poniard
poniard copied to clipboard
rxavier
Streamline scikit-learn model comparison.
OVR can be used to compute a curve for each class. See https://scikit-learn.org/stable/auto_examples/model_selection/plot_roc.html
Things to look out for - `y` type hints should be modified - `check_estimator_type` should be modified - `TargetEncoder` depends on `y` so it shouldn't be used - Maybe include...
Right now preprocessor + estimator pipelines are built during fit. This has two downsides - Can't be previewed without fitting - Make the whole "check if fitted ID exists to...
- Deepchecks - Fugue - Great Expectations - Custom data size reducer (see for example the [Amex Kaggle comp](https://www.kaggle.com/competitions/amex-default-prediction/discussion/328054))
Right now plugins are stored in a list, making it difficult to work with. A dict or named tuple, could be used, or a simple Bunch-like container.
However, within plugins it's also used for other things, making it potentially inconvenient. For example, Weights and Biases trains the model and logs it as an artifact in the project....