How many models does leave-one-out CV train?

Question

You have 250 observations and run leave-one-out cross-validation (LOOCV): each observation takes a turn as the single held-out validation point while the model trains on all the others. How many separate models are trained in total?

Accepted Answer

LOOCV is k-fold cross-validation with k = n: every observation forms its own fold of size 1. So the model is trained once per observation, each time on the other n - 1 points. 	ext{models trained} = n = 250. That is why LOOCV is expensive: it fits the model 250 times here, and for large datasets the count becomes prohibitive, which is why 5- or 10-fold is the usual compromise. Takeaway. LOOCV trains exactly n models, one per held-out observation, making it accurate but costly for large n.

How many models does leave-one-out CV train?

Your answer

Discussion

Learn the concepts

Related questions