Papers and datasets for tensor time series.

• TTS-Norm TTS-Norm: Forecasting Tensor Time Series via Multi-Way Normalization (ACM TKDD 2023) [paper] [code]
• GMRL Learning Gaussian Mixture Representations for Tensor Time Series Forecasting (IJCAI 2023) [paper] [code]
• NET3 Network of Tensor Time Series (WWW 2021) [paper] [code]
• STC-GNN Spatio-Temporal-Categorical Graph Neural Networks for Fine-Grained Multi-Incident Co-Prediction (CIKM 2021) [paper] [code]
• DMSTGCN Dynamic and multi-faceted spatiotemporal deep learning for traffic speed forecasting (KDD 2021) [paper] [code] Here is the list of papers organized in the requested format:
• ST-Norm Spatial and temporal normalization for multi-variate time series forecasting (KDD 2021) [paper] [code]
• ReGENN Pay Attention to Evolution: Time Series Forecasting with Deep Graph-Evolution Learning (TPAMI 2020) [paper] [code]
• MTGNN Connecting the Dots: Multivariate Time Series Forecasting with Graph Neural Networks (NeurIPS 2020) [paper] [code]
• AGCRN Adaptive Graph Convolutional Recurrent Network for Traffic Forecasting (NeurIPS 2020) [paper] [code]
• StemGNN Spectral temporal graph neural network for multivariate time-series forecasting (NeurIPS 2020) [paper] [code]
• STTran Hierarchically structured transformer networks for fine-grained spatial event forecasting (WWW 2020) [paper]
• CoST-Net Co-Prediction of Multiple Transportation Demands Based on Deep Spatio-Temporal Neural Network (KDD 2019) [paper]
• MiST: A Multiview and Multimodal Spatial-Temporal Learning Framework for Citywide Abnormal Event Forecasting (WWW 2019) [paper]
• Graph Wavenet Graph WaveNet for Deep Spatial-Temporal Graph Modeling (IJCAI 2019) [paper] [code]
• DCRNN Diffusion Convolutional Recurrent Neural Network: Data-Driven Traffic Forecasting (ICLR 2018) [paper] [code]
• STGCN Spatio-Temporal Graph Convolutional Networks: A Deep Learning Framework for Traffic Forecasting (IJCAI 2018) [paper] [code]
• MLDS Multilinear dynamical systems for tensor time series (NeurIPS 2013) [paper] [code]
• DynaMMo DynaMMo: mining and summarization of coevolving sequences with missing values (KDD 2009) [paper] [code]

For datasets, please refer to Datasets

Create a virtual environment before we get stated. (Python >= 3.8)

conda create --name TensorTSL

An easy way to install the environment is to use pip install with the config file pyproject.toml.

pip install .

Run a simple task.

python3 main.py

There have been some tasks already. You can try to run python3 run_tasks.py --help for help.

Tasks:

• MTS_Task: run models in MTS_ModelList with specific datasets and data_mode.

# for example
python3 run_tasks.py --his_len 96 --pred_len 12 --dataset Finance --task_name MTS_Task --output_dir './output/'

• TTS_Task: run models in TTS_ModelList with specific datasets and data_mode. (GNN is initialized with 'pearson')

# for example
python3 run_tasks.py --his_len 96 --pred_len 12 --dataset Finance --task_name TTS_Task --output_dir './output/'

• Graph_Init_Task: run Models with prior graph with different graph initialization.

# for example
python3 run_tasks.py --his_len 96 --pred_len 12 --dataset Finance --task_name MTS_Task --output_dir './output/' --graph_init random

Due to its modular design, developing with our framework is straightforward and efficient.

In our framework, there are two types of models: TensorModel and MultiVarModel. These models are categorized based on the shape of the input data.

• TensorModel supports data inputs with the shape (time, dim1, dim2)
• MultiVarModel supports data input with the shape (time, dim)