Gravitational Wave Anomaly Detection using Normalised Autoencoders

This repository implements anomaly detection in gravitational waves using normalised autoencoders, inspired by the Manifold Projection Diffusion Recovery (MPDR) method. It provides scripts to train both standard and normalised autoencoders for detecting anomalies in gravitational wave signals.

---

Installation

Clone the repository and install the required dependencies:

pip install -r requirements.txt

Directory Structure

mpdr_physics/
├── dataset/                # Contains dataset-related files
├── models/                 # Autoencoder and MPDR model definitions
├── train/                  # Training scripts and configurations
├── utils/                  # Helper functions for preprocessing and evaluation
├── train_ae.sh             # Train a standard autoencoder
├── train_mpdr-r.sh         # Train normalised autoencoder (MPDR-r version)
├── train_mpdr-r_best.sh    # Optimized MPDR-r training script
├── train_mpdr-r_optuna.sh  # MPDR-r with hyperparameter tuning (Optuna)
├── train_mpdr-s.sh         # Train MPDR-s version
├── train_mpdr-s_best.sh    # Optimized MPDR-s training script
├── train_mpdr-s_optuna.sh  # MPDR-s with Optuna tuning
├── train_netx.sh           # Train a energy autoencoder
├── requirements.txt        # Required dependencies
└── LICENSE                 # License file