A data scientist uses Platt scaling to calibrate a gradient boosting model's output. What does Platt scaling specifically fit to the raw model scores?
Select an answer to reveal the explanation.
Short Explanation
Here's the deal — Platt scaling fits a two-parameter logistic sigmoid to the model's raw scores using a held-out calibration dataset, so the probabilities better reflect true event rates.
Full Explanation
Platt scaling specifically fits a logistic (sigmoid) regression with two parameters to the model's raw scores using a held-out calibration dataset, producing probabilities that better reflect true event rates. The monotone step function via isotonic regression describes isotonic regression calibration, which is non-parametric. The Beta distribution transformation to map scores to the unit interval describes Beta calibration. The temperature parameter dividing the logit to soften predictions describes temperature scaling, a single-parameter method used mainly with neural networks.