skll

skll copied to clipboard

Use pathlib

1

This is actually a bit involved. It would be best to start by tackling 1-2 functions/methods at a time. Find a function/method that uses a file path string and make...

mulhod

Add feature transformation and truncation to SKLL

3

It would be nice to have some more feature pre-processing functionality in SKLL, e.g., feature truncation and transformation sort of like what's in RSMTool. We could have a separate `preprocessing`...

desilinguist

Investigate a hyperparameter optimization framework like Optuna

1

Investigate whether using something like [optuna](https://optuna.org) is better than plain ol' grid search when it comes to the non-deep-learning algorithms that are in SKLL/scikit-learn.

desilinguist

This will be a pretty large issue. It might make sense to do a module at a time.

mulhod

Consider making solver and penalty grid-searchable for LogisticRegression

1

[`LogisticRegresssion`](https://scikit-learn.org/dev/modules/generated/sklearn.linear_model.LogisticRegression.html#sklearn.linear_model.LogisticRegression) now supports many solvers: Perhaps we should consider making the `solver` grid-searchable (and perhaps also the `penalty` option)?

desilinguist

See these refs: https://packaging.python.org/en/latest/guides/writing-pyproject-toml/ https://setuptools.pypa.io/en/latest/userguide/pyproject_config.html This is in-line with the latest python packaging recommendations.

desilinguist

Since ruff can do formatting now, may be we don't need black anymore.

desilinguist

Once numpy 2 has been out for a while, we should make sure that SKLL works with numpy 2 and drop support for numpy 1, unless there's some serious issues...

desilinguist

Use `None` as the default for "feature scaling" parameter

2

Do it in two stages, i.e. two releases. 1. The first shows a deprecation warning if "none" is specified. 2. Using "none" raises an error.

desilinguist

unpin numpy to use the latest version

1

Closes #760

damien2012eng