Fix: RMSE fitness is displayed as negative for consistency in the fit…

…ness function; If the SimpleThresholdClassifier has <0.50 accuracy in training, it will invert the prediction labels

stgp/Individual.py

@@ -92,6 +92,13 @@ def fit(self, Tr_x, Tr_y):
 
 			self.model.fit(hyper_X,Tr_y)
 
+			# Binary classification with an extra slot for missing values
+			if len(list(set(Tr_y))) <= 3: 
+				# If the accuracy is < 0.5, invert the labels
+				if self.getAccuracy(Tr_x, Tr_y, pred="Tr") < 0.5:
+					self.trainingClassPredictions = None
+					self.model.invertPredictions()
+
 
 	def getSize(self):
 		'''
@@ -139,7 +146,7 @@ def getFitness(self):
 			if self.fitnessType == "RMSE":
 				self.getTrainingValuePredictions()
 				waf = mean_squared_error(self.trainingValuePredictions, self.convertLabelsToInt(self.training_Y))**0.5
-				self.fitness = waf 
+				self.fitness = -waf 
 
 		return self.fitness
 
@@ -194,7 +201,7 @@ def getRMSE(self, X, Y,pred=None):
 		elif pred == "Te":
 			pred = self.getTestValuePredictions(X)
 		else:
-			pred = self.predict(X)
+			pred = self.predict(X, classOutput = False)
 
 		return mean_squared_error(pred, Y)**0.5
 

stgp/Population.py

@@ -129,9 +129,9 @@ def nextGeneration(self):
 				model = pool.map(fitIndividuals, [(ind, self.Tr_x, self.Tr_y) for ind in self.population] )
 				for i in range(len(self.population)):
 					self.population[i].model = model[i][0].model
+					self.population[i].fitness = model[i][1]
 					self.population[i].labelToInt = model[i][0].labelToInt
 					self.population[i].intToLabel = model[i][0].intToLabel
-					self.population[i].trainingPredictions = model[i][1]
 					self.population[i].training_X = self.Tr_x
 					self.population[i].training_Y = self.Tr_y
 		else:
@@ -201,7 +201,7 @@ def fitIndividuals(a):
 	ind,x,y = a
 	ind.fit(x,y)
 
-	return ( ind, ind.predict(x) )
+	return ( ind, ind.getFitness() )
 
 def getTrainingPredictions(ind):
 	return ind.getTrainingPredictions()

stgp/SimpleThresholdClassifier.py

@@ -11,6 +11,7 @@
 class SimpleThresholdClassifier:
 
 	threshold = None
+	isPositive1 = True
 
 	def __init__(self, threshold = 0):
 		self.threshold = threshold
@@ -24,5 +25,11 @@ def predict(self, X):
 		Receives X, a 1-D array of real values
 		Return a list of predictions based on the value
 		"""	
-		predictions = [ 1 if value > self.threshold else 0 for value in X]
-		return predictions
+		if self.isPositive1:
+			predictions = [ 1 if value > self.threshold else 0 for value in X]
+		else:
+			predictions = [ 1 if value < self.threshold else 0 for value in X]
+		return predictions
+
+	def invertPredictions(self):
+		self.isPositive1 = not self.isPositive1