from elm import BvsbClassifier, BvsbUtils
from sklearn import datasets
from elm import elmUtils
from sklearn.preprocessing import StandardScaler
import time
print("---------OSELM-BVSB-----------")
data = elmUtils.readDataFileToData("data/zoo.data", targetIndex=0)
print(f'数据集大小问{data.target.size}')
data.data = StandardScaler().fit_transform(data.data)
(train, iter, test) = elmUtils.splitDataWithIter(data.data, data.target, 0.05,
0.3)
print(f'训练集大小为{train[1].size}')
print(f'迭代训练集大小为{iter[1].size}')
print(f'测试集大小为{test[1].size}')
tic = time.perf_counter_ns()
bvsbc = BvsbClassifier(train[0],
train[1],
iter[0],
iter[1],
test[0],
test[1],
iterNum=0.1)
bvsbc.createOSELM(n_hidden=1000)
bvsbc.trainOSELMWithoutKNN()
toc = time.perf_counter_ns()
print(f'OSELM-BVSB-KNN 正确率为{bvsbc.score(test[0], test[1])}')
print(f'OSELM-BVSB-KNN 项目用时:{(toc - tic) / 1000 / 1000} ms')
print("-----------------以下为OSELM-KNN(本文)算法（10）---------------")
acc_rem = [] #初始化精度列表
for ii in range(10):
#data.target = data.target + 1
#stdsc = StandardScaler()
#label_size = 0.5
(train_data, iter_data,
test_data) = elmUtils.splitDataWithIter(data.data, data.target,
label_size, 0.3)
iter_y = BvsbUtils.KNNClassifierResult(train_data[0], train_data[1],
iter_data[0])
bvsbc = BvsbClassifier(train_data[0],
train_data[1],
iter_data[0],
iter_y,
test_data[0],
test_data[1],
iterNum=0.1)
bvsbc.createOSELM(n_hidden=1000, active_function="sigmoid")
bvsbc.trainOSELMWithKNNButBvsb()
print(f'OSELM-BVSB-KNN 正确率为{bvsbc.score(test_data[0], test_data[1])}')
acc_temp = bvsbc.score(test_data[0], test_data[1]) #记录每次的精度
acc_rem.append(acc_temp) #将每次的精度存入列表
for i in acc_rem:
print(f'{i*100:0.2f}', ) #打印每次精度
acc_mean = np.mean(acc_rem) #求出平均精度
print("{:.2f}".format(acc_mean * 100)) #打印平均精度
print('---------------------以上为OSELM-BVSB-KNN(10次)算法----------------------'
Y_train = train_data[1].copy()
X_iter = iter_data[0].copy()
len_iter = len(X_iter)
i = 1
while len(X_iter) > (len_iter / 2):
nbr = BvsbUtils.KNNClassifier(X_train, Y_train) # KNN
# iter_y=nbr.predict(X_iter)
pred = nbr.predict_proba(X_iter)
iter_y = np.argmax(pred, axis=1)
classMax = np.max(pred, axis=1) # 获取未标记样本的最大隶属度
sortIndex = np.argsort(classMax) # 排序后的原下标
iter_index = np.sort(sortIndex[-select_h:])
sort_h_y = iter_y[iter_index] # 返回原来数据置信度最大的h个数
sort_h_data = X_iter[iter_index]
len_curr_iter = len(sort_h_y)
bvsbc = BvsbClassifier(X_train, Y_train, sort_h_data, sort_h_y, test_data[0], test_data[1], iterNum=0.1)
bvsbc.createELM(n_hidden=hidden_nums, activation_func="tanh", alpha=1.0, random_state=0)
_data_index = bvsbc.fitAndGetUpdateDataIndex(limit=int(0.2 * len_curr_iter))
if len(_data_index) != 0:
X_train = np.r_[bvsbc.X_train, sort_h_data[_data_index]]
Y_train = np.r_[bvsbc.Y_train, sort_h_y[_data_index]]
X_iter = np.delete(X_iter, iter_index[_data_index], axis=0)
else:
print("没有数据被加入训练集，训练结束")
break
print(f"第{ii} 次训练,第{i}次迭代: 正确率为:{bvsbc.score(test_data[0], test_data[1])}")
i += 1
acc_temp = bvsbc.score(test_data[0], test_data[1]) # 记录每次的精度
acc_rem.append(acc_temp) # 将每次的精度存入列表
print("***************ELM-BVSB-KNN加权算法（10次精度）********************")