Usage

Data requirements

The algorithm can handle numerical as well as categorical attributes. There is no requirement to binarize or standardize the data prior to tree building. The only limitation is for the number of values of the nominal attribute. Typically, this number cannot exceed 100.

Model building and testing

Function settings

Overview of algorithm settings

Full specification of the model settings contains the elements of the General Algorithm Settings and Transformation Settings.

The class responsible for providing the algorithm settings specific to Classification Trees is SmartTreeSettings. See javadoc for the programming details of using it in Gython.

Table 31.1. SmartTreeSettings properties summary table

OptionDescriptionPossible valuesDefault value
Aggregating methodMethod used in scoring to assign weights to trees. Method1: validation data. Method2: 0.632*validation + 0.368*train.Method1, Method2Method2
Change usages on each splitTrue: select a random set of active variables on each split; false: active variable configuration is set once for a treetrue/falsetrue
Min Node Variability PrcVariability of the minNodeSizePrc parameter. 0: inactive. This parameter is set once for a treeReal numbers from the interval [0,100]5.0
Multiple treesHow many bootstrapped classification tree models to create.Integer numbers grater or equal than 1100
Number of attributes per each splitNumber of active attributes to select for each split or each tree. -1: sqrt(number of active attr.) is used.-1 or integer numbers greater then 1 -1
PositivePositive target category for lift. Null: the first category will be used.Category names from logicalDatanull
Prob AggregateTrue: each tree outputs a probability value and the probabilities are averaged for the final result. False: each tree produces a decision and the final results is based on the proportions of the votes.true/falsetrue
QuantilesNumber of lift quantiles; if <=0 then don't calculateinteger numbers0
Save TreesTrue: the built trees wil be saved and each tree can be viewed and applied. False: only importance and performance statistics will be saved and the model cannot be applied and tested.true/falsetrue
Split Evaluation VariabilityTrue: each tree will be built using a split evaluation method selected randomly from: entropy, gini, gainRatio. False: only one setting will be used.true/falsetrue
Stratified SamplingTrue: bootstrap samples from each target value separatelly. False: bootstrap samples from the entire training set.true/falsetrue
ThreadsNumber of threads used to build the model. -1: use the maximum number of available threads.integer numbers1
Variable SamplingVariable sampling algorithm. True: sample N variables from all active variables without replacement. False: each tree will take the consecutive N variables from the list without sampling.true/falsetrue
Weight from valid errTrue: during scoring each tree is assigned a weight based on its validation data errortrue/falsetrue
Weighting StatError statistic used to evaluate each model on out-of-bag data and weight it in scoring.sse, acc, mae, continuousmae