Add Jupyter Notebook to train tensorflow-housing model

[piotrpdev]

Description

Whether to include pyenv has been debated, but we already have a working notebook that installs pyenv and miniconda. I think the right approach is for us to include this notebook for now and if there are security concerns/other issues we can change it in the future.

For now, I think we should add instructions on how to use the notebook and tell the user to copy the generated model files to S3, so it can be used in the pipelines. This is a crude approach but is only temporary until #112 #115 are finished.

Note
See #84 for more working examples of notebooks.

A/C

• Add a Jupyter Notebook that:
  • Works in a Data Science Project Workbench created by Open Data Hub
    • Ideally running the latest "Minimal Python" image
  • Trains the MLflow example TensorFlow model
• Add documentation on how to use the notebook

[piotrpdev]

@adelton @LaVLaS Thoughts?

[adelton]

Conceptually majority of the pipelines/* content belongs to some generic Open Data Hub demo / tutorial / learning repo, and the Edge PoC should just build on top of that, adding only the Edge-specific bits. Or to put it differently, the Edge PoC should not really be maintaining these bits and models and their documentation, we should just use some generic Open Data Hub ones.

So I wouldn't spend too much effort on making the training part perfect.

One of the reasons is the maintenance liability that comes with having this content, as well as liability of recommending the pyenv approach which may lead people to use old / insecure versions of the stack.