Details of Lecture 28, ECE662 Spring 2010
April 27, 2010
In Lecture 28, we finished our discussion of decision trees by describing ways to address the "query selection" problem, the "when to stop splitting" problem, and the "what category to assign to the leaves" problem. We briefly discussed the horizon issue and mentioned that pruning can be used to address it. We then began the last section of the course on clustering by explaining the relationship between clustering and decision trees when the data is labeled, as well as giving examples of clustering problems in which there are truly no labels a priori (the animal species, family, classes, etc.). We divided clustering approaches into two distinct types based on the input data: feature vectors, or distance/dissimilarity measures.
Recall that Thursday's lecture (4-29-10) is canceled, and that there is a make up class Friday (4-30-10), 1:30-2:30 in EE117.
Previous: Lecture 27 Next: Lecture 29
