EasyDeepLearn

When is cross-validation NOT necessary?

easy

Answer

  • When you have enough data that a single held-out validation set is already a low-variance estimate (typically  nval    10k  with  wellbehaved  labels)(\mathrm{typically}\;n_{\mathrm{val}}\; \ge \;10k\;\mathrm{with}\;\mathrm{well} - \mathrm{behaved}\;\mathrm{labels}).
  • At Google/Facebook scale you almost never do k-fold — a single split is enough.
  • Also: pure production monitoring uses live traffic slices, not CV.
  • Also: pure inference-time evaluation on a fixed test set.
  • CV shines on small-to-medium data where any single split's variance is meaningful.
Check yourself — multiple choice
  • Always use CV
  • On very large data, a single held-out validation is a low-variance estimate — CV isn't needed
  • CV is only for classification
  • CV is only for tabular data

Very large data ⇒ single val set is already low-variance. Save CV for small/medium data.

#cv#validation

Practise Supervised Learning

215 interview questions in this topic.

Related questions