Demo App

A minimal toy application for demo and testing purposes. It can serve as a reference implementation of current best practices in the project (mirroring the DEEP template).

This demo module implements:

• dummy inference, ie. we return the same inputs we are fed. If some input is not fed we generate a default one.
• dummy training, ie. we sleep for some time and output some random monitoring metrics.

To launch it, first install the package then run deepaas:

git clone https://github.com/deephdc/demo_app 
cd demo_app
pip install -e .
deepaas-run --listen-ip 0.0.0.0

The associated Docker container for this module can be found in deephdc/DEEP-OC-demo_app.

Samples for media files are provided in ./data.

The two branches in this repo cover the two main usecases:

• master: this is a reference implementation on how to return a JSON response for predict().
• return-files: this is a reference implementation on how to return non-JSON responses for predict(). This is particularly useful when returning:
  • long responses (that could better fit better in a txt file),
  • media files (eg. returning an image),
  • multiple files (for example returning an image and a text file at the same time, packing them into a zip file).

The train() function is common for both branches.