This preview shows pages 1–19. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: 1 CMPSCI 383 October 30, 2008 Learning and Overfitting 2 3 4 5 Today ’s topics • Pathologies of learning algorithms • Properties of evaluation functions as statistical estimators • How evaluation functions and search interact in learning • Methods for solving pathologies 6 Amazing Facts • Mutual funds that beat the market! • Winning the lottery, twice! • Hidden prophesies in text! • Amazing coincidences in twins raised apart! • Psychic watch repair! 7 8 Learning terminology • This term refers to a learned structure that expresses a conditional or joint probability distribution. • This mathematical expression is used to “score” models. • In learning, the evaluation function depends on these two arguments. • What is a Model • What is an Evaluation Function • What are Models and Data 9 Example problem 10 Example rule (1) Trains with small closed cars do not carry toxic chemicals 11 Example rule (2) Trains with with only a small and large car, or a jaggedtop car, carry toxic chemicals 12 How did you devise your rules? Did you… • Look for characteristics in one set but missing in the second set? • Examine several potential rules? • Consider simple rules first? • Reject potential rules that didn’t perform well? You performed search 13 How did you evaluate your rules? • Examined how many trains were correctly and incorrectly predicted • Assigned lower value to rules which predicted fewer trains correctly. • Did you do anything else? 14 How reliable are these rules? • For any given train, how confident are you that the answer is correct? • Do we have enough data to construct a reliable rule? How many data points is enough? 15 Pathologies of Learning Algorithms 16 • Overfitting Adding components to models that reduce performance or leave it unchanged • Oversearching Selecting models with lower performance as the size of search space grows • Attribute selection errors Preferring attributes with many possible values despite lower performance (Jensen and Cohen 2000) Pathologies of induction algorithms 17 Sample size (Oates & Jensen 1997, 1998, 1999) Accuracy Tree size Overfitting 18 Overfitting (continued) (Oates & Jensen 1999) 19...
View
Full
Document
This note was uploaded on 02/06/2011 for the course CMPSCI 383 taught by Professor Staff during the Fall '08 term at UMass (Amherst).
 Fall '08
 STAFF

Click to edit the document details