Block A · Lesson 03
Check a model on a new design
Plan how to check a model on a new component design, then see why a different split answers a different question.
PUBLIC LESSON PREVIEW
What you will be able to decide
Choose an evaluation split that answers whether a model can transfer to a complete component design it did not see during training.
- Before you start
- Know why training error is not a final check and why rows from one physical design are related.
- Example you will work through
- Assign six complete designs to training, validation and reserved groups before looking at model results. Keep every row from one design in one group.
- Small check
- Compare a row-random split with a design-grouped split and state which one matches the intended engineering use.
- Evidence boundary
- A grouped split tests a stronger transfer question, but it still proves only the documented dataset, target and design population.
Continue with the interactive lesson
Create an account and confirm your email to continue after lesson 1.
The exercises build on each other and saved work belongs to your learning account. The public preview and related technical resources remain available without signing in.