def get_best_kappa4( self, NE, yPredR ):
#This function returns the thresholds that produce the highest kappa
#when the score can vary from 0 to 4
bestKappa, bthreshold1, bthreshold2, bthreshold3, bthreshold4 = 0, 0, 0, 0, 0
yPred = yPredR
for thresholdV1 in range( 1, 2 * NE ):
for thresholdV2 in range( thresholdV1 + 1 , 2 * NE):
for thresholdV3 in range( thresholdV2 + 1 , 3 * NE):
for thresholdV4 in range( thresholdV3 + 1 , 4 * NE):
threshold1 = thresholdV1 / ( NE * 1.0 )
threshold2 = thresholdV2 / ( NE * 1.0 )
threshold3 = thresholdV3 / ( NE * 1.0 )
threshold4 = thresholdV4 / ( NE * 1.0 )
if threshold1 < np.min( yPredR ):
continue
if min( [ threshold2, threshold3, threshold4 ] ) > np.max( yPredR ) :
continue
yPred = ( yPredR >= threshold1 ) * 1
yPred = ( yPredR >= threshold2 ) * 1 + yPred
yPred = ( yPredR >= threshold3 ) * 1 + yPred
yPred = ( yPredR >= threshold4 ) * 1 + yPred
cKappa = skllMetrics.kappa( self.y, yPred )
if cKappa > bestKappa:
bestKappa, bthreshold1, bthreshold2, bthreshold3, bthreshold4 = cKappa, threshold1, threshold2, threshold3, threshold4
return bestKappa, [ bthreshold1, bthreshold2, bthreshold3, bthreshold4 ]
def get_best_kappa4(self, NE, yPredR):
#This function returns the thresholds that produce the highest kappa
#when the score can vary from 0 to 4
bestKappa, bthreshold1, bthreshold2, bthreshold3, bthreshold4 = 0, 0, 0, 0, 0
yPred = yPredR
for thresholdV1 in range(1, 2 * NE):
for thresholdV2 in range(thresholdV1 + 1, 2 * NE):
for thresholdV3 in range(thresholdV2 + 1, 3 * NE):
for thresholdV4 in range(thresholdV3 + 1, 4 * NE):
threshold1 = thresholdV1 / (NE * 1.0)
threshold2 = thresholdV2 / (NE * 1.0)
threshold3 = thresholdV3 / (NE * 1.0)
threshold4 = thresholdV4 / (NE * 1.0)
if threshold1 < np.min(yPredR):
continue
if min([threshold2, threshold3, threshold4
]) > np.max(yPredR):
continue
yPred = (yPredR >= threshold1) * 1
yPred = (yPredR >= threshold2) * 1 + yPred
yPred = (yPredR >= threshold3) * 1 + yPred
yPred = (yPredR >= threshold4) * 1 + yPred
cKappa = skllMetrics.kappa(self.y, yPred)
if cKappa > bestKappa:
bestKappa, bthreshold1, bthreshold2, bthreshold3, bthreshold4 = cKappa, threshold1, threshold2, threshold3, threshold4
return bestKappa, [bthreshold1, bthreshold2, bthreshold3, bthreshold4]
def evaluateCl1(self, model, SVM=False):
#This function evaluates Logistic Regression or Support Vector Machine Model when the possible
#scores may be 0 or 1. It returns the threshold and current parameters that produce the highest kappa.
yProb = self.cv_estimateCl1(model, SVM)
bestKappa, bthreshold = 0, 0.5
for div in range(1, 100):
threshold = div / 100.0
yPred = yProb > threshold
ckappa = skllMetrics.kappa(self.y, yPred)
if ckappa > bestKappa:
bestKappa = ckappa
bthreshold = threshold
return bestKappa, model.param1, model.param2, bthreshold
def evaluateCl1( self, model , SVM = False):
#This function evaluates Logistic Regression or Support Vector Machine Model when the possible
#scores may be 0 or 1. It returns the threshold and current parameters that produce the highest kappa.
yProb = self.cv_estimateCl1( model, SVM )
bestKappa, bthreshold = 0, 0.5
for div in range(1,100):
threshold = div / 100.0
yPred = yProb > threshold
ckappa = skllMetrics.kappa( self.y, yPred )
if ckappa > bestKappa:
bestKappa = ckappa
bthreshold = threshold
return bestKappa, model.param1, model.param2, bthreshold
def evaluateGB1(self, model, nTree, fixed = False):
#This function evaluates the Gradient Boosting Machine Model when the possible
#scores may be 0 or 1. It returns the tree and threshold that produce the highest kappa.
