NPROS: A Not So Pure Random Orthogonal Search Algorithm –A Suite of Random Optimization Algorithms Driven by Reinforcement Learning.
Published in: Optimization Letters Journal, Springer Publication [SCI Indexed]. Link to paper: https://link.springer.com/article/10.1007/s11590-023-02038-0
PDF of the FULL paper available at: https://rdcu.be/dgCof
NPROS is a suite of algorithms that is an elementary/easy-to-use, Random Optimization based global optimization algorithm. It is an improved/modified/elite version of the PROS (Pure Random Orthogonal Search) algorithm. The original PROS algorithm uses uniform distribution to sample new locations/points in the function domain, whereas our proposed NPROS suite uses Normal and Lognormal distribution. NPROS-1 smartly manipulates the mean and standard deviation parameters of the probability distribution to bring a flavour of 'Exploitation' to the otherwise purely explorative PROS search.
NPROS-2 is an RL based intelligent algorithm, which uses, a Multi-Armed Bandit and decaying epsilon strategy to learn and adopt the most suitable probability distribution automatically without any prior knowledge about the underlying objective function. NPROS-2 exhibits the possibility of incorporating RL strategies at an algorithmic level into optimization algorithms to make intelligent OP algorithms that can 'learn and adapt'. Link to original PROS paper: https://www.mdpi.com/2076-3417/11/11/5053
How to Cite: Hameed, A.S.S.S., Rajagopalan, N. NPROS: A Not So Pure Random Orthogonal search algorithm—A suite of random optimization algorithms driven by reinforcement learning. Optim Lett (2023). https://doi.org/10.1007/s11590-023-02038-0
How to Use: There are two ipython jupyter notebook in this repository. No special prerequisite packages are required. The notebook can be downloaded and executed or the code can be simply copied.
An experiment-ready version titled as: "NPROS (Experiment Ready Version).ipynb". An easier-to-understand version titled as: "NPROS Suite of Algorithms.ipynb", with detailed comments explaining the proposed NPROS algorithm suite. For Any suggestions or doubts mail to: shahulshan81@gmail.com Cite the paper, if you find it useful.