Sunday, March 21, 2010

Least Square Methods: simple form of regression

Sometime simple solutions are the most effective. If you have a series of events e1, ... en, where each ei appears occ(ei) times with probability p(ei), you can represent the events in a Cartesian plan with axis containing the occurrences and the probability, respectively. Then you can imagine a straight line y = m x + b such that the distances from the line to the points in the space is minimized. Surprisingly enough, this method is straightforward has a closed solution and is very effective to predict the behavior of unseen points in the space.

No comments:

Post a Comment