This paper presents a redesigned and modular implementation of the Adaptive Hybrid Feature Selection (AHFS) algorithm in a Python environment, originally developed in MATLAB. The new version incorporates GPU arithmetic support, extensibility for custom measures, and dynamic CPU/GPU switching for efficient computation. Extensive benchmarking on 25 datasets shows that the Python version consistently improves both runtime and feature selection quality. A novel filter component, Fast Correlation-Based Filter in Pieces (FCBFiP), was also integrated and evaluated alongside the already existing selection measures. Results indicate that the redesigned framework enhances scalability and scoring precision while preserving the core behavior of the original algorithm.