SWITCH: An Exemplar for Evaluating Self-Adaptive ML-Enabled Systems

"SWITCH" is an exemplar developed to enhance self-adaptive capabilities in Machine Learning-Enabled Systems through dynamic model switching at runtime. It is designed as a comprehensive web service, catering to a broad range of ML scenarios, with its implementation demonstrated through an object detection use case.

Getting started

The system comprises distinct components, including a frontend written in React, a backend powered by FAST API in Python, and Elasticsearch serving as the database. We will be using YOLOv5u as the object detection model.

Requirements:

• Install Docker Engine and Docker Compose as standalone binaries.
• Minimum free space of 7 GB.
• The system is designed for Linux environments. If you do not have access to a Linux system, you can either utilize a virtual Linux machine or employ a Docker image of a Linux environment.
• If using a virtual environment for setup, pre-install virtualenv.

Installation

Video for steps of installation and running the application can be found at:

1. Youtube
2. Google Drive

Clone the GitHub repository:

            
git clone https://github.com/sa4s-serc/switch
cd SWITCH
            
            Copy
        

Setting up Elastic Search and Kibana

Start the docker engine on your system.

Start Elasticsearch and Kibana containers using Docker Compose with image version 7.9.1, using the command:

            
docker-compose up
            
            Copy
        

Wait until ready; this may take time to install depending on the internet connection.

You can check if it's ready by accessing Elasticsearch at http://localhost:9200/ and Kibana at http://localhost:5601/.

Leave the Docker container running. Use a different terminal for the rest of the steps.

Setting up Virtual Environment (optional)

Using virtualenv allows you to avoid installing Python packages globally.

            
python3 -m venv ./venv
source venv/bin/activate
            
            Copy
        

Installation (only one is required)

Installation using setup.sh

            
chmod +x setup.sh
./setup.sh
            
            Copy
        

OR

Manual Installation

Importing Dashboard

            
    curl -X POST "http://localhost:5601/api/saved_objects/_import" -H "kbn-xsrf: true" --form file=@export.ndjson
            
            Copy
        

Setting up Frontend

            
    npm install
            
            Copy
        

Setting up Backend: Model loader, MAPE-K, Locust a load tester.

            
    cd NAVIE
    pip install -r requirements.txt
    python3 process_model.py
            
            Copy