yProbL = self.cv_estimateGB1( model , nTree)
bestKappa, bthreshold, bTree = 0, 0, 0
NE = 20
treeCandidates = [nTree-1] if fixed else [ 20, 30, 50, 100, 150, 200, 250, 300, 350 , 400, 450, 500 , 550, 600, 750, 1000, 1250, 1500 ]
for j in treeCandidates:
cTree= j - 1
yProb = yProbL[cTree]
for div in range(1,20):
threshold = div / 100.0
yPred = yProb > threshold
ckappa = skllMetrics.kappa( self.y, yPred )
if ckappa > bestKappa:
bestKappa, bthreshold, bTree = ckappa, threshold, cTree
return bestKappa, bTree, model.param2, bthreshold
def get_best_kappa2(self, NE, yPredR):
#This function returns the thresholds that produce the highest kappa
#when the score can vary from 0 to 2
bestKappa, bthreshold1, bthreshold2 = 0, 0, 0
yPred = yPredR
for thresholdV1 in range(1, 2 * NE):
for thresholdV2 in range(thresholdV1 + 1, 2 * NE):
threshold1 = thresholdV1 / (NE * 1.0)
threshold2 = thresholdV2 / (NE * 1.0)
if threshold1 < np.min(yPredR):
continue
if threshold2 > np.max(yPredR):
continue
yPred = (yPredR >= threshold1) * 1
yPred = (yPredR >= threshold2) * 1 + yPred
cKappa = skllMetrics.kappa(self.y, yPred)
if cKappa > bestKappa:
bestKappa, bthreshold1, bthreshold2 = cKappa, threshold1, threshold2
return bestKappa, bthreshold1, bthreshold2
def get_best_kappa2( self, NE, yPredR ):
#This function returns the thresholds that produce the highest kappa
#when the score can vary from 0 to 2
bestKappa, bthreshold1, bthreshold2 = 0, 0, 0
yPred = yPredR
for thresholdV1 in range( 1, 2 * NE ):
for thresholdV2 in range( thresholdV1 + 1 , 2 * NE):
threshold1 = thresholdV1 / ( NE * 1.0 )
threshold2 = thresholdV2 / ( NE * 1.0 )
if threshold1 < np.min( yPredR ):
continue
if threshold2 > np.max( yPredR ):
continue
yPred = ( yPredR >= threshold1 ) * 1
yPred = ( yPredR >= threshold2 ) * 1 + yPred
cKappa = skllMetrics.kappa( self.y, yPred )
if cKappa > bestKappa:
bestKappa, bthreshold1, bthreshold2 = cKappa, threshold1, threshold2
return bestKappa, bthreshold1, bthreshold2
def evaluateCl2(self, model, SVM=False):
#This function evaluates Logistic Regression or Support Vector Machine Model when the possible
#scores may vary from 0 to 2. It returns the threshold and current parameters that produce the highest kappa.
NE = 20
if SVM:
bestKappa, bthreshold1, bthreshold2 = self.evaluateSV2(model)
return bestKappa, model.param1, model.param2, bthreshold1, bthreshold2
yProb1, yProb2 = self.cv_estimateCl2(model, SVM)
bestKappa, bthreshold1, bthreshold2 = 0, 0.5, 0.5
for thresholdV1 in range(1, NE):
for thresholdV2 in range(1, NE):
threshold1 = thresholdV1 / (NE * 1.0)
threshold2 = thresholdV2 / (NE * 1.0)
yPred1 = (yProb1 > threshold1) * 1
yPred2 = (yProb2 > threshold2) * 2
yPred = np.maximum(yPred1, yPred2)
ckappa = skllMetrics.kappa(self.y, yPred)
if ckappa > bestKappa:
bestKappa, bthreshold1, bthreshold2 = ckappa, threshold1, threshold2
return bestKappa, model.param1, model.param2, bthreshold1, bthreshold2
def evaluateCl2( self, model , SVM = False):
#This function evaluates Logistic Regression or Support Vector Machine Model when the possible
#scores may vary from 0 to 2. It returns the threshold and current parameters that produce the highest kappa.
NE = 20
if SVM:
bestKappa, bthreshold1, bthreshold2 = self.evaluateSV2( model )
return bestKappa, model.param1, model.param2, bthreshold1, bthreshold2
yProb1, yProb2 = self.cv_estimateCl2( model , SVM)
bestKappa, bthreshold1, bthreshold2 = 0, 0.5, 0.5
for thresholdV1 in range(1,NE):
for thresholdV2 in range(1,NE):
threshold1 = thresholdV1 / ( NE * 1.0 )
threshold2 = thresholdV2 / ( NE * 1.0 )
yPred1 = ( yProb1 > threshold1 ) * 1
yPred2 = ( yProb2 > threshold2 ) * 2
yPred = np.maximum( yPred1, yPred2)
ckappa = skllMetrics.kappa( self.y, yPred )
if ckappa > bestKappa:
bestKappa, bthreshold1, bthreshold2 = ckappa, threshold1, threshold2
return bestKappa, model.param1, model.param2, bthreshold1, bthreshold2
def evaluateGB1(self, model, nTree, fixed=False):
#This function evaluates the Gradient Boosting Machine Model when the possible
#scores may be 0 or 1. It returns the tree and threshold that produce the highest kappa.
yProbL = self.cv_estimateGB1(model, nTree)
bestKappa, bthreshold, bTree = 0, 0, 0
NE = 20
treeCandidates = [nTree - 1] if fixed else [
20, 30, 50, 100, 150, 200, 250, 300, 350, 400, 450, 500, 550, 600,
750, 1000, 1250, 1500
]
for j in treeCandidates:
cTree = j - 1
yProb = yProbL[cTree]
for div in range(1, 20):
threshold = div / 100.0
yPred = yProb > threshold
ckappa = skllMetrics.kappa(self.y, yPred)
if ckappa > bestKappa:
bestKappa, bthreshold, bTree = ckappa, threshold, cTree
return bestKappa, bTree, model.param2, bthreshold
