Mar-20-2021, 04:37 AM
I use logloss as a quality criterion, but it has a number of disadvantages:
1. Has infinite errors with confident misclassification.
2. Has a slight gradient in the "border" zone, where recognition is especially important.
Are there any modifications to this feature to improve the performance of machine learning algorithms?
1. Has infinite errors with confident misclassification.
2. Has a slight gradient in the "border" zone, where recognition is especially important.
Are there any modifications to this feature to improve the performance of machine learning algorithms?