Skip to content
/ TENT Public

Official implementation code of the paper: "TENT: Tensorized Encoder Transformer for temperature forecasting".

Notifications You must be signed in to change notification settings

onurbil/TENT

Repository files navigation

TENT

Official implementation code of the paper: "TENT: Tensorized Encoder Transformer for temperature forecasting" (ArXiv link).

Model architecture

TENT

(a) Model architectureTensorial. (b) Tensorial self-attention. (c) Tensorial multi-head attention.

Results

TENT The obtained test MAE of the models for USA-Canada dataset averaged over cities (a) and prediction time steps (b).

TENT The comparison between the predictions of TENT model and the real measurements for hourly temperature of the test set of Vancouver.

Attention visualization

TENT Attention visualization for Dallas in USA-Canada dataset.

Data

In order to download the data, please email to the following address:

[email protected]

Usage

Execute the notebook on colab (Use TPU for TENT): TT_All_models_experiments.ipynb

Citation

If you use our data and code, please cite the paper using the following bibtex reference:

@article{bilgin2021tent,
  title={TENT: Tensorized Encoder Transformer for Temperature Forecasting},
  author={Bilgin, Onur and M{\k{a}}ka, Pawe{\l} and Vergutz, Thomas and Mehrkanoon, Siamak},
  journal={arXiv preprint arXiv:2106.14742v2},
  year={2021}
}

About

Official implementation code of the paper: "TENT: Tensorized Encoder Transformer for temperature forecasting".

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published