Nixtla's Open Source Time Series Ecosystem.

Nixtla Community

Ahh. Sorry for the noise. Source Code never lies. mlf_evaluation.models_ it is. Nevermind. Thx anyways for that great toolkit,.

Sorry for the confusion, we're working on a tutorial to access the trained models' parameters that will mention this.

<https://nixtla.github.io/mlforecast/docs/how-to-guides/analyzing_models.html|This> is the guide. please let us know if you feel something's missing

Hi, thx for that. I think that's helpfull.

It's really powerfull that NIXTLA just wraps the original modles. If you think that's desireable I could add another paragprag after the SHAP section where I use a Catboost model and plot the feature importance?  Just to bring home the point => you can access all the models and do the stuff you do with Pytorch/scikit-learn/catboost/lightGBM/XGBoost for model diagnostics?

Sure, go ahead. We use <https://nbdev.fast.ai/|nbdev> so you'll have to modify that notebook (nbs/docs/how-to-guides/analyzing_models.ipynb). We have <https://github.com/Nixtla/mlforecast/blob/main/CONTRIBUTING.md|these instructions> on how to contribute but feel free to reach out if you need any help