Model Training & Adaptation
Principal component analysis
Combining related measurements into fewer new measurements that preserve major patterns of variation.
Example
A collection of information is summarized with a smaller number of principal components.
Why people use it
It summarizes many related measurements with fewer directions of variation.
What you'll hear
“Can we keep most of the variation with fewer measurements?”
What this means for you
Check whether dimension reduction preserves what matters for the task.
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 a small pattern matter more than a large one?
- Yes. A pattern explaining little overall variation could still be important for the particular prediction you care about.
- Can the new directions be hard to name?
- Yes. Each can combine several original measurements rather than correspond to one familiar concept.
- Can different measurement scales affect the result?
- Yes. Large numerical ranges can dominate unless the measurements are prepared appropriately.