AIO2024 MLOps Course

Introduction

1. Basic Deployment

In this first week, we build a simple application (web application) using 2 python packages: Streamlit and Gradio for serving 2 our models.

See the Week 1: Basic Deployment for more details.

2. Client Server Deployment

In this second week, we will re-deploy previous applications using but separate the frontend and backend applications. We use FastAPI for the backend server and Gradio for the frontend application.

See the Week 2: Client Server Deployment for more details.

3. Docker

In this exercise, you will containerize your application using Docker.

• First, cretae a Docker network to connect the backend and frontend containers.
• Build and run the Docker container for the backend server.
• Build and run the Docker container for the frontend application.

See the Week 3: Docker for more details.

4. AWS Services

In this week, we will explore AWS services to deploy our application.

See the Week 4: AWS Services for more details.

5. Airflow

In this week, we build a simple pipeline using Apache Airflow. We will create a DAG to schedule the training and testing of our models.

Dataset

• BTC Price collected from 2016 to 2024 and split into 3 versions:

  • v0.1: 2016 - 2022
  • v0.2: 2016 - 2023
  • v0.3: 2016 - 2024

• Gold Price (Optional) collected from 2016 to 2024.

Lifecycle

• First we will train the model using the v0.1 dataset, taging the model as V1.
• Then we will test the V1 model using the data in 2/2024.
• Next, we will retrain the model using the v0.2 dataset, taging the model as V2.
• Finally, we will test the V2 model using the data in 2/2024.

See the Week 5: Airflow for more details.

6. MLflow

In this week, we will use MLflow to track the model training and testing process. We will also use MLflow to version the models.

See the Week 6: MLflow for more details.

7. Monitoring

In this week, we will use Prometheus and Grafana to monitor the performance of our application.

See the Week 7: Monitoring for more details.