blnetwork 0.1.0

Behavior Learning (BL): Learning Hierarchical Optimization Structures from Data

Behavior Learning (BL)

Behavior Learning (BL) is a general-purpose machine learning framework grounded in behavioral science. It unifies predictive performance and intrinsic interpretability within a single modeling paradigm. BL learns explicit optimization structures from data by parameterizing a compositional utility function built from interpretable modular blocks. Each block represents a Utility Maximization Problem (UMP), a foundational framework of decision-making and optimization. BL supports architectures ranging from a single UMP to hierarchical compositions, enabling expressive yet structurally transparent models. Unlike post-hoc explanation methods, BL provides interpretability by design while maintaining strong empirical performance on high-dimensional tasks.

Installation

blnetwork can be installed via PyPI or directly from GitHub.

Pre-requisites:

Python 3.10.9 or higher
pip

For developers

git clone https://github.com/YueLiang-hye/Behavior-Learning.git
cd blnetwork
pip install -e .

Installation via github

pip install git+https://github.com/YueLiang-hye/Behavior-Learning.git

Installation via PyPI:

pip install blnetwork

Requirements

# python==3.10.9
torch>=2.2
numpy>=1.26
pandas>=2.0
scikit-learn>=1.3

After activating the virtual environment, you can install specific package requirements as follows:

pip install -r requirements.txt

Optional: Conda Environment Setup For those who prefer using Conda:

conda create --name blnetwork-env python=3.10.9
conda activate blnetwork-env
pip install git+https://github.com/YueLiang-hye/Behavior-Learning.git  # For GitHub installation
# or
pip install blnetwork  # For PyPI installation

Computation Requirements

BL is implemented in PyTorch and supports both CPU and GPU training.

• Small-scale tabular examples run on a single CPU within a few minutes.
• High-dimensional settings may benefit from GPU acceleration (e.g., NVIDIA L40).

For most tabular tasks, CPU training is sufficient.

Examples

Start with the notebooks in examples/:

• Example 1: Boston Housing (continuous)
• Example 2: Breast Cancer (classification)

Advice on hyperparameter tuning

In many cases, BL can achieve comparable (or slightly better) performance than an MLP baseline using roughly one third of the hidden width.

Other hyperparameters can be initialized based on standard MLP tuning, and then refined for the specific task.

Contact

If you have any questions, please contact [e-mail]