Example 30.1. Building a tree model

# Description:
#    This example shows how to create a decision tree classifier.
#    The dataset used is 'zoo'. The predicted target id is 'catsize'.
#    The algorithm tries to learn on other attributes, whether 
#    animal size is catsize or not.
#
#   animal attribute is turned off 
#   (by setting its usage to inactive)
#
# Remarks:
#   This script uses data which was generated by the "zoo" script from 
#   the "data" folder

if not tableExists('zoo'):
 raise "Table 'zoo' does not exists. Please run zoo.gy script from data directory first"


# Create building task object, this will produce 
#  sample_cl_settings, and after executing this 
#  task, created_model will also be available
bt = MiningBuildTask('zoo_data', 'zoo_cl_settings', 'created_model')

# Create physical data from a previously prepared dataset
pd = PhysicalData('zoo')

# Generate logical data from physical data
ld = LogicalData(pd)

# New Classificatin function settings
fs = ClassificationFunctionSettings()

# the generated logical data is assigned to function settings
fs.logicalData = ld

# Turn off attribute animal from learning
fs.attributeUsageSet.getAttribute('animal').usage =\
 UsageOption.inactive

# TargetAttribute name describes which value is taken as the 
# target for classification
fs.targetAttributeName = 'catsize'

# Will use Decision Tree algorithm settings
fs.algorithmSettings = TreeSettings()  

# Sets the minimal sample count for node to 1
fs.algorithmSettings.minNodeSizePrc = 1

# Sets the split evaluation fuction used when the tree is built.
fs.algorithmSettings.splitEvaluation = SplitEvaluation.entropy

# Sets the maximum child node count to 10
fs.algorithmSettings.maxNodeRank = 10

# All created objects have to be saved in the metadataRespository
save('zoo_data', pd)
save('zoo_ld_train', ld)
save('zoo_cl_settings', fs)
save('zoo_building', bt)

# And finally build task is executed
print "Executing build task"
execute('zoo_building')
print "Model 'created_model' created"
print "\nEnd of script"

# Description: queries existing model for rules in leaves

# load the model
model = load('created_model')
# printing rules
print "Rules in leaves:" 
# iterate over all leaves, the current leaf is assigned to the 
# leaf variable
for leaf in model.modelStatistics.allLeaves:

    # print some node statistics
    print "**Node %d, count: %g" % (leaf.nodeID, leaf.count)
    print 'RULE:'
    print leaf.nodeRule
    print 'Predicted value:'
    print leaf.getPredictedValue()
    print '---'

Output:

Executing build task
Model 'created_model' created

End of script
Rules in leaves:
**Node 1, count: 4
RULE:
type IN ('amphibian')
Predicted value:
false
---
**Node 3, count: 16
RULE:
type IN ('bird') AND 
airborne IN ('true')
Predicted value:
false
---
**Node 10, count: 4
RULE:
type IN ('bird') AND 
airborne IN ('false')
Predicted value:
true
---
**Node 11, count: 13
RULE:
type IN ('fish')
Predicted value:
false
---
**Node 14, count: 8
RULE:
type IN ('insect')
Predicted value:
false
---
**Node 15, count: 10
RULE:
type IN ('invertebrate')
Predicted value:
false
---
**Node 18, count: 41
RULE:
type IN ('mammal')
Predicted value:
true
---
**Node 29, count: 5
RULE:
type IN ('reptile')
Predicted value:
false
---

Example 30.2. Pruning selected nodes

 
# Description: modifies an existing model using a script 
#  by pruning selected nodes

# list of node id's to prune
nodePruneList = [ 22, 47 ]

# check if model exists
if not exists('created_model'):
    raise 'please run first example for classification trees module'

# load the model
model = load('created_model')

# iterate over all leaves, the current leaf is assigned to
# leaf variable
for node in model.modelStatistics.allNodes:
    if node.nodeID in nodePruneList:
        node.isLeaf = 1

# save the modified model
save('created_model', model)
                

Figure 30.24. Tree structure before running the example code.

Figure 30.25.  Tree structure after running the example code. Two nodes have been pruned.