Nick Erickson
Nick Erickson
*Issue #, if available:* *Description of changes:* - Improve AbstractTrainer Docs TODO: - [ ] Add unit tests to validate methods - [ ] Add validate methods for path, eval_metric,...
### Describe the issue linked to the documentation Currently the install instructions don't describe how to get to the point of being able to do: ``` pip install autogluon ```...
Add option for user-enabled bagging during blend-stacking (such as on `medium_quality`). This will allow users to have some models be bagged while others be single models (standard), at the same...
*Issue #, if available:* - Updated version of #1682 *Description of changes:* ### Added cascade ensemble PoC. The goal of a cascade ensemble is to expand the Accuracy - Inference...
*Issue #, if available:* *Description of changes:* - Allow calling fit_weighted_ensemble when refit models exist, previously crashed. - Allow automatically refitting new ensembles via refit_full arg. - Sped up fit_weighted_ensemble...
Currently, it is not easy for users to run the tutorials themselves such as https://auto.gluon.ai/stable/tutorials/tabular_prediction/tabular-quickstart.html We should provide an option for the user to download the python notebook itself.
Add a feature in Tabular which allows for model training to be interrupted after a given time has passed. Example: time_limits=100, but model has been training for 200 seconds and...
Add an option for users to create a new weighted ensemble after training that uses user-provided model weights. Example usage: ``` my_data = task.Dataset(...) weights = {'CatBoostClassifier_STACKER_l1: 0.8, 'RandomForestClassifierEntr': 0.2}...
Currently as of v0.4, the minimum torch version range is 1.0. This is likely too old and won't work, especially with the use of HuggingFace and PyTorch Lightning in `autogluon.text`....
Related issues: (Python 3.10: #1766) In `autogluon