Environments with local scopes for modules in genetic programming.

Scoped environments are used in various programming languages to provide contexts within which a function's code can bind values and perform computations with them, without affecting bindings used by the rest of the program. The PushGP genetic programming system, which has produced state-of-the-art results in application areas, including software synthesis, allows programs to label sequences of instructions as modules. However, those modules can also modify bindings used by the rest of the program. This increases module coupling and might therefore make modular programs less evolvable than they otherwise could be. To rectify this, and to ensure that modules return single values, we implement scoped environments in PushGP, using a method that allows for the dynamic definition of arbitrary modules in a multi-type setting. We demonstrate that the use of scoped environments leads to an increase in the success rates for multiple software synthesis problems.