CDH-Tree

This is an implementation of the CDH-Tree Algorithm described in Chapter 22 'Classifying Streaming Data II: Time-Dependent Data'
of Principles of Data Mining.


For the experiment with dataset lens24 described in Chapter 22 set
Order of Inputting Records: alternate   Alternation Frequency: 19200   Substitute Attribute Values: 0,2,3,1
Number of Repetitions: 2500  Window Size: 9600  D: 14000  T: 18000  M: 1200
Spypoints: 9600,28000,36000,38400,42000,48000,57600
(use default values for other settings).


Dataset  

Grace Period (records)  

Probability for Use in Calculating Hoeffding Bound   0.9    0.95    0.99    0.999

Order of Inputting Records   standard   reverse   random   alternate

(Only if alternate selected) Alternation Frequency  

(Only if alternate selected) Substitute Attribute Values  

Number of Repetitions of Training Data (minimum 100)  

Use Hoeffding Bound   yes no

acceptequals   yes no

Window Size W (minimum 1000)  

D (minimum 1000)  

T (minimum 1000)  

M (minimum 1000)  

Spypoints (leave blank for none)  

trace1   yes no    Main Trace Facility for Considering Splitting at a Node When Grace Count Achieved
trace2   yes no    Displays class/attval/total Values at All Expandable Leaf Nodes, When Grace Count is Reached at a Node
trace3   yes no    Display Class Totals at All Nodes, When Grace Count is Reached at a Node
trace4   yes no    Displays the Value of Information Gain For Each Available Attribute When Considering Expanding a Leaf Node
trace5   yes no    Displays Information about Branches and Nodes in the Classification Tree
trace6   yes no    When Grace Count is Reached at a Node, Displays Confusion Matrix of Predictions for All Records Since Previous Occurrence
trace7   yes no    Trace 'reviewInternals' Routine to Check for Concept Drift
trace8   yes no    Trace Testing Mode