Model Training & Adaptation
Stacking
Teaching another AI to combine predictions from several AI systems.
Example
Another AI learns how to combine several systems' predictions.
Why people use it
It lets another learner decide how to combine several systems' predictions.
What you'll hear
“Learn which predictions to trust in different cases.”
What this means for you
Keep the final test separate from the examples used to teach the combination.
Can you control it?
Developer-only
The people building or running the AI choose this setup. An everyday user generally needs their help to change how this part works.
Common questions
- Can stacking accidentally leak training information?
- Yes, if the combining AI system learns from improperly generated in-sample predictions.
- Must the original systems use the same method?
- No. Different kinds of predictors can contribute to the combination.
- Can the extra combining step add complexity?
- Yes. It adds another part to train, maintain and explain.