Spaces:
Running
Running
File size: 12,040 Bytes
d03866e |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 |
import torch
import torch.utils.data
import numpy as np
epsilon = 1e-8
class TimeRCDDataset(torch.utils.data.Dataset):
def __init__(self, data, window_size, stride=1, normalize=False, pad_to_multiple=True):
super().__init__()
self.window_size = window_size
self.stride = stride
# Ensure numpy array and a consistent 2D shape (N, C)
data = np.asarray(data)
if data.ndim == 1:
data = data.reshape(-1, 1)
self.original_length = data.shape[0]
self.pad_to_multiple = pad_to_multiple
# Normalize data if other than UCR
self.data = self._normalize_data(data) if normalize else data
# self.data = data
# self.univariate = self.data.shape[0] == 1
# Handle padding if requested
if self.pad_to_multiple:
self.data, self.padding_mask = self._pad_data_to_multiple()
else:
self.padding_mask = np.ones(self.data.shape[0], dtype=bool) # All data is real
def _normalize_data(self, data, epsilon=1e-8):
""" Normalize data using mean and standard deviation. """
mean, std = np.mean(data, axis=0), np.std(data, axis=0)
std = np.where(std == 0, epsilon, std)
return ((data - mean) / std)
def _pad_data_to_multiple(self):
""" Pad data to make its length a multiple of window_size and return padding mask. """
data_length = self.data.shape[0]
remainder = data_length % self.window_size
if remainder == 0:
# No padding needed - all data is real
padding_mask = np.ones(data_length, dtype=bool)
return self.data, padding_mask
# Calculate padding needed
padding_length = self.window_size - remainder
print(f"Padding AnomalyClipDataset: original length {data_length}, window_size {self.window_size}, adding {padding_length} samples")
# Pad by repeating the last row, keeping 2D shape (1, C)
last_row = self.data[-1:, :]
padding_data = np.repeat(last_row, padding_length, axis=0)
padded_data = np.vstack([self.data, padding_data])
# Create padding mask: True for real data, False for padded data
padding_mask = np.ones(data_length + padding_length, dtype=bool)
padding_mask[data_length:] = False # Mark padded samples as False
return padded_data, padding_mask
def __getitem__(self, index):
start = index * self.stride
end = start + self.window_size
if end > self.data.shape[0]:
raise IndexError("Index out of bounds for the dataset.")
# Always return (window_size, num_features)
sample = torch.tensor(self.data[start:end, :], dtype=torch.float32)
mask = torch.tensor(self.padding_mask[start:end], dtype=torch.bool)
# if self.univariate:
# sample = sample.unsqueeze(-1) # Add channel dimension for univariate data
return sample, mask
def __len__(self):
return max(0, (self.data.shape[0] - self.window_size) // self.stride + 1)
class ReconstructDataset(torch.utils.data.Dataset):
def __init__(self, data, window_size, stride=1, normalize=True):
super().__init__()
self.window_size = window_size
self.stride = stride
self.data = self._normalize_data(data) if normalize else data
data = np.asarray(data)
if data.ndim == 1:
data = data.reshape(-1, 1)
self.univariate = data.shape[1] == 1
self.sample_num = max(0, (self.data.shape[0] - window_size) // stride + 1)
self.samples, self.targets = self._generate_samples()
def _normalize_data(self, data, epsilon=1e-8):
mean, std = np.mean(data, axis=0), np.std(data, axis=0)
std = np.where(std == 0, epsilon, std) # Avoid division by zero
return (data - mean) / std
def _generate_samples(self):
data = torch.tensor(self.data, dtype=torch.float32)
if self.univariate:
data = data.squeeze()
X = torch.stack([data[i * self.stride : i * self.stride + self.window_size] for i in range(self.sample_num)])
X = X.unsqueeze(-1)
else:
X = torch.stack([data[i * self.stride : i * self.stride + self.window_size, :] for i in range(self.sample_num)])
return X, X
def __len__(self):
return self.sample_num
def __getitem__(self, index):
return self.samples[index], self.targets[index]
class ForecastDataset(torch.utils.data.Dataset):
def __init__(self, data, window_size, pred_len, stride=1, normalize=True):
super().__init__()
self.window_size = window_size
self.pred_len = pred_len
self.stride = stride
self.data = self._normalize_data(data) if normalize else data
data = np.asarray(data)
if data.ndim == 1:
data = data.reshape(-1, 1)
self.sample_num = max((self.data.shape[0] - window_size - pred_len) // stride + 1, 0)
# Generate samples efficiently
self.samples, self.targets = self._generate_samples()
def _normalize_data(self, data, epsilon=1e-8):
""" Normalize data using mean and standard deviation. """
mean, std = np.mean(data, axis=0), np.std(data, axis=0)
std = np.where(std == 0, epsilon, std) # Avoid division by zero
return (data - mean) / std
def _generate_samples(self):
""" Generate windowed samples efficiently using vectorized slicing. """
data = torch.tensor(self.data, dtype=torch.float32)
indices = np.arange(0, self.sample_num * self.stride, self.stride)
X = torch.stack([data[i : i + self.window_size] for i in indices])
Y = torch.stack([data[i + self.window_size : i + self.window_size + self.pred_len] for i in indices])
