def trainNN(self, classData, dialog):
maxIter = 250
nHidden = 10
seed = np.random.randint(0, 1000000)
trainAUC = np.zeros((self.nFold, maxIter+1))
validAUC = np.zeros((self.nFold, maxIter+1))
partitionGenerator = ml.part.classRandomSubSample(classData,
self.trainFrac, self.nFold)
for fold, trainData, validData in partitionGenerator:
dialog.Update(fold, 'Validation Fold: %d' % fold)
def validTraceCB(optable, iteration, paramTrace, errorTrace, success=True):
if success:
trainAUC[fold,validTraceCB.it:] = optable.auc(trainData)
validAUC[fold,validTraceCB.it:] = optable.auc(validData)
validTraceCB.it += 1
validTraceCB.it = 0
s = np.random.get_state()
np.random.seed(seed)
classifier = ml.FNS(trainData, accuracy=0.0, precision=1.0e-10,
nHidden=nHidden, maxIter=maxIter, optimFunc=ml.optim.scg,
callback=validTraceCB, eTrace=True, verbose=False)
np.random.set_state(s)
dialog.Update(self.nFold, 'Training Final Classifier')
meanValidAUC = np.mean(validAUC, axis=0)
bestIter = np.argmax(meanValidAUC)
bestMeanValidAUC = meanValidAUC[bestIter]
s = np.random.get_state()
np.random.seed(seed)
self.classifier = ml.FNS(classData, accuracy=0.0, precision=1.0e-10,
nHidden=nHidden, maxIter=bestIter, optimFunc=ml.optim.scg,
eTrace=False, verbose=False)
np.random.set_state(s)
finalAUC = self.classifier.auc(classData)
dialog.Destroy()
wx.MessageBox(message=('Best Iteration: %f\n' % bestIter) +
('Mean Validation AUC: %f\n' % bestMeanValidAUC) +
('Final Training AUC: %f' % finalAUC),
caption='Training Completed!', style=wx.OK | wx.ICON_INFORMATION)
def trainWelchNN(self, trainData):
maxIter = 250
nHidden = 10
seed = np.random.randint(0, 1000000)
self.stand = ml.ClassStandardizer(trainData)
trainDataStd = self.stand.apply(trainData)
nFold = self.nTrainTrial
trnCA = np.zeros((nFold, maxIter + 1))
valCA = np.zeros((nFold, maxIter + 1))
dialog = wx.ProgressDialog('Training Classifier',
'Featurizing',
maximum=nFold + 1,
style=wx.PD_ELAPSED_TIME | wx.PD_SMOOTH)
for fold, trnData, valData in ml.part.classStratified(trainDataStd,
nFold=nFold):
dialog.Update(fold, 'Validation Fold: %d' % fold)
stand = ml.ClassStandardizer(trnData)
trnData = stand.apply(trnData)
valData = stand.apply(valData)
def valTraceCB(optable,
iteration,
paramTrace,
errorTrace,
success=True):
if success:
trnCA[fold, valTraceCB.it] = optable.ca(trnData)
valCA[fold, valTraceCB.it] = optable.ca(valData)
valTraceCB.it += 1
valTraceCB.it = 0
np.random.seed(seed)
classifier = ml.FNS(trnData,
accuracy=0.0,
precision=0.0,
nHidden=nHidden,
maxIter=maxIter,
optimFunc=ml.optim.scg,
callback=valTraceCB,
eTrace=True,
verbose=False)
dialog.Update(nFold, 'Training Final Classifier')
meanTrnCA = np.mean(trnCA, axis=0)
meanValCA = np.mean(valCA, axis=0)
bestIter = np.argmax(meanValCA)
bestMeanTrnCA = meanTrnCA[bestIter]
bestMeanValCA = meanValCA[bestIter]
np.random.seed(seed)
self.classifier = ml.FNS(trainDataStd,
accuracy=0.0,
precision=0.0,
nHidden=nHidden,
maxIter=bestIter,
optimFunc=ml.optim.scg,
eTrace=False,
verbose=False)
trainCA = self.classifier.ca(trainDataStd)
trainConfusion = np.round(100 *
self.classifier.confusion(trainDataStd))
dialog.Destroy()
resultText = (('Best Num Iterations: %f\n' % bestIter) +
('Best Mean Training CA: %f\n' % bestMeanTrnCA) +
('Best Mean Validation CA: %f\n' % bestMeanValCA) +
('Final Training CA: %f\n' % trainCA) +
('Confusion Matrix:\n' + str(trainConfusion) + '\n') +
('Choices: ' + str(self.choices)))
wx.MessageBox(message=resultText,
caption='Training Completed!',
style=wx.OK | wx.ICON_INFORMATION)
self.saveResultText(resultText)
def trainNN(self, trainData, dialog):
maxIter = 250
nHidden = 10
seed = np.random.randint(0, 1000000)
trnCA = np.zeros((self.nFold, maxIter + 1))
valCA = np.zeros((self.nFold, maxIter + 1))
for fold, trnData, valData in ml.part.classStratified(
trainData, nFold=self.nFold):
dialog.Update(fold, 'Validation Fold: %d' % fold)
stand = ml.ClassStandardizer(trnData)
trnData = stand.apply(trnData)
valData = stand.apply(valData)
def valTraceCB(optable,
iteration,
paramTrace,
errorTrace,
success=True):
if success:
trnCA[fold, valTraceCB.it] = optable.ca(trnData)
valCA[fold, valTraceCB.it] = optable.ca(valData)
valTraceCB.it += 1
valTraceCB.it = 0
np.random.seed(seed)
classifier = ml.FNS(trnData,
accuracy=0.0,
precision=0.0,
nHidden=nHidden,
maxIter=maxIter,
optimFunc=ml.optim.scg,
callback=valTraceCB,
eTrace=True,
verbose=False)
dialog.Update(self.nFold, 'Training Final Classifier')
meanValCA = np.mean(valCA, axis=0)
bestIter = np.argmax(meanValCA)
bestMeanValCA = meanValCA[bestIter]
self.stand = ml.ClassStandardizer(trainData)
trainData = self.stand.apply(trainData)
np.random.seed(seed)
self.classifier = ml.FNS(trainData,
accuracy=0.0,
precision=0.0,
nHidden=nHidden,
maxIter=bestIter,
optimFunc=ml.optim.scg,
eTrace=False,
verbose=False)
trainCA = self.classifier.ca(trainData)
trainConfusion = self.classifier.confusion(trainData)
dialog.Destroy()
wx.MessageBox(message=('Best Iteration: %f\n' % bestIter) +
('Mean Validation CA: %f\n' % bestMeanValCA) +
('Final Training CA: %f' % trainCA),
caption='Training Completed!',
style=wx.OK | wx.ICON_INFORMATION)