Logit
Definition
A logit is, depending on context, the log-odds value or the unnormalized output of a classification model before softmax/sigmoid.
In binary logistic regression, the logit is log(p/(1-p)). In neural classifiers, raw scores before probability normalization are often also called logits.
Simply explained
Like having a raw value that is only translated into probabilities by sigmoid or softmax.
Why it matters in practice
Relevant for classification, sampling, and model APIs.
Example from business
A classifier outputs logits that are converted into class probabilities via softmax.
Limits and pitfalls
- Reading logit directly as probability; interpreting logit differences without transformation.
Origin & history
The logit concept originates from statistical modeling and was established through logistic regression.
Use in mid-sized companies
- Possible application in SMEs: Clearly distinguish raw outputs, probabilities, and calibration.
Frequently asked questions
Is a logit a probability?
No. How to get one? Via sigmoid or softmax.
Sources
- Logistic function program LOGIT · NIST, Logistic function program LOGIT
Sources checked on 10.10.2026.
Editorially reviewed on: