HWM Documentation

The hwm package offers a powerful and flexible implementation of the Adaptive Hammerstein-Wiener Model (HWM), designed to handle dynamic system modeling in intelligent computing environments. By integrating both linear and nonlinear dependencies, the HW model is capable of effectively capturing complex, time-dependent patterns in sequential data.

The package is particularly suitable for advanced applications such as network intrusion detection, financial forecasting, industrial automation, and other tasks where accurate predictions and adaptability to dynamic systems are essential. The model’s architecture supports both regression and classification tasks, making it versatile for a variety of machine learning challenges.

This documentation will guide you through installing and using the hwm package, and provide detailed explanations of its components and functionalities.

Overview

The Adaptive Hammerstein-Wiener (HW) Model is an extension of traditional Hammerstein-Wiener systems, optimized for machine learning tasks. It consists of the following components:

• Nonlinear Input Transformation: This module encodes the complex relationships between input features. It applies a nonlinear transformation to the input data to model its intricate characteristics before passing it to the linear dynamic block. This step is crucial for capturing the nonlinearities in real-world data.

• Linear Dynamic System Block: The linear dynamic system block captures the temporal dependencies across multiple time steps. It models how current states depend on past states, facilitating the modeling of dynamic relationships in sequential data. This block can handle both stationary and non-stationary data and is a core feature for time series forecasting.

• Nonlinear Output Transformation: After the linear dynamic block, the output is transformed through a nonlinear mapping to produce task-specific predictions. This ensures that the model can adapt to various types of outputs, such as categorical labels (for classification) or continuous values (for regression).

The unique architecture of the HW model is designed to deliver both high interpretability and computational efficiency. It is particularly useful in scenarios where accurate predictions and temporal stability are required, making it well-suited for real-time applications.

Additionally, the package includes custom evaluation metrics:

• Prediction Stability Score (PSS): hwm.metrics.prediction_stability_score() Measures the consistency of the model’s predictions over time, assessing its robustness in dynamic environments.

• Time-weighted Accuracy (TWA): hwm.metrics.twa_score() Evaluates the model’s accuracy while taking into account the timing of predictions, ensuring that the model performs well in time-sensitive tasks.

Exploration

Table of Contents:

• Installation Guide
• User Guide
• hwm API Reference
• Compatibility Utilities
• Datasets
• Dynamic System Estimators
• Utils
• Metrics
• Examples and Tutorials
• Package Architecture
• Development Guide
• Release Notes

Navigation

The hwm package is divided into several subpackages, each serving a specific function to ensure the model’s proper operation in various machine learning workflows:

• installation: This subpackage provides all the necessary instructions to install and configure the hwm package. It includes detailed steps for setting up dependencies and environment requirements, ensuring users can get started quickly and without hassle. It also includes troubleshooting tips for common installation issues.

• examples: The examples subpackage includes a collection of scripts designed to demonstrate the HW model’s application across multiple domains. Examples cover tasks such as network intrusion detection, time series forecasting, and industrial system monitoring, showcasing how the HW model can be used to solve real-world problems.

• datasets: The hwm.datasets subpackage includes utilities for managing datasets, including data loaders and preprocessors. The primary file, _datasets.py, provides functions for loading common datasets, handling missing data, and preprocessing input features. It is designed to make it easier to work with large datasets and integrate them with the HW model for training and evaluation.

• estimators: The hwm.estimators subpackage contains the core model classes that implement the HW model’s nonlinear and dynamic processing blocks. The files dynamic_system.py define the internal structure of the model, including the nonlinear input transformation, linear dynamic system, and nonlinear output transformation. These estimators can be customized for specific applications, such as classification or regression tasks, and are the heart of the model’s predictive capabilities.

• utils: The hwm.utils subpackage provides a variety of general-purpose utility functions that support the overall functionality of the hwm package. Key modules include:

  • hwm.utils.context : Manages the context in which the model is run, helping to organize and track the model’s state during training and evaluation.

  • hwm.utils.validator : Validates input data, ensuring that the data used for training or testing the model is correctly formatted and meets the necessary requirements.

  • Other utility functions support tasks like logging, error handling, and model evaluation.

• metrics: The hwm.metrics subpackage defines custom evaluation metrics that are specifically designed for the HW model. These include:

  • Prediction Stability Score (PSS): A metric that measures the consistency of the model’s predictions over time, highlighting the model’s ability to handle dynamic, time-varying data.

  • Time-weighted Accuracy (TWA): A metric that evaluates the model’s time-sensitive performance, giving more weight to accurate predictions made during critical time windows.

Indices and Tables

The following resources are available to help you navigate the hwm documentation:

• Index

• Module Index

• Search Page