Hands-On Machine Learning with C++

Log-Loss

The logistic loss function value (the Log-Loss) is used as a target loss function for optimization purposes. It is given by the following equation:

We can understand the Log-Loss value as the accuracy being corrected, but with penalties for incorrect predictions. This function gives significant penalties, even for single miss-classified objects, so all outlier objects in the data should be processed separately or removed from the dataset.

There is a class called CrossEntropy in the Shark-ML library that can be used to calculate this metric and use it as the loss function. The Shogun library has a class called CLogLoss for the same ...