Training-set size in each fold of 8-fold CV

Question

You have 1000 observations and run 8-fold cross-validation, splitting the data into 8 equal folds. How many observations are used for training in each fold (that is, everything except the held-out fold)?

Accepted Answer

Each fold holds 1000 / 8 = 125 observations for validation. The model trains on everything else: 1000 - 125 = 875 	ext{ observations}. So each of the 8 models trains on 875 points and is validated on the remaining 125. More folds means each training set is larger (closer to the full dataset), which reduces the pessimistic bias of the error estimate at the cost of more compute. Takeaway. Each fold's training set has n - n/k observations; with n = 1000 and k = 8 that is 875.

Training-set size in each fold of 8-fold CV

Your answer

More Statistics questions