Glossary term
Glossary term
Foundations
The vector of raw (non-normalized) predictions that a classification model generates, which is ordinarily then passed to a normalization function. If the model is solving a multi-class classification problem, logits typically become an input to the softmax function. The softmax function then generates a vector of (normalized) probabilities with one value for each possible class.
Created for this library
An ML team logs raw logits alongside probabilities so downstream calibration checks can be performed independently.
A research team monitors logit distributions during training to spot saturating outputs that may indicate poor scaling.
A risk modeling team works with logits during calibration so it can compose multiple calibrated models without re-applying the softmax.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License