def checkPerformance(self, valSet, trainSet, k):
result = 0
for i in range(len(valSet)):
# perform knn
predicted = Knn.knn(trainSet, valSet[i], k, True)
# check if predicted class and actual class match
result += self.checkPrediction(predicted, valSet[i][-1])
# return percent correctly predicted
return result / len(valSet)
def checkPerformance(self, valSet, trainSet, k):
result = 0
for i in range(len(valSet)):
# perform knn
predicted = Knn.knn(trainSet, valSet[i], k, True)
# check if predicted class and actual class match
result += self.checkPrediction(predicted, valSet[i][-1])
# return percent correctly predicted
return result / len(valSet)
# testing data
#testClass = np.array([[1, 2, 1], [2, 3, 0], [7, 8, 1], [4, 5, 0], [6, 6, 1],
# [6, 5, 1], [0, 1, 0], [1, 1, 0], [2, 2, 0], [3, 7, 1],
# [3, 1, 1], [9, 7, 1], [1, 5, 0], [7, 2, 1], [5, 5, 0]])
#editednn = EditedNN()
#print(editednn.eknn(testClass, 3))
def eknn(self, trainSet, k):
# separate into training and validation sets
ennSet, val = train_test_split(trainSet, test_size=0.2, random_state=0)
# repeat until performance stops improving on validation set
prevPerformance = 0.0
perfImprove = True
loopCount = len(ennSet)
while (perfImprove):
# copy previous enn set
prevTrain = ennSet.copy()
# loop through training set
i = 0
while (i < loopCount):
# remove ith point from training set b/c it will be test point for knn
knnTestPoint = ennSet[i]
tempTrain = np.delete(ennSet, i, 0)
# perform knn
predicted = Knn.knn(tempTrain, knnTestPoint, k, True)
# check if predicted class and actual class match
result = self.checkPrediction(predicted, knnTestPoint[-1])
# if not equal then keep set with removed point
if result == 0:
ennSet = tempTrain
loopCount -= 1
else:
i += 1
# check performance and only continue if not degrading
curPerformance = self.checkPerformance(val, tempTrain, k)
if curPerformance > prevPerformance:
prevPerformance = curPerformance
else:
ennSet = prevTrain
perfImprove = False
pd.DataFrame(ennSet).to_csv("reduced_datasets/enn.csv",
header=None,
index=None)
# return edited-nn set
return ennSet
