Benchmark against state of the art LSTM architectures

[firobeid]

Is your feature request related to a problem? Please describe. A simple benchmark against powerful LSTMs that are able to handle long forecasts. If you library beats LSTMS then going with the faster, easier to implement and more interpretable model is best!

Describe the solution you'd like Showing that your framework beats LSTMs or other ,ML techniques for forecasting, give more exposure so that it can be adopted faster in the industry.

[kdgutier]

Hi @firobeid,

Thanks for the recommendation, we have compared in the past to stronger neural forecasting baselines. Here is an experiment where we compare the automatic ETS with Facebook's NeuralProphet algorithm. In the experiments on hundreds of thousands of series ETS was 32% more accurate and 104x faster.

If you are interested in testing other neural-forecasting methods, we are actively working in the NeuralForecast library which we hope is of use to your projects.

Here is a quick start tutorial for NeuralForecast.

[firobeid]

Thank you @kdgutier! Will definitely be showcasing in my lectures at Berkeley!

[kdgutier]

Awesome!