sevisal / RFF-BLR

RFF-BLR: Random Fourier Features Bayesian Linear Regression

Welcome to the RFF-BLR repository! This repository houses the implementation of the Random Fourier Features Bayesian Linear Regression (RFF-BLR) presented in the paper Bayesian learning of feature spaces for multitask problems.

Overview

In this repository, we delve into the world of advanced regression techniques with a focus on Bayesian inference and the Random Fourier Features. We provide an in-depth exploration through an ablation study, comparing various model extensions such as Bayesian formulation, Links, and Gamma optimization. Additionally, we conduct a rigorous comparison against different baselines using synthetic datasets, shedding light on the profound impact of RFF and its interaction with Bayesian Sparse formulations. Lastly, we offer a computational cost analysis, illuminating the efficiency of the RFF-BLR model in diverse scenarios using synthetic datasets.

Installation

To get started, follow these simple installation steps:

# Clone the repository
git clone https://github.com/sevisal/RFF-BLR.git

# Navigate into the cloned directory
cd RFF-BLR

# Install the required Python packages
pip install -r requirements.txt

Usage

This repository offers various scripts designed for different aspects of the study:

1. Baseline Comparison: Run the BaselineComparison.py script to initiate the baseline comparison:

   python BaselineComparison.py

   Witness the comparative performance of the RFF-BLR model against other baseline models using synthetic datasets.

2. Ablation Study: Execute the AblationStudy.py script to run the ablation study:

   python AblationStudy.py

   Dive into the realm of model extensions with functions meticulously crafted for each combination, exploring Bayesian formulations with gamma optimization or with RVFL-like links.

3. Computational Cost Analysis: To run the computational cost analysis, execute the CostAnalysis.py script:

   python CostAnalysis.py

   Experience a detailed analysis of computational costs, providing invaluable insights into the performance of the RFF-BLR model using synthetic datasets.