A few scripts showing how to use udocker with Docker hub images (e.g. Tensorflow, MXNet)

Official TensorFlow Docker images from https://hub.docker.com/r/tensorflow/tensorflow/tags/

Example how to use official TensorFlow images with udocker:

1. install udocker in your preferred directory:

$> curl https://raw.githubusercontent.com/mariojmdavid/udocker/devel/udocker.py > udocker
$> chmod u+rx ./udocker
$> ./udocker install 

2. look into udocker_pull.sh script and specify which TensorFlow you want to pull. Note the name of the created container. List of available containers is given by udocker ps

$> ./udocker_pull.sh tensorflow/tensorflow:1.7.0-gpu

or just

$> ./udocker_pull.sh

(default is set in udocker_pull.sh, e.g. DEFAULTIMG=tensorflow/tensorflow:1.6.0-gpu)"

3. have a look into job_udocker.sh to adjust e.g. what example script testscript-tf.py or testscript-mx.py to run. Run the script with the proper udocker container name (check with udocker ps).

Other important variables to check in the script:

UDOCKER_DIR="$PROJECT/.udocker" - udocker main directory.

HOSTDIR=$PROJECT - directory at your host to mount inside the container.

USERScriptDirHost=$HOSTDIR/workspace/udocker-example - location of the user program (host)

DIRINCONTAINER="/home" - mount point inside container