awslabs / gluonts

Probabilistic time series modeling in Python
https://ts.gluon.ai
Apache License 2.0
4.61k stars 752 forks source link

Feature request: add example for classification #420

Open lsorber opened 4 years ago

lsorber commented 4 years ago

Description

timoschowski commented 4 years ago

A simple example for anomaly detection is here: https://github.com/awslabs/gluon-ts/blob/master/examples/anomaly_detection.py

Classification would indeed be nice, but we haven't thought about this much. May be this is something you could contribute?

lsorber commented 4 years ago

I'm not experienced enough with this library to be able to contribute, unfortunately. In fact, I submitted this feature precisely because we would benefit from an example or docs to get started with that use case.

adriang133 commented 4 years ago

For classification problems, we already get probability distributions over the output, with a softmax layer essentially outputting a Categorical distribution.

Is there anything fancier we can do with probabilistic reasoning ? Maybe somehow also output a measure of uncertainty in our predictions ? Although uncertainty is already implied by the magnitude of the probabilities in some sense.

In other words, would a classification example be just a softmax layer added to any of the models ?

MrArnon commented 1 year ago

Hi all Do you have any updates on that topic? Or maybe an example of adding a softmax layer to a model?

elisim commented 1 year ago

Any news for a classfication example? I have troubles represnt classifcation data (e.g. JapaneseVowels) in GluonTS format.

In classifcation, the data is (instance, variable, time index), whereas in gluonTS forcasting is (instance, time index).

Any ideas?