Transformer Temporal Fusion

Pytorch implementation of Transformer Temporal Fusion (TFT) for time series forecast. This implementation, adapted from the referenced repository pytorch-forecasting, simplified for developers outside the computer science field.

Installation

Create Miniconda Environment

We recommend installing miniconda for managing Python environment, yet this repo works well with other alternatives e.g., venv.

Install miniconda by following these instructions

Create a conda environment

conda create --name your_env_name python=3.10

Activate conda environment
```
conda activate your_env_name
```

Install all required libraries

pip install -r requirements.txt

Example of Power Usage Prediction

Download Example Data

Either run the following command to download the data or directly visit the provided URL. NOTE: data must be saved in a folder data/

wget https://archive.ics.uci.edu/ml/machine-learning-databases/00321/LD2011_2014.txt.zip
unzip LD2011_2014.txt.zip

Train TFT Model

python runner.py

Prediction Interpretation

Directory Structure

.
├── runner.py                       # Run the training step
├── config                          # User-specified data variables and hyperparameters for TFT
├── data_preprocessor.py            # Preprocess data for TFT format
├── metric.py                       # Loss function for training e.g., Quantile loss
├── model.py                        # All model architectures required for TFT 
├── tft.py                          # TFT model
├── tft_interp.py                   # Output interpretation and visual tools

License

This implementation is released under the MIT license.

THIS IS AN OPEN SOURCE SOFTWARE FOR RESEARCH PURPOSES ONLY. THIS IS NOT A PRODUCT. NO WARRANTY EXPRESSED OR IMPLIED.

Acknowledgement

This code is adapted from the repository pytorch-forecasting. In addition, we would like to acknowledge and give credit to UC Irvine Machine Learning Repository for providing the sample dataset that we utilized for this project.

lhnguyen102 / tft-sgd