"Freecell" neural network heuristics

In areas, such as planning, state space searches are often conducted to find solutions. Usually, the heuristic is derived from knowledge of the domain. In many cases the knowledge of a domain is limited or the domain is so complex that an effective heuristic cannot be formulated. As an alternative, machine-learning techniques such as neural networks may be used to derive the heuristic. The game of Freecell was selected as a suitable benchmark domain, in which "knowledge based heuristics" and "neural heuristics" were employed to find solutions for randomly generated games. An amalgamation of the two, in which the neural network developed a heuristic from several knowledge based heuristics, was also used. Of the neural derived heuristics, the best-case architecture did not employ the "knowledge based heuristics". Moreover, neural heuristics were not able to improve upon those defined a priori.