DATA MINING Desktop Survival Guide by Graham Williams
Desktop Survival Project Home List of Figures List of Tables Data Mining with Rattle Introduction Data Mining with Rattle Data Exploring Data Transforming Data Cluster Analysis Association Analysis Classification and Regression Models Evaluation and Deployment Issues Moving into R Troubleshooting R for the Data Miner R Data Graphics in R Understanding Data Preparing Data Descriptive and Predictive Analytics Issues Evaluating Models Reporting Cluster Analysis Text Mining Text Mining Algorithms Bagging Bayes Classifier Cluster Analysis Conditional Trees Hierarchical Clustering K-Nearest Neighbours Linear Models Neural Networks Support Vector Machines Open Products AlphaMiner Borgelt Data Mining Suite KNime R Rattle Weka Closed Products C4.5 Clementine Equbits Foresight GhostMiner InductionEngine ODM Enterprise Miner Statistica Data Miner TreeNet Virtual Predict Appendicies Installing GTK, R, and Rattle Glossary Bibliography Index

Min Bucket (minbucket)

The minbucket is the minimum number of observations in any terminal leaf node.

The two variables minbucket and minsplit are closely related. In rpart if either is not specified then by default the other is calculated as $minsplit = 3*minbucket$.

Using rpart directly we specify minbucket within an option called control which takes the results from a function called rpart.control. In this example we

> audit <- read.csv(url("http://rattle.togaware.com/audit.csv"))
> audit.rpart <- rpart(TARGET_Adjusted ~ Age + Marital 
                                             + Occupation 
                                             + Deductions, 
                       data=audit,
                       method="class", 
                       control=rpart.control(minbucket=100))
> audit.rpart

Changing minbucket can result in different variables being chosen at different nodes. Compare the tree obtain with the command above (with minbucket set to 100) to the result when minbucket is set to 10. Note how node 7 was originally split using XnullXRattle!VariablesR functions (R function)Rattle!VariablesR libraries (R library)Rattle!VariablesR option (R option)Rattle!VariablesR packages (R package)Rattle!VariablesDatasets (Dataset)Rattle!VariablesRattle!VariablesAge but with the minimum bucket size set to 10 the node is split on XnullXRattle!VariablesR functions (R function)Rattle!VariablesR libraries (R library)Rattle!VariablesR option (R option)Rattle!VariablesR packages (R package)Rattle!VariablesDatasets (Dataset)Rattle!VariablesRattle!VariablesDeductions. We can see why -- the resulting node 15 has only 30 entities:

[...] 
  control=rpart.control(minbucket=100))
[...]
   7) Occupation=Clerical [...] 516 207 1 (0.40116279 0.59883721)  
    14) Age< 36.5 151  72 0 (0.52317881 0.47682119) *
    15) Age>=36.5 365 128 1 (0.35068493 0.64931507) *


[...]
  control=rpart.control(minbucket=10))
[...]
    7) Occupation=Clerical [...] 516 207 1 (0.40116279 0.59883721)  
     14) Deductions< 1299.833 486 207 1 (0.42592593 0.57407407)  
[...]
     15) Deductions>=1299.833 30   0 1 (0.00000000 1.00000000) *

Whilst the default is to set minbucket to be one third of minsplit there is no requirement for minbucket to be less than minsplit. A node will always have at least minbucket entities, and it will be considered for splitting if it has at least minsplit entities and on splitting, each of its children have at least minbucket entities.

Support further development through the purchase of the PDF version of the book.
PDF version is properly formatted and forms a comprehensive book (draft with over 600 pages).
Brought to you by Togaware.