return X, Y # Inputs & targets
def __len__(self):
return self.sample_num
def __getitem__(self, index):
return self.samples[index], self.targets[index]
# class ForecastDataset(torch.utils.data.Dataset):
# def __init__(self, data, window_size, pred_len, normalize=True):
# super().__init__()
# self.normalize = normalize
# if self.normalize:
# data_mean = np.mean(data, axis=0)
# data_std = np.std(data, axis=0)
# data_std = np.where(data_std == 0, epsilon, data_std)
# self.data = (data - data_mean) / data_std
# else:
# self.data = data
# self.window_size = window_size
# if data.shape[1] == 1:
# data = data.squeeze()
# self.len, = data.shape
# self.sample_num = max(self.len - self.window_size - pred_len + 1, 0)
# X = torch.zeros((self.sample_num, self.window_size))
# Y = torch.zeros((self.sample_num, pred_len))
# for i in range(self.sample_num):
# X[i, :] = torch.from_numpy(data[i : i + self.window_size])
# Y[i, :] = torch.from_numpy(np.array(
# data[i + self.window_size: i + self.window_size + pred_len]
# ))
# self.samples, self.targets = torch.unsqueeze(X, -1), torch.unsqueeze(Y, -1)
# else:
# self.len = self.data.shape[0]
# self.sample_num = max(self.len - self.window_size - pred_len + 1, 0)
# X = torch.zeros((self.sample_num, self.window_size, self.data.shape[1]))
# Y = torch.zeros((self.sample_num, pred_len, self.data.shape[1]))
# for i in range(self.sample_num):
# X[i, :] = torch.from_numpy(data[i : i + self.window_size, :])
# Y[i, :] = torch.from_numpy(data[i + self.window_size: i + self.window_size + pred_len, :])
# self.samples, self.targets = X, Y
# def __len__(self):
# return self.sample_num
# def __getitem__(self, index):
# return self.samples[index, :, :], self.targets[index, :, :]
class TSDataset(torch.utils.data.Dataset):
def __init__(self, X, y=None, mean=None, std=None):
super(TSDataset, self).__init__()
self.X = X
self.mean = mean
self.std = std
def __len__(self):
return self.X.shape[0]
def __getitem__(self, idx):
if torch.is_tensor(idx):
idx = idx.tolist()
sample = self.X[idx, :]
if self.mean is not None and self.std is not None:
sample = (sample - self.mean) / self.std
# assert_almost_equal (0, sample.mean(), decimal=1)
return torch.from_numpy(sample), idx
class ReconstructDataset_Moment(torch.utils.data.Dataset):
def __init__(self, data, window_size, stride=1, normalize=True):
super().__init__()
self.window_size = window_size
self.stride = stride
self.data = self._normalize_data(data) if normalize else data
self.univariate = self.data.shape[1] == 1
self.sample_num = max((self.data.shape[0] - window_size) // stride + 1, 0)
self.samples = self._generate_samples()
self.input_mask = np.ones(self.window_size, dtype=np.float32) # Fixed input mask
def _normalize_data(self, data, epsilon=1e-8):
mean, std = np.mean(data, axis=0), np.std(data, axis=0)
std = np.where(std == 0, epsilon, std) # Avoid division by zero
return (data - mean) / std
def _generate_samples(self):
data = torch.tensor(self.data, dtype=torch.float32)
indices = np.arange(0, self.sample_num * self.stride, self.stride)
if self.univariate:
X = torch.stack([data[i : i + self.window_size] for i in indices])
else:
X = torch.stack([data[i : i + self.window_size, :] for i in indices])
return X
def __len__(self):
return self.sample_num
def __getitem__(self, index):
return self.samples[index], self.input_mask
class TACLipDataset(torch.utils.data.Dataset):
def __init__(self, data, win_size, step=1, flag="test"):
self.flag = flag
self.step = step
self.win_size = win_size
self.test = data
print("Before normalization", self.test[:20])
self.test = self._normalize_data(self.test)
print("After normalization", self.test[:20])
self.test_labels = np.zeros(self.test.shape[0])
def _normalize_data(self, data, epsilon=1e-8):
mean, std = np.mean(data, axis=0), np.std(data, axis=0)
std = np.where(std == 0, epsilon, std) # Avoid division by zero
return (data - mean) / std
def __len__(self):
"""
Number of images in the object dataset.
"""
if self.flag == "train":
return (self.train.shape[0] - self.win_size) // self.step + 1
elif (self.flag == 'val'):
return (self.val.shape[0] - self.win_size) // self.step + 1
elif (self.flag == 'test'):
return (self.test.shape[0] - self.win_size) // self.step + 1
else:
return (self.test.shape[0] - self.win_size) // self.win_size + 1
def __getitem__(self, index):
index = index * self.step
if self.flag == "train":
return np.float32(self.train[index:index + self.win_size]), np.float32(self.test_labels[0:self.win_size])
elif (self.flag == 'val'):
return np.float32(self.val[index:index + self.win_size]), np.float32(self.test_labels[0:self.win_size])
elif (self.flag == 'test'):
return np.float32(self.test[index:index + self.win_size]), np.float32(
self.test_labels[index:index + self.win_size])
else:
return np.float32(self.test[
index // self.step * self.win_size:index // self.step * self.win_size + self.win_size]), np.float32(
self.test_labels[index // self.step * self.win_size:index // self.step * self.win_size + self.win_size])
|