Controlling Travel Path of Original Cobra

The appeal of original COBRA reduces due to its discrete structure in weight calculation, which increases the computational burden as it adapts a Grid search approach to choose its optimal parameters. A kernel-based ensemble learning [Guedj and Srinivasa Desikan (2020)] and its implementation through python [Guedj and Desikan (2017)] makes faster implementation of COBRA. However, it uses a different set of parameters independent of the threshold parameters of the original COBRA estimator that consists of a smooth weight function. In our paper, we plan to tame the original COBRA by choosing a suitable kernel that depends on the threshold parameter of the original COBRA. The methodology will help us to train faster than Grid search method for choosing its optimal parameters. In fact the methodology can help us to achieve much better the accuracy level than similar algorithms that take different weak learners like Ridge regression, Lasso, and Decision Tree inside the strategy. There is a wide variety of ensemble algorithms [Dietterich (2000), Giraud (2014), Shalev-Shwartz and Ben-David (2014)], with a crushing majority devoted to linear or convex combinations. A non-linear way of combining estimators is available by [Mojirsheibani (1999)].