iflow / zero-shot-prediction-dashboard

Machine-Learning deployment pipeline and Covid-Dashboard (MDS-SDC Project)

ML deployment with Tensorflow and Gradio

A MDS-SDC project

Overview

Our project contains three docker containers

1. GradioApp for using a ML models to classify images
2. Tensorflow Model Server providing the image classifier
3. Covid Dashboard build with Mercury

The docker container can be used during development on a local docker desktop instance.

For CI/CD the docker container is 0) triggerd by a push of the new sources into this Github repository,

1. build using Githab actions in this repo,
2. pushed to DockerHub,
3. deployed to Azure Services by
4. pulling the images from DockerHub

The interface of the covid dashboard:

The GradioApp interface

for classifying images waits for an image as input and outputs the predictions together with the probabilities and below a generated wordcloud of the predictions in their relative importances.

Further links

Wordcloud in Python
https://pypi.org/project/wordcloud/\

Gradio as Docker image
https://github.com/njanakiev/minimal-gradio

TF-Serving using Huggingface models
https://huggingface.co/blog/tf-serving-vision

Tensorflow-Serving

Run setup.py to generate model files for the docker image

Docker

How to run

docker-compose up -d

How to stop

docker-compose stop

Local usage

Gradio App

http://localhost:7861

Dashboard

http://localhost:7862

TF-Modelserver

http://localhost:8501