https://github.com/vpavlin/odh-tensorflow-jobs
https://github.com/vpavlin/odh-tensorflow-jobs
Last synced: 2 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/vpavlin/odh-tensorflow-jobs
- Owner: vpavlin
- License: mit
- Created: 2018-11-12T09:54:40.000Z (over 6 years ago)
- Default Branch: master
- Last Pushed: 2019-03-11T12:14:17.000Z (about 6 years ago)
- Last Synced: 2025-01-23T03:45:15.367Z (4 months ago)
- Language: Jupyter Notebook
- Size: 8.84 MB
- Stars: 0
- Watchers: 2
- Forks: 2
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Open Data Hub Tensorflo Jobs demo
This repository contains artifacts for demoing a machine learning pipeline with Jupyter and Tensorflow on top of OpenShift.
# Content
## OpenShift Templates
You can find all the necessary OpenShift templates in `openshift` directory. Namely those are:
* Jupyter Notebook workspace
* Tensorflow training job
* Tensorflow serving container
* OpenShift secret for object storage configuration and credentials
## Training
You can find a script which runs the training and its dependencies in the `training` directory.
The training dataset was generated using (https://github.com/CermakM/char-generator) and can be found in the [`training` directory as well](training/num-dataset.tar.xz)
## Serving
The `serving` directory contains `s2i` build and run scripts which are used for TF serving container
# How to run
You can use `install.sh` script in `openshift` directory to deploy all templates into your OpenShift cluster:
```
cd openshift
bash install.sh
```
You can specify `GIT_REF` environment variable to pick a git refernce (branch or commmit) to use - `master` is used by default. You can also use `LOCAL` environment variable (with any value) to deploy from local clone of the repo.
Once the templates are imported you can either deploy them from OpenShift Catalog - you can filter for provider **Open Data Hub**; or directly from command line using OpenShift clients.
## Controlling from Jupyter Notebook
There is an example notebook ready in this repository - [odh-tensorflow-jobs.ipynb](https://github.com/vpavlin/odh-tensorflow-jobs/blob/master/odh-tensorflow-jobs.ipynb). To be able to run it successfully you will first need to deploy Jupyter Notebook server - simply head over to OpenShift Catalog in your namespace and click the Jupyter Workspace template. Follow the wizard and you should get a URL to your Jupyter Notebook server displayed in your namespace overview.
Login to your Jupyter server and upload the notebook from this repository. Follow the notebook to get a model trained and served.