DanilaEremenko

@6sy666 I am reporting a significant drop in accuracy for regression task. I am including the model vs. model game matrices on the test sets for the classification task (Figure...

@Yinhuaihong Hi, I've created a repository to reproduce the issue ([LAMDA-TALENT-MNCA-REG-ISSUE](https://github.com/DanilaEremenko/LAMDA-TALENT-MNCA-REG-ISSUE)). The main experiment uses nested cross-validation with 5 inner and 2/5 outer folds. Model hyperparameters are optimized using Optuna...

@andreasgoethals I'm not sure; I have an inverse normalization for the regression within `LamdaLow.predict(..),` as you can see.

@Yinhuaihong @6sy666 This problem also occurs in classification tasks when using log loss for early stopping and log loss as the test metric for comparison.