Why is a trained model not the end of an ML project?
Options
- Because training results are random and cannot be trusted
- Because real-world data changes, so the model must be deployed, monitored and updated (Correct option)
- Because models stop working after a fixed number of days
- Because a model cannot be saved to a file
Answer
Because real-world data changes, so the model must be deployed, monitored and updated
Why
A model is trained on a snapshot of the past. Customers, prices, sensors and fraud patterns keep changing, so its quality can fall without any code change.
Shipping it is the start of the work: serving, watching it and retraining when needed.
No model has a built-in expiry date; it is the data around it that moves.
Remember
The world changes, so the model needs care after launch.