Child pages
  • Expected Log-Loss (7.0)

Contents

Context

Monitor's Contextual Menu

New Feature: Expected Log-Loss

One of the key metric in Information Theory is Entropy:

The entropy is thus the sum the Expected Log-Loss of each state  of variable  when using network ;

where:

As of version 7.0, you can now select to get the Expected Log-Loss of the states in the monitors instead of their probabilities.

Example

This insteresting example shows that setting Bronchitis = Yes reduces the Entropy of Dyspnea, without changing the Expected Log-Loss of Dyspnea = False.

Indeed, below are curves that show how the entropy is decomposed for a binary variable; this confirm that the Expected Log-Loss is similar for probabilities equal to 0.19 and 0.57.

 

Updated Feature: Monitor's Tooltip

The Expected Log-Losses are now also available int the tooltip associated with the monitors, in the Information Mode , while hovering over the monitor: