Reference Code for the Paper: "Unsupervised Novelty Detection Methods Benchmarking with Wavelet Decomposition"

This repository contains the reference code for the paper titled "Unsupervised Novelty Detection Methods Benchmarking with Wavelet Decomposition", which will be published as a proceeding in the 8th International Conference on System Reliability and Safety (ICSRS), held in Sicily, Italy, from November 20-22, 2024.

• Preprint link: https://arxiv.org/abs/2409.07135
• Conference link: https://www.icsrs.org/
• Author contact: https://orcid.org/0009-0006-2741-8054

Setup Instructions for Linux-Based Machine

Follow these steps to set up and run the project on a Linux-based system.

Prerequisites

Before starting, ensure you have the following installed:

1. Git: Download and install Git by following instructions from the github official website.

2. Docker: Install Docker from the docker official website.

3. PDF Viewer: Ensure you have a PDF viewer installed to open PDF files.

Step-by-Step Instructions

1. Pull the Docker Image: Download the pre-built Docker image.

   docker pull arielpriarone/unsupervised-novelty-detection-methods-benchmarking-with-wavelet-decomposition:latest

2. Tag the Docker Image: Tag the downloaded image for easier reference.

   docker image tag arielpriarone/unsupervised-novelty-detection-methods-benchmarking-with-wavelet-decomposition:latest undbwwd:latest

3. Clone the Git Repository: Clone the repository containing the project code.

   git clone https://github.com/PIC4SeR/Unsupervised-Novelty-Detection-Methods-Benchmarking-with-Wavelet-Decomposition.git UNDBwWD

4. Navigate to the Project Directory: Change the directory to the cloned repository.

   cd UNDBwWD

5. Run the Docker Container: Start the container with necessary settings.

   docker run -it -v $(pwd):/code --runtime=nvidia --gpus all undbwwd:latest

   now the container should be running. Please don't exit and execute the following steps 6 to 12 inside the container.

6. Create Database Directory: Set up a directory for MongoDB.

   mkdir -p /data/db

7. Set Directory Permissions: Change the ownership of the directory.

   chown `id -u` /data/db

8. Start MongoDB: Run MongoDB in the background.

   mongod --fork --logpath /var/log/mongodb/mongod.log

9. Navigate to the Code Directory: Go to the code folder inside the container.

   cd code

10. Run the Demo Script: Execute the demo script.

python3 demo.py

11. Go to the Results Directory: Navigate to the results directory.

cd results

12. Compile the LaTeX File: Use pdflatex to generate the table.

pdflatex table.tex

13. Exit the Container: Press Ctrl+D to exit the Docker container.

14. Open the Results: Use a PDF viewer to open the results.

    xdg-open results/Correlation_table.pdf
    xdg-open results/Fig5.pdf
    xdg-open results/Fig7.pdf
    xdg-open results/table.pdf

---

Follow these steps to set up the environment, run the project, and view the results.

Customizing Script Settings

The demo.py script includes several configuration flags that allow you to customize its behavior. You can modify these settings to control which steps are executed when running the script.

To change the script settings, open the demo.py file in a text editor and locate the following section:

load_timeseries_flag    = True	# if True, the timeseries are loaded to MongoDB
run_FA_flag             = True	# if True, the feature agent is run
train_flag              = False	# if True, the models are trained
NormalizeNovMetric      = True	# if True, the novelty metric is normalized
Generate_PCA            = False	# if True, the PCA models are generated
Optimize_Autoencoder    = False	# if True, the Autoencoder models are optimized
Fit_Autoencoder         = False	# if True, the Autoencoder models are fitted
plot_patches            = False	# if True, the variance is visualized as a patch in the plot

Configuration Flags Description

• load_timeseries_flag:

  • Purpose: Loads the time series data into MongoDB.
  • Set to True: If you want to load the time series data. Useful for the initial setup.
  • Set to False: If the data is already loaded and you want to skip this step.

• run_FA_flag:

  • Purpose: Executes the feature agent process.
  • Set to True: To run the FA, which is necessary for generating feature sets.
  • Set to False: To skip the FA process the features are already loded in the MongoDB database

• train_flag:

  • Purpose: Trains the machine learning models used in the analysis.
  • Set to True: If you need to train the models from scratch.
  • Set to False: If you prefer to use pre-trained models.

• NormalizeNovMetric:

  • Purpose: Normalizes the novelty metric for consistent scaling.
  • Set to True: To normalize the novelty scores.
  • Set to False: To use raw novelty scores without normalization.

• Generate_PCA:

  • Purpose: Generates Principal Component Analysis (PCA) models.
  • Set to True: If you need to create PCA models for dimensionality reduction.
  • Set to False: If PCA models are already generated.

• Optimize_Autoencoder:

  • Purpose: Optimizes the hyperparameters of Autoencoder models.
  • Set to True: To perform hyperparameter optimization on Autoencoders.
  • Set to False: To skip optimization and use existing parameters.

• Fit_Autoencoder:

  • Purpose: Fits the Autoencoder models to the data.
  • Set to True: If you need to fit the Autoencoders after optimization.
  • Set to False: If the Autoencoders are already fitted.

• plot_patches:

  • Purpose: Visualizes variance as patches in the generated plot.
  • Set to True: To include variance patches in your visualizations.
  • Set to False: To generate plots without variance patches.