lob-latent-regimes/Notebook/Reproducibility Note.md

Reproducibility Guide

This repository contains the complete pipeline for generating the synthetic Limit Order Book (LOB) data, detection signals, and evaluation metrics presented in the paper.

💻 Environment & Hardware

Experiments were conducted across the following environments:

• Primary Runtime: Google Colab (NVIDIA T4 GPU)
• Local Testing: Apple MacBook (M4 Architecture)

Note

All experiments are fully deterministic using specified random seeds. Minor variations may occur due to differences in hardware floating-point precision or library versions.

🛠 Prerequisites

Ensure you have Python 3.x installed. The core logic relies on the following stack:

• NumPy & SciPy (Numerical processing)
• scikit-learn (Evaluation metrics)
• hmmlearn (Hidden Markov Modeling)

🚀 Quick Start

To reproduce the paper's results, figures, and tables:

1. Install Dependencies:

   pip install -r requirements.txt
   

2. Execute Pipeline: Run the notebook cells sequentially. The pipeline is designed to be executed end-to-end to generate:
   • Synthetic LOB Data: Simulated market depth and flow.
   • Detection Signals: Primary output of the proposed model.
   • Metrics & Figures: All visualizations and tables used in the final publication.

---