Last modified: Thu Oct 27 14:46:59 EDT 2005 by Dean Foster

Statistical Data mining: Streaming searching

Admistrivia

• Homework 3 next class.

ID1 stepwise regression

• Nearest neighbor 1
• As we discussed previously, need to do subspace selection
• Modelling after forward stepwise
  • Test a bunch of features
  • Evaluate each on to see how it performs
  • Add the best feature

Classification evaluation

• calinski-harabasz
• T. Calinski and J. Harabasz. A dendrite method for cluster analysis. Communications in statistics, 3(1):1--27, 1974. 2.
• "C&H use the Variance Ratio Criterion which is analogous to F-Statistics to estimate the number of clusters a given data naturally falls into. They minimize Within Cluster/Group Sum of Squares (WGSS) and maximize Between Cluster/Group Sum of Squares (BGSS)" cpan

Beam search and streaming

• Beam as a compromize between breadth first and depth first
• Beams of width one: Same as depth first search
• Beams of unbounded with is stepwise in this context
• Not address in paper: order of variables