Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -103,93 +103,7 @@ UTSD is constructed with hierarchical capacities, namely **UTSD-1G, UTSD-2G, UTS
|
|
103 |
|
104 |
## Usage
|
105 |
|
106 |
-
You can load UTSD
|
107 |
-
|
108 |
-
```python
|
109 |
-
import datasets
|
110 |
-
import numpy as np
|
111 |
-
from torch.utils.data import Dataset
|
112 |
-
from sklearn.preprocessing import StandardScaler
|
113 |
-
from tqdm import tqdm
|
114 |
-
|
115 |
-
class UTSDDataset(Dataset):
|
116 |
-
def __init__(self, remote=True, root_path=r'UTSD-1G', flag='train', input_len=None, pred_len=None, scale=True,
|
117 |
-
stride=1, split=0.9):
|
118 |
-
self.input_len = input_len
|
119 |
-
self.pred_len = pred_len
|
120 |
-
self.seq_len = input_len + pred_len
|
121 |
-
assert flag in ['train', 'val']
|
122 |
-
assert split >= 0 and split <=1.0
|
123 |
-
type_map = {'train': 0, 'val': 1, 'test': 2}
|
124 |
-
self.set_type = type_map[flag]
|
125 |
-
self.flag = flag
|
126 |
-
self.scale = scale
|
127 |
-
self.split = split
|
128 |
-
self.stride = stride
|
129 |
-
self.remote = remote
|
130 |
-
|
131 |
-
self.data_list = []
|
132 |
-
self.n_window_list = []
|
133 |
-
|
134 |
-
self.root_path = root_path
|
135 |
-
self.__read_data__()
|
136 |
-
|
137 |
-
def __read_data__(self):
|
138 |
-
if self.remote:
|
139 |
-
dataset = datasets.load_dataset("thuml/UTSD", "UTSD-1G")['train']
|
140 |
-
else:
|
141 |
-
dataset = datasets.load_from_disk(self.root_path)
|
142 |
-
|
143 |
-
print(dataset)
|
144 |
-
for item in tqdm(dataset):
|
145 |
-
self.scaler = StandardScaler()
|
146 |
-
data = item['target']
|
147 |
-
data = np.array(data).reshape(-1, 1)
|
148 |
-
num_train = int(len(data) * self.split)
|
149 |
-
border1s = [0, num_train - self.seq_len]
|
150 |
-
border2s = [num_train, len(data)]
|
151 |
-
|
152 |
-
border1 = border1s[self.set_type]
|
153 |
-
border2 = border2s[self.set_type]
|
154 |
-
|
155 |
-
if self.scale:
|
156 |
-
train_data = data[border1s[0]:border2s[0]]
|
157 |
-
self.scaler.fit(train_data)
|
158 |
-
data = self.scaler.transform(data)
|
159 |
-
|
160 |
-
data = data[border1:border2]
|
161 |
-
n_window = (len(data) - self.seq_len) // self.stride + 1
|
162 |
-
if n_window < 1:
|
163 |
-
continue
|
164 |
-
|
165 |
-
self.data_list.append(data)
|
166 |
-
self.n_window_list.append(n_window if len(self.n_window_list) == 0 else self.n_window_list[-1] + n_window)
|
167 |
-
|
168 |
-
|
169 |
-
def __getitem__(self, index):
|
170 |
-
dataset_index = 0
|
171 |
-
while index >= self.n_window_list[dataset_index]:
|
172 |
-
dataset_index += 1
|
173 |
-
|
174 |
-
index = index - self.n_window_list[dataset_index - 1] if dataset_index > 0 else index
|
175 |
-
n_timepoint = (len(self.data_list[dataset_index]) - self.seq_len) // self.stride + 1
|
176 |
-
|
177 |
-
s_begin = index % n_timepoint
|
178 |
-
s_begin = self.stride * s_begin
|
179 |
-
s_end = s_begin + self.seq_len
|
180 |
-
p_begin = s_end
|
181 |
-
p_end = p_begin + self.pred_len
|
182 |
-
seq_x = self.data_list[dataset_index][s_begin:s_end, :]
|
183 |
-
seq_y = self.data_list[dataset_index][p_begin:p_end, :]
|
184 |
-
|
185 |
-
return seq_x, seq_y
|
186 |
-
|
187 |
-
def __len__(self):
|
188 |
-
return self.n_window_list[-1]
|
189 |
-
|
190 |
-
dataset = UTSDDataset(input_len=1440, pred_len=96)
|
191 |
-
print(len(dataset))
|
192 |
-
```
|
193 |
|
194 |
It should be noted that due to the construction of our dataset with diverse lengths, the sequence lengths of different samples vary. You can construct the data organization logic according to your own needs.
|
195 |
|
|
|
103 |
|
104 |
## Usage
|
105 |
|
106 |
+
You can access and load UTSD based on [the following code](https://github.com/thuml/Large-Time-Series-Model/tree/main/scripts/UTSD) in our repo.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
107 |
|
108 |
It should be noted that due to the construction of our dataset with diverse lengths, the sequence lengths of different samples vary. You can construct the data organization logic according to your own needs.
|
109 |
|