def compareRoc(self, selectedResults=None, append=''):
if selectedResults:
results = {
k: self.models[k].testPrediction()
for k in selectedResults if k in self.models
}
else:
results = {
k: self.models[k].testPrediction()
for k in self.models.keys()
}
#As the splitting is done with the same pseudorandom initialisation the test set is always the same
results['truth'] = self.models.values()[0].data.y_test
rocCurve(results, output=self.output, append=append)
pass
def rocCurve(self,doEvalSet=False):
if self.doRegression:
print 'Cannot make ROC curve for a regression problem, skipping'
return None
if doEvalSet:
rocCurve(self.model.predict(self.data.X_eval.as_matrix()), self.data.y_eval,self.output,append='_eval')
rocCurve(self.model.predict(self.data.X_train.as_matrix()),self.data.y_train,self.output,append='_dev')
else:
rocCurve(self.model.predict(self.data.X_test.as_matrix()), self.data.y_test,self.output)
rocCurve(self.model.predict(self.data.X_train.as_matrix()),self.data.y_train,self.output,append='_train')
def rocCurve(self, doEvalSet=False):
if doEvalSet: #produce report for dev and eval sets instead
rocCurve(self.bdt.decision_function(self.data.X_eval),
self.data.y_eval,
output=self.output,
append='_eval')
rocCurve(self.bdt.decision_function(self.data.X_dev),
self.data.y_dev,
output=self.output,
append='_dev')
else:
rocCurve(self.bdt.decision_function(self.data.X_test),
self.data.y_test,
output=self.output)
rocCurve(self.bdt.decision_function(self.data.X_train),
self.data.y_train,
output=self.output,
append='_train')