Counteracting drift
As discussed in the introduction, model drift is bound to happen. Maybe it happens very slowly or maybe it occurs quickly, but it is something that cannot really be avoided if we don't try to actively counteract it.
What you will realize in the coming section is that online learning, which has been covered extensively in this book, is actually a very performant method against drift. Although we had not yet clearly defined drift, you will now come to understand that online learning has a strong added value here.
We will now recapitulate two approaches for counteracting drift, depending on the type of work you are doing, as follows:
- Retraining for offline learning
- Online learning
Let's start with the most traditional case, which is offline learning with retraining strategies implemented against model decay.
Offline learning with retraining strategies against drift
Offline learning is still the most commonly used method for...
