Time Series Datasets

We are the Time Series Analysis Team of Southeast University.

email: zysong@seu.edu.cn

We are developing a new benchmark that includes a broader dataset and a more lightweight code framework to address the issue of excessive encapsulation in current time series forecasting libraries.

COMMON: such as ETT, Traffic, Electricity, PEMS
TFB: form paper "TFB: Towards Comprehensive and Fair Benchmarking of Time Series Forecasting Methods"
WORKLOAD: form paper "Fremer: Lightweight and Effective Frequency Transformer for Workload Forecasting in Cloud Services", ByteDance

Dataset format

dict:

'data': np.array, shape: (length, num_variates)

'time_date': the DatetimeIndex, shape: (length,)

'columns': np.array, shape: (num_variates,)

'freq': np.str

'cycle': np.array, the series cycles (*e.g.*, 24, 24*7 and so on). We don't test the cycle if the value is -1.

Loading Dataset

import numpy as np
import pandas as pd

df_data_columns_date = np.load(file_path, allow_pickle=True).item()
df_data = df_data_columns_date["data"]  # shape (seq_len, num_features)
# df_columns = df_data_columns_date["columns"]  # list of column names
df_date = df_data_columns_date["time_date"]  # pd.DatetimeIndex (seq_len, )