y_uncertainty for unknown example
Currently, one can determine noisy label detection following the procedure described in section 3 of the paper Tang, Y.-H., Zhu, Y. & de Jong, W. A. Detecting Label Noise via Leave-One-Out Cross-Validation. 1–11 (2021)
.
From the source code, y_uncertainty computes the magnitude of the noise for a dataset. It needs sigma and y_std
. I understand sigma is estimated during the fitting procedure for the "entire dataset", but I wonder if there is any way to predict y_uncertainty
after fitting the GPROutlierDetector
for an unknown example? If not, are there any plans to add something like it?