def train_classifier(X_train, X_test, y_train, y_test, sa_index, p_Group,
dataset, mutex, mode, base_learners):
if mode == 0:
classifier = AdaFairSP(n_estimators=base_learners,
saIndex=sa_index,
saValue=p_Group,
cumul=False,
CSB="CSB1")
elif mode == 1:
classifier = AdaFairSP(n_estimators=base_learners,
saIndex=sa_index,
saValue=p_Group,
CSB="CSB1")
classifier.fit(X_train, y_train)
y_pred_labels = classifier.predict(X_test)
mutex.acquire()
infile = open(dataset + "_sp_acc", 'rb')
dict_to_ram = pickle.load(infile)
infile.close()
dict_to_ram.performance.append(
calculate_performance_SP(X_test, y_test, y_pred_labels, sa_index,
p_Group))
outfile = open(dataset + "_sp_acc", 'wb')
pickle.dump(dict_to_ram, outfile)
outfile.close()
mutex.release()
def train_classifier(X_train, X_test, y_train, y_test, sa_index, p_Group,
dataset, mutex, mode, base_learners, c):
if mode == 0:
classifier = AdaCostClassifier(saIndex=sa_index,
saValue=p_Group,
n_estimators=base_learners,
CSB="CSB1")
elif mode == 1:
classifier = AdaFairSP(n_estimators=base_learners,
saIndex=sa_index,
saValue=p_Group,
CSB="CSB2",
c=c)
elif mode == 2:
classifier = SMOTEBoost(n_estimators=base_learners,
saIndex=sa_index,
n_samples=10,
saValue=p_Group,
CSB="CSB1")
else:
train = []
for i in range(len(X_train)):
train.append((X_train[i], y_train[i]))
classifier = boostLearner(train, sa_index, p_Group)
if mode in [0, 1, 2]:
classifier.fit(X_train, y_train)
y_pred_probs = classifier.predict_proba(X_test)[:, 1]
y_pred_labels = classifier.predict(X_test)
else:
y_pred_labels = [classifier(x) for x in X_test]
y_pred_probs = [0 for i in y_pred_labels]
mutex.acquire()
infile = open(dataset, 'rb')
dict_to_ram = pickle.load(infile)
infile.close()
dict_to_ram.performance.append(
calculate_performance_SP(X_test, y_test, y_pred_labels, y_pred_probs,
sa_index, p_Group))
outfile = open(dataset, 'wb')
pickle.dump(dict_to_ram, outfile)
outfile.close()
mutex.release()
def train_classifier(X_train, X_test, y_train, y_test, sa_index, p_Group, dataset, mutex, mode, base_learners, c, dataset_name):
if mode == 0:
classifier = AdaCostClassifier(saIndex=sa_index, saValue=p_Group, n_estimators=base_learners, CSB="CSB1")
elif mode == 1:
classifier = AdaFairSP(n_estimators=base_learners, saIndex=sa_index, saValue=p_Group, CSB="CSB1", c=c)
elif mode == 2:
if dataset_name == 'adult-gender' or dataset == 'bank':
samples = 100
elif dataset_name == 'compass-gender':
samples = 2
else:
samples = 500
classifier = SMOTEBoost(n_estimators=base_learners,saIndex=sa_index, n_samples=samples, saValue=p_Group, CSB="CSB1" )
else:
classifier=marginAnalyzer(X_train,y_train,sa_index=sa_index,p_Group=p_Group,numRounds=200)
if mode in [0,1,2]:
classifier.fit(X_train, y_train)
# y_pred_probs = classifier.predict_proba(X_test)[:, 1]
y_pred_labels = classifier.predict(X_test)
else:
y_pred_labels=classifier.pred(X_test)
# y_pred_probs=classifier.margin(classifier.clf,X_test)
# y_pred_probs=[abs(prob) for prob in y_pred_probs]
mutex.acquire()
infile = open(dataset, 'rb')
dict_to_ram = pickle.load(infile)
infile.close()
dict_to_ram.performance.append(
calculate_performance_SP(X_test, y_test, y_pred_labels, sa_index, p_Group))
outfile = open(dataset, 'wb')
pickle.dump(dict_to_ram, outfile)
outfile.close()
mutex.release()
def predict(clf, X_test, y_test, sa_index, p_Group):
y_pred_probs = clf.predict_proba(X_test)[:, 1]
y_pred_labels = clf.predict(X_test)
return calculate_performance_SP(X_test, y_test, y_pred_labels,
y_pred_probs, sa_index, p_Group)