Timeline for High variance of leave-one-out cross-validation
Current License: CC BY-SA 4.0
5 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| May 4, 2023 at 4:58 | comment | added | paperskilltrees | @adosar Alijonov's answer is not well-written. But to address your comment, I believe, they meant the following. Split the dataset in $K$ folds: $D=D_1 \cup \dots \cup D_K$. CV estimation of performance is based on fitting a model to $D\setminus D_k$ (denote it $\hat f_{k}$) and evaluating it on $D_k$, and repeating for $k=1,\dots,K$. The "variance in the model" is how $\hat f_{k}$ are different from each other. In LOOCV $D\setminus D_1$ differs from $D\setminus D_2$ only in one point, so all $\hat f_{k}$ can be very alike, yet very different from the optimal model for infinite data. | |
| Nov 22, 2022 at 12:01 | comment | added | Antonios Sarikas | What is variance in the model? Cross validation techniques are used to estimate performance. | |
| Jun 14, 2022 at 12:10 | review | Late answers | |||
| Jun 14, 2022 at 12:59 | |||||
| S Jun 14, 2022 at 11:51 | review | First answers | |||
| Jun 14, 2022 at 13:06 | |||||
| S Jun 14, 2022 at 11:51 | history | answered | Alijonov | CC BY-SA 4.0 |