You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
If you want to use a Darts is a lightweight environment, you can use the onnx format (check the code snippet at the end of issue #553). However, you will be responsible for reshaping/preparing the samples for inference and won't have access to the TimeSeries data-structure anymore.
Maintaining a separated version of the library for each model would actually not solve the problem since a lot of dependencies would remain (and it would be not realistically maintainable).
Feel free to fork Darts and remove all the features/models you don't need, hopefully this will reduce the requirements enough so that you can use our TFT implementation with your current constraints.
It is essential to have isolated versions for each model.
For example, I only need TFTModel in my project but I have to install ALL the models (>3GB!!).
This is horrible. I have to interrupt my entire workflow to adapt a project with just the parts I want.
The text was updated successfully, but these errors were encountered: