我想重新采样我的数据集。这包括具有 3 个类别标签的分类转换数据。每类样本数量为：
没有标签的数据形状是 (16661, 1000, 256)。这意味着 (1000,256) 的 16661 个样本。我想要的是将数据上采样到多数类的样本数，即A类->（6945）
但是，调用时：
from imblearn.over_sampling import SMOTE
print(categorical_vector.shape)
sm = SMOTE(random_state=2)
X_train_res, y_labels_res = sm.fit_sample(categorical_vector, labels.ravel())
它一直说 ValueError: Found array with dim 3. Estimator expected <= 2.
如何以估算器可以拟合并且也有意义的方式展平数据？此外，在获得 X_train_res 后如何展开（使用 3D 维度）？
我正在考虑一个虚拟3d数组并自己假设一个2d数组大小，
arr = np.random.rand(160, 10, 25)
orig_shape = arr.shape
print(orig_shape)
输出：(160, 10, 25)
arr = np.reshape(arr, (arr.shape[0], arr.shape[1]))
print(arr.shape)
输出：(4000, 10)
arr = np.reshape(arr, orig_shape))
print(arr.shape)
输出：(160, 10, 25)
from imblearn.over_sampling
import RandomOverSampler
import numpy as np
oversample = RandomOverSampler(sampling_strategy='minority')
X 可以是时间步长的 3D 数据，例如 X[sample,time,feature]，而 y 可以是每个样本的二进制值。例如：(1,1),(2,1),(3,1) -> 1
X = np.array([[[1,1],[2,1],[3,1]],
[[2,1],[3,1],[4,1]],
[[5,1],[6,1],[7,1]],
[[8,1],[9,1],[10,1]],
[[11,1],[12,1],[13,1]]
])
y = np.array([1,0,1,1,0])
无法使用 3D X 值训练 OVERSAMPLER，因为如果使用 2D，您将获得 2D 数据。
Xo,yo = oversample.fit_resample(X[:,:,0], y)
Xo:
[[ 1 2 3]
[ 2 3 4]
[ 5 6 7]
[ 8 9 10]
[11 12 13]
[ 2 3 4]]
yo:
[1 0 1 1 0 0]
但是如果您使用 2D 数据 (sample,time,0) 来拟合模型，它将返回索引，并且足以创建 3D 过采样数据
oversample.fit_resample(X[:,:,0], y)
Xo = X[oversample.sample_indices_]
yo = y[oversample.sample_indices_]
Xo:
[[[ 1 1][ 2 1][ 3 1]]
[[ 2 1][ 3 1][ 4 1]]
[[ 5 1][ 6 1][ 7 1]]
[[ 8 1][ 9 1][10 1]]
[[11 1][12 1][13 1]]
[[ 2 1][ 3 1][ 4 1]]]
yo:
[1 0 1 1 0 0]
I will create each point for a 2-dim array and then reshape it as 3 dim array. I have provided my scripts. If there is any confusion, comment; please reply.
x_train, y_train = zip(*train_dataset)
x_test, y_test = zip(*test_dataset)
dim_1 = np.array(x_train).shape[0]
dim_2 = np.array(x_train).shape[1]
dim_3 = np.array(x_train).shape[2]
new_dim = dim_1 * dim_2
new_x_train = np.array(x_train).reshape(new_dim, dim_3)
new_y_train = []
for i in range(len(y_train)):
# print(y_train[i])
new_y_train.extend([y_train[i]]*dim_2)
new_y_train = np.array(new_y_train)
# transform the dataset
oversample = SMOTE()
X_Train, Y_Train = oversample.fit_sample(new_x_train, new_y_train)
# summarize the new class distribution
counter = Counter(Y_Train)
print('The number of samples in TRAIN: ', counter)
x_train_SMOTE = X_Train.reshape(int(X_Train.shape[0]/dim_2), dim_2, dim_3)
y_train_SMOTE = []
for i in range(int(X_Train.shape[0]/dim_2)):
# print(i)
value_list = list(Y_Train.reshape(int(X_Train.shape[0]/dim_2), dim_2)[i])
# print(list(set(value_list)))
y_train_SMOTE.extend(list(set(value_list)))
## Check: if there is any different value in a list
if len(set(value_list)) != 1:
print('\n\n********* STOP: THERE IS SOMETHING WRONG IN TRAIN ******\n\n')
dim_1 = np.array(x_test).shape[0]
dim_2 = np.array(x_test).shape[1]
dim_3 = np.array(x_test).shape[2]
new_dim = dim_1 * dim_2
new_x_test = np.array(x_test).reshape(new_dim, dim_3)
new_y_test = []
for i in range(len(y_test)):
# print(y_train[i])
new_y_test.extend([y_test[i]]*dim_2)
new_y_test = np.array(new_y_test)
# transform the dataset
oversample = SMOTE()
X_Test, Y_Test = oversample.fit_sample(new_x_test, new_y_test)
# summarize the new class distribution
counter = Counter(Y_Test)
print('The number of samples in TEST: ', counter)
x_test_SMOTE = X_Test.reshape(int(X_Test.shape[0]/dim_2), dim_2, dim_3)
y_test_SMOTE = []
for i in range(int(X_Test.shape[0]/dim_2)):
# print(i)
value_list = list(Y_Test.reshape(int(X_Test.shape[0]/dim_2), dim_2)[i])
# print(list(set(value_list)))
y_test_SMOTE.extend(list(set(value_list)))
## Check: if there is any different value in a list
if len(set(value_list)) != 1:
print('\n\n********* STOP: THERE IS SOMETHING WRONG IN TEST ******\n\n')