🎯  Platt Scaling

Platt Scaling#

Fitting a logistic function to scores to produce calibrated probabilities.

Important

✨ AI-generated content. This page was written with the assistance of an AI language model and is provided as a learning aid. Despite careful review, it may still contain mistakes, omissions, or out-of-date information. Whether you are new to the topic, a team lead, or a senior practitioner, treat it as a starting point rather than an authoritative reference: read it critically and independently verify anything you act on (code, commands, figures, and factual claims) against official documentation and primary sources before relying on it.

What it is#

Platt scaling calibrates a classifier by fitting a logistic (sigmoid) function on its raw scores, mapping them to probabilities:

\[\hat{Q} = \sigma(a\,z + b).\]

The two parameters a and b are fit by negative log-likelihood on a validation set. It was invented by John Platt for SVMs.

How it relates#

It is a parametric method that assumes a sigmoid-shaped miscalibration. Temperature scaling is essentially its one-parameter, multi-class special case (fixing the slope and dropping the offset), so the two are close cousins.

When to use it#

Platt scaling is a solid default for binary classifiers with monotonic score miscalibration and limited calibration data, since two parameters rarely overfit — but if the true miscalibration isn’t sigmoid-shaped, a more flexible method like isotonic regression fits better.


Theme: Probability Calibration  ·  All terminology



See also

Source article Adapted (context, re-expressed) in our own words from: Platt Scaling (insightful-data-lab.com).

Tags: purpose: reference topic: terminology level: advanced