def supervised_opf_feature_selection(opytimizer):
# Gathers features
features = opytimizer[:, 0].astype(bool)
# Remaking training and validation subgraphs with selected features
X_train_selected = X_train[:, features]
X_val_selected = X_val[:, features]
# Creates a SupervisedOPF instance
opf = SupervisedOPF(distance='log_squared_euclidean',
pre_computed_distance=None)
# Fits training data into the classifier
opf.fit(X_train_selected, Y_train)
# Predicts new data
preds = opf.predict(X_val_selected)
# Calculates accuracy
acc = g.opf_accuracy(Y_val, preds)
return 1 - acc
class US(object):
def __init__(self, path_output):
self.opfSup = SupervisedOPF(distance='log_squared_euclidean', pre_computed_distance=None)
self.path_output=path_output
def __classify(self, x_train,y_train, x_valid, y_valid, minority_class):
# Training the OPF
indexes = np.arange(len(x_train))
self.opfSup.fit(x_train, y_train,indexes)
# Prediction of the validation samples
y_pred,_ = self.opfSup.predict(x_valid)
y_pred = np.array(y_pred)
# Validation measures for this k nearest neighbors
accuracy = accuracy_score(y_valid, y_pred)
recall = recall_score(y_valid, y_pred, pos_label=minority_class) # assuming that 2 is the minority class
f1 = f1_score(y_valid, y_pred, pos_label=minority_class)
return accuracy, recall, f1, y_pred
def __saveResults(self, X_train,Y_train, X_test, Y_test, ds,f, approach, minority_class):
path = '{}/down_{}/{}/{}'.format(self.path_output,approach,ds,f)
if not os.path.exists(path):
os.makedirs(path)
results_print=[]
accuracy, recall, f1, pred = self.__classify(X_train,Y_train, X_test, Y_test, minority_class)
results_print.append([0,accuracy, recall, f1])
np.savetxt('{}/pred.txt'.format(path), pred, fmt='%d')
np.savetxt('{}/results.txt'.format(path), results_print, fmt='%d,%.5f,%.5f,%.5f')
def __saveDataset(self, X_train,Y_train, pathDataset,ds_name):
DS = np.insert(X_train,len(X_train[0]),Y_train , axis=1)
np.savetxt('{}/train_{}.txt'.format(pathDataset, ds_name),DS, fmt='%.5f,'*(len(X_train[0]))+'%d')
def __computeScore(self, labels, preds, conqs, score):
for i in range(len(labels)):
if labels[i]==preds[i]:
score[conqs[i]]+=1
else:
score[conqs[i]]-=1
def major_negative(self, output, X, Y, X_test, Y_test, path, majority_class, ds, f, minority_class):
#1st case: remove samples from majoritary class with negative scores
output_majority = output[output[:,1]==majority_class]
output_majority_negative = output_majority[output_majority[:,2]<0]
X_train = np.delete(X, output_majority_negative[:,0],0)
Y_train = np.delete(Y, output_majority_negative[:,0])
self.__saveDataset(X_train,Y_train, path,'major_negative')
self.__saveResults(X_train,Y_train, X_test, Y_test, ds,f, 'major_negative', minority_class)
def major_neutral(self, output, X, Y, X_test, Y_test, path, majority_class, ds, f, minority_class):
#2st case: remove samples from majoritary class with negative or zero scores
output_majority = output[output[:,1]==majority_class]
output_majority_neutal = output_majority[output_majority[:,2]<=0]
X_train = np.delete(X, output_majority_neutal[:,0],0)
Y_train = np.delete(Y, output_majority_neutal[:,0])
self.__saveDataset(X_train,Y_train, path,'major_neutral')
self.__saveResults(X_train,Y_train, X_test, Y_test, ds,f, 'major_neutral', minority_class)
def negative(self, output, X, Y, X_test, Y_test, path, majority_class, ds, f, minority_class):
#3st case: remove all samples with negative
output_negatives = output[output[:,2]<0]
X_train = np.delete(X, output_negatives[:,0],0)
Y_train = np.delete(Y, output_negatives[:,0])
self.__saveDataset(X_train,Y_train, path,'negative')
self.__saveResults(X_train,Y_train, X_test, Y_test, ds,f, 'negative', minority_class)
def negatives_major_zero(self, output, X, Y, X_test, Y_test, path, majority_class, ds, f, minority_class):
#4st case: remove samples from majoritary class with negative or zero scores
# and from minoritary class with negative scores
output_negatives = output[output[:,2]<0]
output_negatives_major_zero = output_negatives[output_negatives[:,1]==majority_class]
