def main_lr(dataset, range_th):
data = np.load('Datasets/' + dataset + '.npy')
np.random.shuffle(data)
num_samples = data.shape[0]
data_x = data[:num_samples, :-1]
data_y = data[:num_samples, -1]
accs = []
ths = []
for th in range_th:
step = data_x.shape[0] // k_fold
num_of_class = int(np.max(data_y) + 1)
acc = 0.
for i in range(k_fold):
test_x, test_y = data_x[i:i + step], data_y[i:i + step]
train_x, train_y = np.concatenate((data_x[0:i], \
data_x[i+step:]), axis=0), \
myutility.convertToOneHot(np.concatenate((data_y[0:i], \
data_y[i+step:]), axis=0), num_of_class)
LR = logistic.logistic_regression(train_x.shape[1],
train_y.shape[1],
thres=th / 1000)
LR.fit(train_x, train_y)
predict_y = LR.predict(test_x)
acc += np.sum(predict_y == test_y) / test_y.shape[0]
accs.append(acc / k_fold)
ths.append(th / 1000)
return (accs)
def main_lr(dataset):
data = np.load('Datasets/' + dataset + '.npy')
np.random.shuffle(data)
num_samples = data.shape[0]
size_of_dataset = []
acclr_model = []
acclr_sk = []
for m in range(num_samples):
if ((m % 10 == 0) & (m != 0)):
data_x = data[:m, :-1]
data_y = data[:m, -1]
step = data_x.shape[0] // k_fold
num_of_class = int(np.max(data_y) + 1)
acc = 0.
for i in range(k_fold):
test_x, test_y = data_x[i:i + step], data_y[i:i + step]
train_x, train_y = np.concatenate((data_x[0:i], \
data_x[i+step:]), axis=0), \
myutility.convertToOneHot(np.concatenate((data_y[0:i], \
data_y[i+step:]), axis=0), num_of_class)
LR = logistic.logistic_regression(train_x.shape[1],
train_y.shape[1])
LR.fit(train_x, train_y)
predict_y = LR.predict(test_x)
acc += np.sum(predict_y == test_y) / test_y.shape[0]
acclr_model.append(acc / k_fold)
# print(acc/k_fold)
# acc = 0.
# for i in range(k_fold):
# test_x, test_y = data_x[i:i+step], data_y[i:i+step]
# train_x, train_y = np.concatenate((data_x[0:i], \
# data_x[i+step:]), axis=0), \
# np.concatenate((data_y[0:i], data_y[i+step:]), axis=0)
# LR2 = LogisticRegression(solver='lbfgs', multi_class='auto')
# LR2.fit(train_x, train_y)
# predict_y = LR2.predict(test_x)
# acc += np.sum(predict_y == test_y) / test_y.shape[0]
# # print(acc/k_fold)
# acclr_sk.append(acc/k_fold)
size_of_dataset.append(m)
return acclr_model, size_of_dataset
def main_lr(dataset):
data = np.load('Datasets/' + dataset + '.npy')
np.random.shuffle(data)
num_samples = data.shape[0]
data_x = data[:num_samples, :-1]
data_y = data[:num_samples, -1]
step = data_x.shape[0] // k_fold
num_of_class = int(np.max(data_y) + 1)
acc = 0.
for i in range(k_fold):
test_x, test_y = data_x[i:i + step], data_y[i:i + step]
train_x, train_y = np.concatenate((data_x[0:i], \
data_x[i+step:]), axis=0), \
myutility.convertToOneHot(np.concatenate((data_y[0:i], \
data_y[i+step:]), axis=0), num_of_class)
LR = logistic.logistic_regression(train_x.shape[1], train_y.shape[1])
LR.fit(train_x, train_y)
predict_y = LR.predict(test_x)
acc += np.sum(predict_y == test_y) / test_y.shape[0]
print(acc / k_fold)
print(metrics.classification_report(test_y, predict_y))
acc = 0.
for i in range(k_fold):
test_x, test_y = data_x[i:i + step], data_y[i:i + step]
train_x, train_y = np.concatenate((data_x[0:i], \
data_x[i+step:]), axis=0), \
np.concatenate((data_y[0:i], data_y[i+step:]), axis=0)
LR2 = LogisticRegression(solver='lbfgs', multi_class='auto')
LR2.fit(train_x, train_y)
predict_y = LR2.predict(test_x)
acc += np.sum(predict_y == test_y) / test_y.shape[0]
print(acc / k_fold)
print(metrics.classification_report(test_y, predict_y))
return predict_y, test_y, num_of_class
def main_nb(dataset):
data = np.load('Datasets/' + dataset + '.npy')
np.random.shuffle(data)
num_samples = data.shape[0]
data_x = data[:num_samples, :-1]
data_y = data[:num_samples, -1]
step = data_x.shape[0] // k_fold
num_of_class = int(np.max(data_y) + 1)
acc = 0.
for i in range(k_fold):
test_x, test_y = data_x[i:i + step], data_y[i:i + step]
train_x, train_y = np.concatenate((data_x[0:i], \
data_x[i+step:]), axis=0), \
myutility.convertToOneHot(np.concatenate((data_y[0:i], \
data_y[i+step:]), axis=0), num_of_class)
NB = naivebayes.NaiveBayes(train_x.shape[1], train_y.shape[1])
NB.fit(train_x, train_y)
predict_y = NB.predict(test_x)
acc += np.sum(predict_y == test_y) / test_y.shape[0]
print(acc / k_fold)
print(metrics.classification_report(test_y, predict_y))
acc = 0.
for i in range(k_fold):
test_x, test_y = data_x[i:i + step], data_y[i:i + step]
train_x, train_y = np.concatenate((data_x[0:i], data_x[i+step:]), axis=0), \
np.concatenate((data_y[0:i], data_y[i+step:]), axis=0)
NB2 = MultinomialNB()
NB2.fit(train_x, train_y)
predict_y = NB2.predict(test_x)
acc += np.sum(predict_y == test_y) / test_y.shape[0]
print(acc / k_fold)
print(metrics.classification_report(test_y, predict_y))
data_y = data[:num_samples, -1]
iteration = 200
k_fold = 5
step = data_x.shape[0] // k_fold
num_of_class = int(np.max(data_y) + 1)
train_acc_all = [0] * iteration
test_acc_all = [0] * iteration
train_accs_avg = [] * len(datasets)
test_accs_avg = [] * len(datasets)
for i in range(k_fold):
train_accs = []
test_accs = []
test_x, test_y = data_x[i:i + step], data_y[i:i + step]
train_x, train_y = np.concatenate((data_x[0:i], \
data_x[i+step:]), axis=0), \
myutility.convertToOneHot(np.concatenate((data_y[0:i], \
data_y[i+step:]), axis=0), num_of_class)
LR = logistic.logistic_regression(train_x.shape[1],
train_y.shape[1],
max_iter=1000,
thres=0.05,
lr=0.0001)
for i in range(iteration):
LR.fit(train_x, train_y)
train_predict_y = LR.predict(train_x)
# if i == 0:
# print(train_predict_y.shape)
# print(train_y.shape)
# print(train_predict_y == train_y)
test_predict_y = LR.predict(test_x)
train_acc = np.sum(np.argmax(train_y, axis=1) ==
train_predict_y) / train_y.shape[0]