rnkuhns / base_object Goto Github PK

To make the testing process more portable to other packages that use BaseObject, it should have a testing interface similar to the testing interface in sktime's BaseObject.

However, unlike sktime this should be made up of non-public methods by default. A mixin should also be provided to make it easy to add public methods that interface the corresponding non-public methods. This will give downstream users the choice in making the testing interface user facing or not.

It would be useful to think through whether it makes sense to include a base fittable estimator that inherits from BaseObject.

If this is decided upon, A key design point is whether to include a single BaseEstimator, or also try and add a BaseSklearnEstimator that inherits from BaseObject and also passes the sklearn user function for checking estimators (check_is_estimator?) and entirely adheres to their API.

Need to add unit tests that cover BaseObjects functionality.

In the github action workflows it is easiest to install all the required dependencies using requirements.txt or environment.yml file. However, it is not desirable to maintain these files separately from pyproject.toml.

A Python script (using toml or tomli) that reads pyproject.toml and outputs either requirements.txt or environment.yml would be useful. Ideally it would have a CLI interface and could be made into a pre-commit hook (think everytime pyproject.toml changes, this runs and generates the updated files).