FloList / EMPL

The Earth Mover's Pinball Loss: Quantiles for Histogram-Valued Regression
7 stars 0 forks source link

can it be used for tabular data ? #1

Closed Sandy4321 closed 11 months ago

Sandy4321 commented 2 years ago

can it be used for tabular data ?

like https://towardsdatascience.com/deep-quantile-regression-in-tensorflow-1dbc792fe597

FloList commented 2 years ago

Hi Sandy, This Github repo is specifically for the task of histogram-valued regression (so the label belonging to each input should be an entire histogram), and a neural network trained with the Earth Mover's Pinball Loss will estimate the quantiles of the cumulative histogram belonging to each input. If you're looking for a loss function to do quantile regression for scalar labels as in the blog post you mentioned, you can directly take the pinball loss (or quantile loss), for example here: https://github.com/strongio/quantile-regression-tensorflow/blob/master/Quantile%20Loss.ipynb. Also, the pinball loss function is available as a Tensorflow Add-on. Cheers, Florian

Sandy4321 commented 11 months ago

thanks for answer for Also, the pinball loss function is available as a Tensorflow Add-on. is written

Warning: This project is deprecated. TensorFlow Addons has stopped development, The project will only be providing minimal maintenance releases until May 2024. See the full announcement here or on github.

Sandy4321 commented 11 months ago

it is questionabe if https://github.com/strongio/quantile-regression-tensorflow/blob/master/Quantile%20Loss.ipynb

will work for new TF versions