skbase

0.13.1 · active · verified Sat Apr 11

skbase provides base classes for creating scikit-learn-like parametric objects, along with tools to make it easier to build custom packages that follow these design patterns. It is a foundational library, notably used by `sktime`. The current version is 0.13.1, and it has a frequent release cadence with minor and patch updates occurring roughly monthly or bi-monthly.

Warnings

breaking The meaning of `filter_tags` arguments (type `str` or `iterable of str`) in `skbase.lookup.all_objects` changed. Previously, it selected objects possessing the tag(s) of any value. From version 0.9.0, it selects objects where the tag's value is explicitly `True` (boolean).
Fix: To retain the old behavior (selecting objects with the tag of any value), pass a dictionary with the tag name as key and `re.Pattern('.*')` as value. For example, `from re import Pattern; all_objects(filter_tags={'tag_name': Pattern('.*')})`.
gotcha As of version 0.4.2, `skbase` explicitly removed all core external PyPI dependencies, intending to be a truly base module without implicit requirements beyond Python itself. Developers building on `skbase` should be aware of this and explicitly manage any dependencies their custom objects require.
Fix: Ensure all necessary dependencies for your project are explicitly listed in your `pyproject.toml` or `setup.py` and installed, rather than relying on `skbase` to pull them in indirectly.
gotcha Users often confuse `skbase` with `scikit-learn`. `skbase` provides base classes for *creating* scikit-learn-like parametric objects and design patterns, but it is not `scikit-learn` itself and does not contain machine learning algorithms for direct use. Importing `sklearn.base.BaseEstimator` is for `scikit-learn`, not `skbase.base.BaseObject`.
Fix: If you intend to use existing machine learning algorithms, use `scikit-learn` (e.g., `from sklearn.ensemble import RandomForestClassifier`). If you are building new estimators that adhere to the `scikit-learn` API, use `skbase.base.BaseObject`.

Install

pip install scikit-base Install core package
pip install scikit-base[all_extras] Install with all optional dependencies

Imports

BaseObject
```
from skbase.base import BaseObject
```
skbase provides its own BaseObject for creating scikit-learn-like classes; it is not the same as scikit-learn's BaseEstimator.
all_objects
```
from skbase.lookup import all_objects
```
Used to retrieve registered skbase objects, often filtered by tags.

Quickstart

This quickstart demonstrates defining a custom object inheriting from `skbase.base.BaseObject`, illustrating how to initialize parameters, retrieve them with `get_params`, modify them with `set_params`, and create a deep copy using `clone()`.

from skbase.base import BaseObject

class MyCustomObject(BaseObject):
    """A simple custom object demonstrating skbase.base.BaseObject."""

    def __init__(self, value_a=1, value_b="default_string", random_state=None):
        self.value_a = value_a
        self.value_b = value_b
        self.random_state = random_state
        super().__init__()

# Create an instance of our custom object
my_obj = MyCustomObject(value_a=10)
print(f"Initial parameters: {my_obj.get_params()}")

# Modify parameters using set_params
my_obj.set_params(value_b="new_string_value")
print(f"Parameters after set_params: {my_obj.get_params()}")

# Demonstrate cloning (a common scikit-learn-like pattern)
cloned_obj = my_obj.clone()
print(f"Cloned object parameters: {cloned_obj.get_params()}")
print(f"Is cloned_obj the same instance as my_obj? {cloned_obj is my_obj}")

view raw JSON →