ML Monitor

This codebase builds on the original ML_monitor as presented in https://github.com/wiatrak2/ml_monitor This repository work on local installation of Python only and DOES NOT Support Colab at the moment.

Requirements:

• python >= 3.6
• docker

Installation

1. Clone and enter the repository

git clone https://github.com/sohiniroych/ML-Monitoring-with-Grafana.git
cd ML-Monitoring-with-Grafana

2. Install package using pip

pip install .

3. Install and Setup docker

sudo snap install docker
sudo apt install docker-compose

This process will set up the docker image that houses the Grafana Dashboad!

Processing (shown in the video)

1. Enable docker using the command

cd docker
sudo docker-compose up

• docker directory contains an easy setup of tools used for metrics visualization and analysis. These are Prometheus and Grafana. You should firstly start these programs, with docker-compose up command. Now you should be able to reach the Grafana admin panel on http://localhost:3000.
• Logging in for the first time use username: admin, password: ml_monitor

2. To set up the ml_monitor sensing through post 9090, open a separate terminal and enter the following:

python
import ml_monitor
ml_monitor.control.start()

You will get a message that a Prometheus UI should be reachable on http://localhost:9090.

3. Setup Jupyter notebook and open the file "ML_minitoring_Grafana.ipynb" Run this file. The first part will train the model and in the end random test samples will be generated to mimic a production setup. Increase the number of "epochs" to run for a longer Production-like setup

4. Open the Grafana dashboard at http://localhost:3000 and monitor the "n_rmse" metric.

Congratulations, now you know how to monitor metrics at production time!