Implementation:Interpretml Interpret APLRClassifier And APLRRegressor

Knowledge Sources	Interpretml Interpret
Domains	Machine_Learning, Interpretability
Last Updated	2026-02-07 12:00 GMT

Overview

APLRRegressor and APLRClassifier are interpretable glassbox model wrappers around the Automatic Piecewise Linear Regression (APLR) algorithm, providing both global and local explanations within the InterpretML framework.

Description

This module wraps the external aplr package to provide APLR models that conform to the InterpretML explainer API. APLR builds piecewise linear models that are inherently interpretable, supporting both main effects and two-way interactions.

• APLRRegressor: Extends APLRRegressorNative (from the aplr package), RegressorMixin, and ExplainerMixin. Supports regression tasks with global feature importance visualizations (horizontal bar charts and per-term shape plots) and local per-instance contribution explanations.

• APLRClassifier: Extends APLRClassifierNative (from the aplr package), ClassifierMixin, and ExplainerMixin. Supports multi-class classification by internally using per-class logit APLR models. Provides global explanations broken down by class and local explanations for each instance.

• APLRExplanation: Custom explanation class extending FeatureValueExplanation that handles visualization for global summaries (bar charts of term importances), per-term shape functions (line plots for univariate, heatmaps for interactions), and local breakdowns.

The module also includes helper functions for density calculation (calculate_densities), feature name management (define_feature_names), unique value counting (calculate_unique_values), and local explanation value creation (create_values).

Usage

Use APLRRegressor for regression tasks where you need an inherently interpretable piecewise linear model with automatic term selection. Use APLRClassifier for classification tasks with the same interpretability properties. Both require the external aplr package to be installed (pip install aplr).

Code Reference

Source Location

• Repository: Interpretml_Interpret
• File: python/interpret-core/interpret/glassbox/_aplr.py

Signature

class APLRRegressor(APLRRegressorNative, RegressorMixin, ExplainerMixin):
    available_explanations = ["local", "global"]
    explainer_type = "model"

    def __init__(self, **kwargs):
    def fit(self, X, y, **kwargs):
    def explain_global(self, name: Optional[str] = None):
    def explain_local(self, X: FloatMatrix, y: FloatVector = None, name: Optional[str] = None):


class APLRClassifier(APLRClassifierNative, ClassifierMixin, ExplainerMixin):
    available_explanations = ["local", "global"]
    explainer_type = "model"

    def __init__(self, **kwargs):
    def fit(self, X, y, **kwargs):
    def explain_global(self, name: Optional[str] = None):
    def explain_local(self, X: FloatMatrix, y: FloatVector = None, name: Optional[str] = None):


class APLRExplanation(FeatureValueExplanation):
    def __init__(self, explanation_type, internal_obj, feature_names=None,
                 feature_types=None, name=None, selector=None):
    def visualize(self, key=None):

Import

from interpret.glassbox import APLRRegressor, APLRClassifier

I/O Contract

APLRRegressor

Constructor Inputs

Name	Type	Required	Description
**kwargs	varies	No	Keyword arguments passed to the underlying APLRRegressorNative constructor

fit Inputs

Name	Type	Required	Description
X	numpy array, pandas DataFrame, or list	Yes	Feature matrix (numeric values only)
y	numpy array	Yes	Target vector for regression
**kwargs	varies	No	Additional fit arguments; X_names can specify feature names

explain_global Outputs

Name	Type	Description
explanation	APLRExplanation	Global explanation with term importances and per-term shape functions

explain_local Outputs

Name	Type	Description
explanation	APLRExplanation	Local explanation with per-instance feature contribution breakdowns

APLRClassifier

Constructor Inputs

Name	Type	Required	Description
**kwargs	varies	No	Keyword arguments passed to the underlying APLRClassifierNative constructor

fit Inputs

Name	Type	Required	Description
X	numpy array, pandas DataFrame, or list	Yes	Feature matrix (numeric values only)
y	array-like	Yes	Target labels (converted to strings internally)
**kwargs	varies	No	Additional fit arguments; X_names can specify feature names

explain_global / explain_local Outputs

Name	Type	Description
explanation	APLRExplanation	Explanation object with global term importances per class or local per-instance breakdowns

Usage Examples

Regression Example

from interpret.glassbox import APLRRegressor
import numpy as np

X_train = np.random.randn(100, 5)
y_train = X_train[:, 0] * 2 + X_train[:, 1] + np.random.randn(100) * 0.1

model = APLRRegressor()
model.fit(X_train, y_train, X_names=["f0", "f1", "f2", "f3", "f4"])

# Global explanation
global_exp = model.explain_global(name="APLR Global")
global_exp.visualize(key=None)  # Overall feature importances

# Local explanation
local_exp = model.explain_local(X_train[:5], y_train[:5], name="APLR Local")
local_exp.visualize(key=0)  # First instance explanation

Classification Example

from interpret.glassbox import APLRClassifier
from sklearn.datasets import load_iris

X, y = load_iris(return_X_y=True)
model = APLRClassifier()
model.fit(X, y, X_names=["sepal_l", "sepal_w", "petal_l", "petal_w"])

global_exp = model.explain_global(name="APLR Classifier")
global_exp.visualize(key=None)

Related Pages

• Principle:Interpretml_Interpret_APLR_Regression