def Validate(self, subset):
# Ensure that given the subset is valid.
for feat in subset:
if (0 > feat) or (feat > self.classifier.num_features):
print("ERROR: Subset does not match features.")
return 0
# Make copy of the dataset for feature selection.
# temp_normalized = self.classifier.normalized.copy()
to_delete = []
# Trim down to given subset.
if subset != [0]:
for col in range(self.classifier.num_features + 1)[1:]:
if col not in subset:
to_delete.append(col)
# K-fold vars.
fold_list = []
folded = []
num_correct = 0
print("\n--------------------------------------")
print("\nPerforming k-fold cross validation.\n")
# Perform k-fold cross validation.
for leave_out in range(self.classifier.num_points):
# Prepare temporary datasets for k-folding.
fold_list = self.classifier.normalized.copy()
folded = self.classifier.normalized[leave_out].copy()
del fold_list[leave_out]
# Remove the columns not specified by the subset.
fold_list = np.delete(fold_list, to_delete, axis=1)
folded = np.delete(folded, to_delete)
# Create classifier for folded data.
temp_c = Classifier()
temp_c.Train(fold_list)
result = temp_c.Test(folded)
# Increment if classification accurate.
if result == folded[0]:
num_correct += 1
return num_correct / self.classifier.num_points
def intro():
print("\n--------------------------------------")
print("CS170 Project 02 / Part 02\n")
print("Otniel Thehumury 862029595")
print("Matthew Walsh 862088280")
print("--------------------------------------")
# Collect the number of features from the user.
# num_features = input("Please enter total number of features: ")
# print("\nType the number of the algorithm you want to run.")
# print("\n1) Forward Selection")
# print("\n2) Backward Elimination")
# print("\n3) NN Classifier")
# choice = input("\nHere: ")
# choice = int(choice)
# Project 2, Part 2:
choice = 3
print("")
p = Problem()
if choice == 1:
print("Chosen Forward Elimination")
# p.forward_selection(num_features)
elif choice == 2:
print("Chosen Backward Elimination")
# p.backward_elimination(num_features)
elif choice == 3:
print("Nearest Neighbor Classifier:\n")
training_data = read_dataset()
print("Please select a feature subset.\n")
print("For example, type: 1 3 5 7 9")
print("Enter zero or blank for all features.\n")
sub_input = input("Enter your choice here: ")
# Separate input into int list.
subset = []
for feat in sub_input.split():
subset.append(int(feat))
# Zero indicates all features.
if sub_input == "":
subset = [0]
# Train the classifier.
c = Classifier()
c.Train(training_data)
# c.Test([2.0, 2.1530859, 4.4095784, 3.6216757, 3.8451064, 2.9807186, 2.0171732, 0.5397355, 3.3933456, 2.2950856, 3.0431002])
# Implement validator.
v = Validator(c)
print(v.Validate(subset))
else:
print("Incorrect choice - closing program.")
Q.queryRanker()
if arguments.run == 'cluster' or arguments.run == 'clusterclassify' or arguments.run == 'all':
print "-- Clustering --"
bestRankersFile = 'QueryData/' + dataset + str(iterations) + '.data'
KM = KMeans(arguments.fromrangek, arguments.torangek, bestRankersFile,
dataset, iterations)
(queryToCluster, clusterToRanker) = KM.runScript()
#print 'queryToCluster', queryToCluster
if arguments.run == 'classify' or arguments.run == 'clusterclassify' or arguments.run == 'all':
print "-- Classification --"
clusterPath = "ClusterData/" + dataset + str(iterations) + ".data"
rankerPath = "QueryData/generalRanker.data"
C = Classifier(clusterPath, path_train, rankerPath, iterations)
C.Train()
if arguments.run == 'compare':
print "-- Comparison --"
classifierPath = "Classifier/" + dataset + str(iterations) + ".data"
basic_ranker_path = "QueryData/generalRanker.data"
clusterPath = "ClusterData/" + dataset + str(iterations) + ".data"
compare.compareSystems(path_validate, classifierPath, basic_ranker_path,
clusterPath, click)
if arguments.run == "fake":
print "--Fake clustering and learning--"
rankerPath = "QueryData/generalRanker.data"
bestRankersFile = 'QueryData/' + dataset + str(iterations) + '.data'
clusterPath = "ClusterData/" + dataset + str(iterations) + ".data"
rankerPath = "QueryData/generalRanker.data"
