Reverse Engineering a Commercial Assessment Formula

Particular A.I. Methodology

  1. Genetic Programming manipulates the GP-trees as normal.

  2. Individuals are evaluated but rather than return an answer, the GP tree constitutes a set of instructions that manipulate pointers to memory. Upon evaluation the GP tree has produced a variable length vector. Each element of this vector is a real number that is bounded to be between 0.0 and 1.0.

  3. The problem in question is defined by means of grammatical rules that govern how the independent variables of the problem are to be combined together to achieve a candidate solution to the problem in question.

  4. The vector of real numbers guides the production of this grammar.

  5. Hence, this hybrid method combines all of the advantages and power of the standard Genetic Programming which have been tested and investigated in three books by John Koza as well as countless of other researchers, with the advantages of grammars that can overcome the needs of strongly typed GP. This is done without the complication of introducing a new GA type method as the Grammatical Evolution method and in this very real sense it is superior to the GE method.

  6. The method can easily abide by problem solution search with restrictions and constraints imposed by the user.

Numerical Example

Example of the Solution Interface

Example of User Specified Constraints

Evolution under some user constraints