output_negatives_major_zero = output_negatives_major_zero[output_negatives_major_zero[:,2]<=0]
X_train = np.delete(X, output_negatives_major_zero[:,0],0)
Y_train = np.delete(Y, output_negatives_major_zero[:,0])
self.__saveDataset(X_train,Y_train, path,'negatives_major_zero')
self.__saveResults(X_train,Y_train, X_test, Y_test, ds,f, 'negatives_major_zero', minority_class)
def balance(self, output, X, Y, X_test, Y_test, path, majority_class, ds, f, minority_class):
#5st case: remove samples from majoritary class until balancing the dataset
# find the number of samples to remove
n_samples = len(output)
n_samples_minority = len(output[output[:,1]==2])
n_samples_to_remove = n_samples - (n_samples_minority*2)
# sort samples from majority class by score
output_majority= output[output[:,1]==majority_class]
order = np.argsort(output_majority[:,2])
output_majority_ordered = output_majority[order,:]
# remove samples
output_to_remove = output_majority_ordered[:n_samples_to_remove,:]
X_train = np.delete(X, output_to_remove[:,0],0)
Y_train = np.delete(Y, output_to_remove[:,0])
# save new dataset and results
self.__saveDataset(X_train,Y_train, path,'balance')
self.__saveResults(X_train,Y_train, X_test, Y_test, ds,f, 'balance', minority_class)
def __runOPF(self, X_train,y_train,index_train,X_test,y_test,index_test, score):
# Creates a SupervisedOPF instance
opf = SupervisedOPF(distance='log_squared_euclidean',
pre_computed_distance=None)
# Fits training data into the classifier
opf.fit(X_train, y_train, index_train)
# Predicts new data
preds, conqs = opf.predict(X_test)
self.__computeScore(y_test, preds, conqs, score)
def run(self, X, Y, indices):
# Create stratified k-fold subsets
kfold = 5 # no. of folds
skf = StratifiedKFold(kfold, shuffle=True,random_state=1)
skfind = [None] * kfold # skfind[i][0] -> train indices, skfind[i][1] -> test indices
cnt = 0
for index in skf.split(X, Y):
skfind[cnt] = index
cnt += 1
score = np.zeros((5,len(X)))
for i in range(kfold):
train_indices = skfind[i][0]
test_indices = skfind[i][1]
X_train = X[train_indices]
y_train = Y[train_indices]
index_train = indices[train_indices]
X_test = X[test_indices]
y_test = Y[test_indices]
index_test = indices[test_indices]
self.__runOPF(X_train,y_train,index_train,X_test,y_test,index_test, score[i])
output= np.zeros((len(indices),8))
score_t = np.transpose(score)
output[:,0] =indices
output[:,1] =Y
output[:,2] =np.sum(score_t,axis=1)
output[:,3:] =score_t
return output
txt = l.load_txt('data/boat.txt')
# Parsing a pre-loaded numpy array
X, Y = p.parse_loader(txt)
# Splitting data into training and validation sets
X_train, X_val, Y_train, Y_val = s.split(X, Y, percentage=0.5, random_state=1)
# Creates a always true loop
while True:
# Creates a SupervisedOPF instance
opf = SupervisedOPF(distance='log_squared_euclidean',
pre_computed_distance=None)
# Fits training data into the classifier
opf.fit(X_train, Y_train)
# Predicts new data
preds = opf.predict(X_val)
# Calculating accuracy
acc = g.opf_accuracy(Y_val, preds)
print(f'Accuracy: {acc}')
# Gathers which samples were missclassified
errors = np.argwhere(Y_val != preds)
# If there are no missclassified samples
if len(errors) == 0:
# Breaks the process
grdSearch = GridSearchCV(modelXgbEmpty, {
'max_depth': [2, 4, 8, 10],
'n_estimators': [50, 100, 200, 400]
},
verbose=1,
error_score='accuracy')
grdSearch.fit(X_train, y_train)
grdSearch.best_score_, grdSearch.best_params_
y_pred = grdSearch.predict(X_test)
print(classification_report(y_test, y_pred))
print(accuracy_score(y_test, y_pred))
#Testando Algoritmo OPF (Não é Essemble apenas para demonstração)
modelOPF = SupervisedOPF(distance='manhattan')
# manhattan = 74
# squared_euclidean 72
# log_euclidean 74
# bray_curtis 71
# canberra 71
# log_squared_euclidean 74
# squared_euclidean 72
# gaussian 37
# squared_cord 53
y_train_opf = y_train + 1
y_test_opf = y_test + 1
modelOPF.fit(X_train, y_train_opf)
predsOPF = modelOPF.predict(X_test)
print(accuracy_score(y_test_opf, predsOPF))
