Data Science Process in the Google Cloud

The presentation were created in the scope of a project on tensorflow and Google Cloud Platform together with Fabien Tarrade, please visit github/tarrade/proj_DL_models_and_pipelines_with_GCP for latest developments.

Slides for

• Windows Users, underlying notebook (in nbviewer)
• Linux/Mac Users, view underlying notebook (in nbviewer)

Follow me on .

Setup to execute presentation notebooks

Fresh installation Anaconda

Install Anaconda Distribution on your system. This provides you with the most important packages on your system.

Install Jupter Notebook extension into base environment.

conda activate base
conda install -c conda-forge jupyter_contrib_nbextensions #triggers system-wide installation

Download the code from GitHub

• go to the directory in which you want to download the package from git
• download the package from Github:
  • git clone https://github.com/enryH/gcp_ml_engine_talk/
  • or download zipped repository.
• open a command line in the directory that contains the code from GitHub.

Create the python conda env

This will provide you a virtual environment of python packages needed to run the code.

• create a python env based on a list of packages from environment.yml
  conda env create -f environment.yml -n gcp_dl

• activate the env
  conda activate gcp_dl

In case of issue clean all the cache in conda conda clean -a

• In case you want to use nbdime for checking differences in notebooks:

  nbdime config-git --enable --global
  

• In case you want to use interactive slides using RISE:

  conda activate gcp_dl
  conda install -c conda_forge rise
  

Update or delete the python conda env

• update a python env based on a list of packages from environment.yml
  conda env update -f environment.yml -n gcp_dl

• delete the env to recreate it when too many changes are done
  conda env remove -n gcp_dl

TensorBoard

• open a command line
• go to the directory in which you want to download the package from git
• activate the env:
  conda activate gcp_dl
• execute TensorBoard:
  tensorboard --logdir ./results
• open a web browser (Firefox) and copy the link that appear in the prompt above:
  http://<something>:6006