Interpret-Core

Warnings

• breaking The shape of the `bags` parameter in EBMs was changed from (n_outer_bags, n_samples) to (n_samples, n_outer_bags). In v0.7.0, the old format issued a warning and was accepted, but this behavior may be fully deprecated or removed in future major versions.
  Fix: Ensure the `bags` parameter is passed with the shape (n_samples, n_outer_bags) to align with `X` parameter's shape.
• breaking The `ComputeProvider` abstraction was removed in v0.7.3, simplifying the interface. Code relying on this abstraction will break.
  Fix: Refactor code to remove dependencies on `ComputeProvider`. Refer to the latest documentation for the updated simplified interface.
• gotcha Older versions of `interpret-core` (prior to v0.7.4) had incompatibilities with `scikit-learn` versions 1.8 and above, specifically due to changes in `is_classifier` and `is_regressor` only accepting valid estimators.
  Fix: Upgrade `interpret-core` to v0.7.4 or newer to resolve `scikit-learn` compatibility issues. Alternatively, pin `scikit-learn` to a version below 1.8 if an upgrade is not possible.
• gotcha When using NumPy arrays as input to explainers, feature names might not appear in visualizations. This occurs because NumPy arrays lack inherent column names.
  Fix: Pass a Pandas DataFrame with meaningful column names, or explicitly set the `feature_names` property when initializing or calling explain functions for array inputs.
• gotcha For `ExplainableBoostingClassifier`, the y-axis values in the generated global explanation graphs are in 'logit' space, not direct probabilities. This requires careful interpretation for classification tasks.
  Fix: Remember that the y-axis represents log-odds (logits). Positive values push towards the positive class, but transformations are needed to convert to probabilities if desired for direct comparison.

Install

• pip install interpret-core Basic installation
• pip install interpret-core[required,ebm,plotly,dash] Installation with common optional dependencies (EBM, visualizations)

Imports

• ExplainableBoostingClassifier

  from interpret.glassbox import ExplainableBoostingClassifier

• ExplainableBoostingRegressor

  from interpret.glassbox import ExplainableBoostingRegressor

• show

  from interpret import show

  The `show` function for visualizations is directly available from the top-level `interpret` package, not a submodule like `visualize`.

Quickstart

This quickstart demonstrates how to train an Explainable Boosting Machine (EBM) for a classification task and generate both global and local explanations. It covers data preparation, model fitting, and accessing explanation data programmatically. For interactive visualizations, `show(explanation_object)` would be used in a compatible environment like a Jupyter notebook.

import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
from interpret.glassbox import ExplainableBoostingClassifier
from interpret import show

# Generate some synthetic data
np.random.seed(0)
X = pd.DataFrame({
    'feature_a': np.random.rand(100) * 10,
    'feature_b': np.random.randint(0, 3, 100).astype(str),
    'feature_c': np.random.randn(100)
})
y = (X['feature_a'] + (X['feature_b'].astype(int) * 2) + np.random.randn(100) > 10).astype(int)

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Initialize and fit an Explainable Boosting Machine (EBM)
ebm = ExplainableBoostingClassifier(random_state=42)
ebm.fit(X_train, y_train)

# Get global explanations (feature importances and shapes)
ebm_global = ebm.explain_global()

# In a notebook environment, you would call show(ebm_global) to visualize
# For script execution, we can print summary or a representation of the explanation
print(f"Global Explanation for EBM:\n{ebm_global.data()}")

# Get local explanations for a specific sample
ebm_local = ebm.explain_local(X_test[:1], y_test[:1])
print(f"\nLocal Explanation for first test sample:\n{ebm_local.data()}")