How do I deploy a Cloud ML Training job with additional python dependencies?

Question

I use additional python dependencies in my tensorflow program. How do I deploy the training job through Cloud ML?

Answer 1

If your dependencies are on PyPI, you can declare them as dependencies in a custom setup.py file. See http://python-packaging.readthedocs.io/en/latest/dependencies.html . Once you have authored your setup.py file, use gcloud beta ml jobs submit training as normal and your code will be built using your setup.py file.

Run gcloud beta ml jobs submit training --help for more information.

If your dependency is not on PyPI, you can follow the instructions here, http://python-packaging.readthedocs.io/en/latest/dependencies.html#packages-not-on-pypi, or you can manually pass your tar.gz archives to gcloud with, the --packages flag.