Magnetic serves its customers by showing internet users advertisements that they are likely to click on or will lead them to make a purchase ("convert").
In other words, we attempt to place ads in a way that will maximize their click-through rates or conversion rates.
In order to bid on auctions for online ad placement in real time, we need to estimate the click or conversion probabilities to determine how much to bid. Most predictive models produce scores which indicate event ranks rather than event probabilities.
This paper describes the calibration algorithm - Monotonically Increasing Multi-Interval Continuous Calibration - that we developed to convert the model scores to probability estimates.
Sam Steingold has been doing data science since before it got that swanky name. He is the lead data scientist at Magnetic Media Online (they're hiring- http://www.magnetic.com/careers/ )and holds a PhD in Math from UCLA.He contributed to various open source projects (e.g., GNU Emacs,CLISP, Vowpal Wabbit).