Base plugin

All plugins should subclass BasePlugin.

---

source

BasePlugin

 BasePlugin ()

Base plugin class. New plugins should inherit from this class.

BasePlugin defines a set of common plugin hooks that allow plugin classes to execute actions during the life of a Poniard estimator. These could be thought of hooks in callbacks in other libraries like Keras, Transformers or fastai.

We have named them plugins as callbacks in other libraries are generally not expected to significantly alter what the main code does, and instead add funcionality like logging, model saving, etc. Poniard plugins have no such restriction.

Developing plugins

Plugins allow devs to extend Poniard funcionality beyond what the main module offers. Doing so is straightforward: subclass BasePlugin and implement the desired methods.

Crucially, Poniard estimators inject themselves to all plugins during initialization, meaning that plugin instances have access to the estimator on the attribute _poniard.

The following minimal example builds a plugin that adds a new (useless) feature and modifies the preprocessor.

import numpy as np
import pandas as pd
from sklearn.feature_selection import VarianceThreshold

from poniard import PoniardClassifier

class StringFeaturePlugin(BasePlugin):
    """A plugin that adds a feature comprised of a single string.

    Parameters
    ----------
    string :
        The string to add as a feature.
    """

    def __init__(self, string: str):
        super().__init__()
        self.string = string

    def on_setup_data(self):
        data = self._poniard.X
        if hasattr(data, "iloc"):
            self._poniard.X = data.assign(**{self.string: self.string})
        else:
            self._poniard.X = np.append(data, self.string, axis=1)
        return

    def on_setup_preprocessor(self):
        old_preprocessor = self._poniard.preprocessor
        if isinstance(old_preprocessor[-1], VarianceThreshold):
            self._poniard.preprocessor = old_preprocessor[:-1]
            self._poniard.pipelines = self._poniard._build_pipelines()
        return


features = pd.DataFrame(
    np.random.normal(size=(20, 2)), columns=[f"X_{i}" for i in range(2)]
)
target = np.random.choice([0, 1], size=20)
pnd = PoniardClassifier(plugins=StringFeaturePlugin("foobar")).setup(features, target)
pnd.preprocessor

Target info
-----------
Type: binary
Shape: (20,)
Unique values: 2

Main metric
-----------
roc_auc

Thresholds
----------
Minimum unique values to consider a feature numeric: 2
Minimum unique values to consider a categorical high cardinality: 20

Inferred feature types
----------------------

	numeric	categorical_high	categorical_low	datetime
0	X_0		foobar
1	X_1

Pipeline(steps=[('type_preprocessor',
                 ColumnTransformer(transformers=[('numeric_preprocessor',
                                                  Pipeline(steps=[('numeric_imputer',
                                                                   SimpleImputer()),
                                                                  ('scaler',
                                                                   StandardScaler())]),
                                                  ['X_0', 'X_1']),
                                                 ('categorical_low_preprocessor',
                                                  Pipeline(steps=[('categorical_imputer',
                                                                   SimpleImputer(strategy='most_frequent')),
                                                                  ('one-hot_encoder',
                                                                   OneHotEncoder(drop='if_binary',
                                                                                 handle_unknown='ignore',
                                                                                 sparse=False))]),
                                                  ['foobar'])]))],
         verbose=0)

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.