ML Model Life Cycle
A typical Machine Learning project life cycle:
Image credit: "Machine Learning Engineering" by Andriy Burkov
Business Problem
Consists of understanding the business problem at hand and propose how to solve it with one or multiple machine learning (ML) projects.
Goal Definition
The goal of a single ML project is usually just a part of the business objective, which is what your organization is trying to achieve. The defined goal must specify the inputs of our statistical model, its outputs and its accepted tolerance to be considered a "good" model.
Data Collection and Preparation
Collection and preparation of the raw data to be later used in a ML model.
Feature Engineering
Transformation of raw data into feature vectors.
Model Training
The pre-selected ML model is trained with the collected data recently transformed into feature vectors.
Model Evaluation
The performance of the model is evaluated before we deployed it.
Model Deployment
After testing the ML model, it can now be deployed. The model can accept queries generated by the user.
Model Serving, Monitoring, and Maintenance
Once the model is deployed it must be constantly monitored in case it needs maintance due to new incoming data (ie. Model Serving).
