Cross Validation under Machine Learning


Under introduction to Machine Learning on Cross Validation,
the Learn Page states the followings;

In the following code block, we display the results of varying k from 3 to 23 . For each k value, we calculate and display the average RMSE value across all of the folds and the standard deviation of the RMSE values. Across the many different k values, it seems like the average RMSE value is around 128 . You’ll notice that the standard deviation of the RMSE increases from approximately 1.1 to 37.3 as we increase the number of folds.

but the output is;
3 folds: avg RMSE: 127.19146799819767 std RMSE: 7.80114274447321
5 folds: avg RMSE: 130.57004998596955 std RMSE: 15.968993082617418
7 folds: avg RMSE: 124.74000565490935 std RMSE: 23.009326104623764
9 folds: avg RMSE: 133.85427296864364 std RMSE: 20.275996691809862
10 folds: avg RMSE: 134.50358073016668 std RMSE: 30.83892745302988
11 folds: avg RMSE: 129.58548991863123 std RMSE: 22.39316430178567
13 folds: avg RMSE: 133.05101345639838 std RMSE: 27.88932598342725
15 folds: avg RMSE: 124.86715246014936 std RMSE: 37.03384132069149
17 folds: avg RMSE: 131.3786960290144 std RMSE: 40.043451719093724
19 folds: avg RMSE: 129.0143524209374 std RMSE: 44.3383982741942
21 folds: avg RMSE: 125.49498964946545 std RMSE: 41.03033829748872
23 folds: avg RMSE: 125.27939162120605 std RMSE: 41.668089858618046

The lowest avg RMSE is circa 124 while the RMSE STD runs from 7.80 to 41.66
from the output.

Please clarify

If you are asking about why the std looks bigger, an intuitive answer is as the folds get more numerous, the number of points in each test set gets smaller, then mean in the RMSE is averaging over less points and the RMSE is varying more wildly based on which test set it is evaluating on.

You can plot the prediction errors for a better view of how number of folds affect the model trained and the errors generated.