A model learns from examples. How do you know it learned the pattern and not just the examples?
40 points; the model learns from 24, is compared on 8 and tested on 8.
RMSE: a typical error, in the same unit as y.
The weights are learned; the degree and λ are hyperparameters.
| x | y | Set |
|---|
The 32 points that are not test are split into k parts; each part validates once.
Every combination of degree and λ, scored with cross-validation.
lower error higher error
Train the winner on all 32 points and try it on the 8 test points: that is the number to report.