def one_step():
"""
一步到位运行所有对比方法
"""
x, y = read_data.get_data([6], -1, "ecoli.dat", show_info=True)
k = 5 # 交叉验证次数
# 期望每折交叉验证样本数量 >= 100
# while len(y) / k < 100:
# x, y = read_data.upsampling_copy(x, y, 1)
# print("复制一份后：%d/%d" % (len(y[y == 1]), len(y[y == 0])))
print("|%-20s|%-20s|%-20s|%-20s|%-20s" % ("", "f1score", "auc", "gmean", "bACC"))
print("|%-20s|%-20s|%-20s|%-20s|%-20s" % ("----", "----", "----", "----", "----"))
method = ("KNN", "DT", "RandomForest", "AdaBoost", "EasyEnsemble", "BalancedBagging")
sampling = ("RUS", "SMOTE")
for m in method:
if m in ("KNN", "DT"):
for s in sampling:
result = kFoldTest(x.copy(), y.copy(), sampler=s, classifier=m, k=k)
print(result[1])
else:
result = kFoldTest(x.copy(), y.copy(), sampler="", classifier=m, k=k)
print(result[1])
def main():
params = Params()
if params.GPU:
configure_gpu()
images, angles = get_data(params.parametrized)
params.set_angles(angles)
X_train, X_test, y_train, y_test = train_test_split(images,
angles,
test_size=0.2,
random_state=42)
generator, discriminator = train(params, X_train, y_train)
save_generator(params, generator)
def one_step():
"""
一步到位运行所有对比方法
"""
x, y = read_data.get_data([2], -1, "zoo.dat", show_info=True)
k = 5
while len(y[y == 0]) / k < 50:
x, y = read_data.upsampling_copy(x, y, 1)
print("复制一份后：%d/%d" % (len(y[y == 1]), len(y[y == 0])))
print("|%-20s|%-20s|%-20s|%-20s|%-20s|%-20s|%-20s" % ("", "val_acc", "val_precision", "val_recall", "val_f1", "auc_value", "val_gmean"))
print("|%-20s|%-20s|%-20s|%-20s|%-20s|%-20s|%-20s" % ("----", "----", "----", "----", "----", "----", "----"))
classify_method = ("KNN", "DT", "SVC")
sampling_method = ("",)
for classifier in classify_method:
for sampler in sampling_method:
result = compare.kFoldTest(x.copy(), y.copy(), sampler=sampler, classifier=classifier, k=k)
print(result[0])
# 测试
y_proba = clf.predict_proba(x_val)
y_pred = np.argmax(y_proba, axis=1)
# 进化前的表现
save_metric(val_history, y_val, y_pred, y_proba)
print("进化前：")
show_last_data(val_history)
# 进化
s = equation.get_S_matrix(x_val)
y_proba = PSOEvolutor(s).evolve(y_proba, max_steps=200)
y_pred = np.argmax(y_proba, axis=1)
# 进化后的表现
save_metric(evo_history, y_val, y_pred, y_proba)
print("进化后：")
show_last_data(evo_history)
# 统计，求平均值和标准差
print("进化前平均：")
show_mean_data(val_history)
print("进化后平均：")
show_mean_data(evo_history)
if __name__ == '__main__':
x, y = read_data.get_data([0, 6], -1, "yeast.dat", show_info=True)
kFoldEvolution(x, y)
:param x: 集合
:return: 欧式距离，矩阵
"""
dis = np.zeros((len(x), len(x)))
for i, xi in enumerate(x):
for j, xj in enumerate(x):
dis[i][j] = np.sqrt(np.sum(np.square(xi - xj)))
return dis
def get_neighbor(self, d):
"""
获取最近的 k 个邻居编号
:param d: 所有邻居的距离
:return: k 个最近邻居的编号，从近到远
"""
t = sorted(enumerate(d), key=lambda x: x[1])
t = [x[0] for x in t]
return t[:self.k]
if __name__ == '__main__':
x, y = read_data.get_data([1], -1, "banana.dat", show_info=True)
x_neg = x[y == 0]
print("采样前：%d" % len(x_neg))
x_neg = SmoteSampler(5).resample(x_neg, 100)
print("采样后：%d" % len(x_neg))
y = -1
else:
y = 1
if y_pred[i] == 0:
h = -1
else:
h = 1
Z += C[Y[i]] * D[i] * np.exp(-alpha * y * h)
return Z
if __name__ == '__main__':
# 获取原始数据
x, y = get_data([0, 6], -1, "1到5/yeast.dat")
x, y = shuffle_data(x, y)
x = np.array(x)
y = np.array(y, dtype=np.int8)
# y[y == 0] = -1
# 代价项
C = {1: 0.1, 0: 1}
x_train = x[:-100]
y_train = y[:-100]
x_val = x[-100:]
y_val = y[-100:]
print("进化前平均：")
experiment_helper.show_mean_data(val_history)
print("进化后平均：")
experiment_helper.show_mean_data(evo_history)
print("结合后平均：")
experiment_helper.show_mean_data(mean_history)
auc_record["%d:%d" % (n_under, n_up)] = get_auc(mean_history)
print(auc_record)
else:
experiment_helper.show_mean_data(val_history)
def get_auc(history):
return np.mean(history["auc"])
auc_record = {}
if __name__ == '__main__':
N_under = (5, 10, 15, 20)
N_up = (5, 10, 15, 20)
for i in N_under:
for j in N_up:
x, y = read_data.get_data([6], -1, "yeast.dat", show_info=False)
# 期望每折交叉验证样本数量 >= 100
# for i in range(1):
# x, y = read_data.upsampling_copy(x, y, 1)
# print("复制一份后：%d/%d" % (len(y[y == 1]), len(y[y == 0])))
kFoldEvolution(x, y, evolution=True, n_under=i, n_up=j)