Example 29.1. Bivariate Probit

if not tableExists('biv_credit'):
 raise "Table 'biv_credit' does not exists. Please run Appendix Example for Bivariate"

pd = PhysicalData('biv_credit')
save('biv_pd', pd)

ld  = LogicalData(pd)
save('biv_ld',ld)

fs = BivariateFunctionSettings()
fs.setLogicalData(ld)


target = ['good','Accepted_los']

active1 = ['checking','duration']
active2 = ['history','purpose','savings']
active = ['employed']


as = BivariateProbitSettings()

vss = VariableSelectionSettings()
vss.variableSelectionMethod = VariableSelectionMethod.forward
vss.modelEntryLevel = 0.05

as.logisticPreselection = TRUE
as.randomization = TRUE

opt = OptimizationAlgorithmSettings()
opt.convergenceCriterion = ConvergenceCriterion.likelihood
opt.convergenceThreshold = 0.01
opt.iterMax = 20

fs.set1stTargetPositiveCategory('true')
fs.set2ndTargetPositiveCategory('true')

as.optimizationAlgorithmSettings = opt
as.variableSelectionSettings = vss
fs.algorithmSettings = as


us = fs.getAttributeUsageSet()
usages = us.getAttributes()
for i in range(0,usages.size()):
    usage = usages.get(i)
    usage.setUsage(UsageOption.inactive)
for attr in active:
    us.getAttribute(attr).setUsage(UsageOption.active)
for attr in active1:
    us.getAttribute(attr).setUsage(BivariateUsageOption.active1)
for attr in active2:
    us.getAttribute(attr).setUsage(BivariateUsageOption.active2)

us.getAttribute(target[0]).setUsage(UsageOption.target)
us.getAttribute(target[1]).setUsage(BivariateUsageOption.target2)

save('biv_fs', fs)

bt = MiningBuildTask('biv_pd', 'biv_fs', None)

def buildSequentially(fromSeed, tillSeed):
    for i in range(fromSeed,tillSeed):
        if i > 0:
            bt.inputModelName = 'seed%i' % (i-1)
        bt.modelName = 'seed%i' % i
        save('biv_bt', bt)
        execute('biv_bt')
        model = load('seed%i' % i)
        lf = model.getLikelihoodFunction()
        beta = lf.getSignatureBeta()
        print "\nbivariate seed model %i beta:" % i
        print "correlation (RO) = %.4f"% beta[2]

buildSequentially(0,4)

Output:

bivariate seed model 0 beta:
correlation (RO) = -0.9059

bivariate seed model 1 beta:
correlation (RO) = -0.9059

bivariate seed model 2 beta:
correlation (RO) = -0.9059

bivariate seed model 3 beta:
correlation (RO) = -0.9059