Readme.MD

TSMixer forecasting

This package is a demonstration of how to train and use TSMixer for forecasting.

Installation

pip install -e .[all]

Alternatively, you don't need to include any optional dependencies (i.e. pip install -e .)

Datasets

To download the datasets either:

make datasets

Training

To train the models as per the paper run:

make train-weather [out_dir="."]
make train-ETTm2 [out_dir="."]
make train-electricity [out_dir="."]
make train-traffic [out_dir="."]

To train the model with custom parameters, refer to the help via:

python3 -m tsmixer train --help

Example

For the following example, we train on the weather dataset. Our losses and errors are as follow:

Sample forecasts

Forecasts were run using:

make forecast out_dir="./weather"