4. Model, training, inference and features

These four terms are fundamental.

Feature: An input value used to make a prediction.

For customer-churn prediction, features might include:

  • Account age
  • Monthly amount
  • Number of complaints
  • Days since last login

Label: The correct historical outcome we want the system to learn.

Churned = true or false

Training: The process that uses historical examples to find useful patterns.

Model: The learned mathematical representation of those patterns.

Inference: Using the trained model to make a prediction for new input.

You train occasionally, but inference may happen thousands of times per minute.

PreviousNext