Validation set
Also called Validation Set / Dev Set
A slice of data the model never trains on, used to tune choices and check progress honestly.
Think of it like
The practice exam you grade yourself on — not the real test, but a fair read on whether you are ready.
Example
The team compares two learning rates by which one scores better on the validation set, then locks that choice in.
How it actually works
Data is split into training, validation, and test. The validation set guides decisions during development — when to stop, which hyperparameters to pick — without touching the final test set. Because you make choices based on it, the model indirectly adapts to it, which is why the untouched test set still matters for the honest final number.
For product teams
It is how you tell real progress from wishful thinking while a model is still being built.
For engineers
A held-out split used for hyperparameter selection and early stopping, kept separate from the final test set.
Related
- Training set — The data the model actually learns from.
- Test set — The untouched final exam.
- Early stopping — Uses this signal to know when to stop.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome