Model Training & Adaptation
Cross-validation
Checking AI using several different divisions of examples for learning and testing.
Example
A collection of information is divided into folds so different portions take turns being held out.
Why people use it
It checks whether performance holds up across several divisions of the available examples.
What you'll hear
“Does it work well across different held-out groups?”
What this means for you
Use splits suitable for the data's grouping and time structure.
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
- Does cross-independent checking remove all checking bias?
- No. Leakage, inappropriate splits and repeated tuning can still distort results.
- Does every example get tested in the same round?
- No. Different portions take turns being held aside while the system learns from the others.
- Can it cost more than one split?
- Yes. The system may need to be trained and checked several times.