search

oegedijk / explainerdashboard

Quickly build Explainable AI dashboards that show the inner workings of so-called "blackbox" machine learning models.
http://explainerdashboard.readthedocs.io
MIT License
2.32k stars 332 forks source link

Logodds: difference between contributions plot and prediction box #294

Open rita-a-oliveira-alb opened 10 months ago

rita-a-oliveira-alb commented 10 months ago

From the code

image

I built an explainer dashboard, which includes the following outputs:

ExplainerDashboard_PredictionBox ExplainerDashboard_ContributionsPlot

The underlying problem is one of multiclassification, with classes 0, 1 and 2. The model used was LGBM.

In the previous images, I chose class 2 as the positive class. My question is why the logodd predicted in the Contributions Plot is different from the logodd of class 2 in the Prediction box and how is the first logodd calculated?