Fast Ridge Regression

Fast Ridge regression with multiple ridge parameters and response variables.

ridge_fast is a high-performance MATLAB function designed to perform ridge regression with multiple response variables and multiple ridge penalty values—significantly faster than MATLAB’s built-in ridge function.

Key Features:

• Speed & Efficiency:Utilizes optimized linear algebra routines—Cholesky factorization for single (scalar) ridge penalties and a fully vectorized eigen-decomposition for multiple penalties—to deliver results in a fraction of the time required by the built-in function, especially when processing multiple ridge parameters.
• Versatility:Supports simultaneous processing of multiple response variables and multiple ridge penalty values. This removes the need for iterative loops and repetitive function calls, streamlining large-scale analyses.
• Accuracy:Produces highly accurate coefficient estimates, with the mean squared error (MSE) between ridge_fast and MATLAB’s ridge function (when properly scaled) on the order of 1e-30.
• GPU Compatibility:Accepts GPU arrays in single precision, which can further accelerate computations on large datasets when a supported GPU is available.
• Comprehensive Benchmarking:Includes example code to measure the speed and accuracy of the ridge regression, so you can easily compare performance and verify precision.