Problem
This is a simple end-to-end mlops project which takes data from capital bikeshare and transforms it with machine learning pipelines from training, model tracking and experimenting with mlflow, ochestration with prefect as workflow tool to deploying the model as a web service.
The project runs locally and uses AWS S3 buckets to store model artifacts during model tracking and experimenting with mlflow.
Dataset
The chosen dataset for this project is the Capital Bikeshare Data
Improvements
In the future I hope to improve the project by having the entire infrastructure moved to cloud using AWS cloud(managing the infrastructure with iac tools such as terraform), have model deployment as either batch or streaming with AWS lambda and kinesis streams, a comprehesive model monitoring.
Project Setup
Clone the project from the repository
Change to mlops-project directory
Setup and install project dependencies
Add your current directory to python path
Start Local Prefect Server
In a new terminal window or tab run the command below to start prefect orion server
Start Local Mlflow Server
The mlflow points to S3 bucket for storing model artifacts and uses sqlite database as the backend end store
Create an S3 bucket and export the bucket name as an environment variable as shown below
In a new terminal window or tab run the following commands below
Start the mlflow server
Running model training and model registery staging pipelines locally
Model training
Register and Stage model
Create scheduled deployments and agent workers to start the deployments
Create work queues
Run deployments locally to schedule pipeline flows
Deploy model as a web service locally
Change to webservice directory and follow the instructions here
