Skip to content

Tensorflow Workflow Example

Tensorflow is an open-source machine learning platform. It provides an ecosystem of tools and libraries that allow researchers to build and deploy machine learning applications.

Note

The flight environment will need to be activated before the environments can be created so be sure to run flight start or setup your environment to automatically activate the flight environment.

Installing Tensorflow

  • Create a singularity software environment: 
    [flight@chead1 (mycluster1) ~]$ flight env create singularity
  • Activate the environment: 
    [flight@chead1 (mycluster1) ~]$ flight env activate singularity
  • Create a conda software environment:
[flight@chead1 (mycluster1) ~]$ flight env create conda
  • Activate the environment:
[flight@chead1 (mycluster1) ~]$ flight env activate conda
  • Create a Python environment for tensorflow:
(base) <conda> [flight@chead1 (mycluster1) ~]$ conda create -n tensorflow python=3.6
  • Activate the Python environment:
(base) <conda> [flight@chead1 (mycluster1) ~]$ source activate tensorflow
  • Install the tensorflow package:
(tensorflow) <conda> [flight@chead1 (mycluster1) ~]$ pip install tensorflow==1.15

Running the Job

  • Download the example job models: 
    <singularity> [flight@chead1 (mycluster1) ~]$ git clone -b v1.13.0 https://github.com/tensorflow/models.git
  • Launch the tensorflow docker container with singularity to run the job: 
    <singularity> [flight@chead1 (mycluster1) ~]$ singularity exec docker://tensorflow/tensorflow:1.15.0 python ./models/tutorials/image/mnist/convolutional.py
  • Download the example job models: 
    (tensorflow) <conda> [flight@chead1 (mycluster1) ~]$ git clone -b v1.13.0 https://github.com/tensorflow/models.git
  • Execute the job with python: 
    (tensorflow) <conda> [flight@chead1 (mycluster1) ~]$ python ./models/tutorials/image/mnist/convolutional.py