Good example of tabnet on a regression problem

Feature request

What is the expected behavior? Example to reproduce results from the tabnet paper on Rossmann Store Sales

What is motivation or use case for adding/changing the behavior? The current example on regression is a bit confusing to me. The example predicts for “income level” which only has two options <=50K and >50k. Maybe I missed something, but I don’t think this is a regression problem?

How should this be implemented in your opinion? An example using Rossmann Store Sales

Are you willing to work on this yourself? No, I am trying to learn tabnet on regression (so I think I lack the skills to do this currently), but need a good example to learn on that is why I am asking this 😕

Thank you.

[Optimox]

To add a few words, best way to use a Kaggle Notebook is probably to fork it, you’ll then be able to run it in your web browser within a docker container with all the libraries installed, or you can work in docker container yourself: https://github.com/Kaggle/docker-python (kaggle container is huge so not the best way).

About time series, TabNet is similar to XGBoost on this, you’ll need to engineer explicit lag features in order to do time series forecasting. It’s definitely doable and might be the good way to go BUT TabNet has nothing specific to time series like Temporal Convolutional Networks.

[eduardocarvp]

Hi @turmeric-blend ,

TabNet does not support time series classification or regression. You can have a look here, for example, on how to adapt a time series problem to a classic regression algorithm like XGBoostRegressor or TabNetRegressor.

If you really need some specific algorithm for time series data I’d refer to gluonTS or some similar library.