
  • tidyr
  • ggplot2
  • ...
• data.table
• xgboost
• caret
• rjags
• tensorflow**
• keras**
• ... Run installed.packages() inside R for the full list.

To install other packages: install.packages("<packages_name>")

** tensorflow and keras for R have already been properly configured. There is no need to run install_keras() or install_tensorflow().