Variable neighborhood programming for symbolic regression

In the field of automatic programming (AP), the solution of a problem is a program, which is usually represented by an AP-tree. A tree is built using functional and terminal nodes. For solving AP problems, we propose a new neighborhood structure that adapts the classical elementary tree transformation (ETT) into this specific AP-tree. The ETT is the process of removing an edge and adding another one to obtain a new feasible tree. Experimental comparison with reduced VNP, i.e., with VNP without local search, genetic programming, and artificial bee colony programming shows clearly advantages of the new proposed BVNP method, in terms of speed of convergence and computational stability.

Automated summary

In the field of automatic programming, an AP-tree is used to represent a program for solving AP problems. A new neighborhood structure, based on the elementary tree transformation, is proposed to adapt the AP-tree. Experimental comparison shows clear advantages of the new BVNP method in terms of convergence speed and computational stability.