Gene clustering: 2001:
2
Some Problems
§
Naive Bayes
•
good accuracy
•
ignores dependencies
•
discriminant model, not characteristic
§
PF
OIL
search is limited
•
“waterfall” effect
•
Hard to find good heuristics for top-down rule builders
§
Predictive accuracy is not the real goal
•
no unknown instances to predict
•
expect low coverage
Ž
poor accuracy
•
small data set
Ž
test set costly and unreliable