treemind

treemind

treemind is a high-performance library for interpreting tree-based models. It supports regression, binary and multiclass classification, and handles both numerical and categorical features. By analyzing split intervals and feature interactions, treemind helps you understand which features drive predictions and how they interact making it ideal for model explanation, debugging, and auditing.

A formal research paper detailing the theoretical foundation of treemind is forthcoming.

Installation

Install treemind via pip:

pip install treemind

Key Features

• Feature Analysis Quantifies how individual features influence predictions across specific decision boundaries.

• Interaction Detection Detects and visualizes interaction effects between two or more features at any order n, constrained by memory and time.

• Optimized Performance Fast even on deep models thanks to efficient Cython-backed core.

• Rich Visualizations Interactive and static plots to visualize importance, split intervals, and interaction strength.

• Broad Model Support Compatible with xgboost, lightgbm, catboost, sklearn, and perpetual. Works with regression, binary, and multiclass tasks. Supports categorical features.

Algorithm & Performance

The treemind algorithm analyzes how often features and their combinations appear in decision paths, then summarizes their behavior over split intervals.

• Algorithm Overview
• Performance Benchmarks

Quickstart Example

This walkthrough shows how to use treemind.Explainer with a LightGBM model trained on the Breast Cancer dataset.

from lightgbm import LGBMClassifier
from sklearn.datasets import load_breast_cancer

from treemind import Explainer
from treemind.plot import (
    feature_plot,
    interaction_plot,
    interaction_scatter_plot,
    importance_plot,
)

# Load sample data
X, y = load_breast_cancer(return_X_y=True, as_frame=True)

# Train a model
model = LGBMClassifier(verbose=-1)
model.fit(X, y)

# Create an explainer
explainer = Explainer(model)

Count Feature Appearances

To see how often each feature (or feature pair) appears in the decision trees:

explainer.count_node(degree=1)  # Individual feature usage

| column_index | count |
|--------------|-------|
| 21           | 1739  |
| 27           | 1469  |

explainer.count_node(degree=2)  # Pairwise feature usage

| column1_index | column2_index | count |
|---------------|---------------|-------|
| 21            | 22            | 927   |
| 21            | 23            | 876   |

One-Dimensional Feature Analysis

Analyze how a single feature influences the model:

result1_d = explainer.explain(degree=1)

Inspect a specific feature (e.g., feature 21):

result1_d[21]

| worst_texture_lb | worst_texture_ub | value     | std      | count  |
|------------------|------------------|-----------|----------|--------|
| -inf             | 18.460           | 3.185128  | 8.479232 | 402.24 |
| 18.460           | 19.300           | 3.160656  | 8.519873 | 402.39 |

Feature Visualization

feature_plot(result1_d, 21)

Feature Importance

result1_d.importance()

| feature_0            | importance |
|----------------------|------------|
| worst_concave_points | 2.326004   |
| worst_perimeter      | 2.245493   |

importance_plot(result1_d)

Two-Dimensional Interaction Analysis

Evaluate how two features interact to influence predictions:

result2_d = explainer.explain(degree=2)
result2_d[21, 22]

| worst_texture_lb | worst_texture_ub | worst_concave_points_lb | worst_concave_points_ub | value    | std      | count  |
|------------------|------------------|--------------------------|--------------------------|----------|----------|--------|
| -inf             | 18.46            | -inf                     | 0.058860                 | 4.929324 | 7.679424 | 355.40 |

Interaction Importance

result2_d.importance()

| feature_0         | feature_1            | importance |
|------------------|----------------------|------------|
| worst_perimeter  | worst_area           | 2.728454   |
| worst_texture    | worst_concave_points | 2.439605   |

importance_plot(result2_d)

Interaction Plots

interaction_plot(result2_d, (21, 22))

interaction_scatter_plot(X, result2_d, (21, 22))

Contributing

Contributions are welcome! If you'd like to improve treemind or suggest new features, feel free to fork the repository and submit a pull request.

License

treemind is released under the MIT License. See the LICENSE file for details.