report1/eval.tex

   1 \section{Conclusion}
   2 Detailed benchmarking and thorough comparison of the results still has to be
   3 done.
   4
   5 The object-based approach shows to heavily depend on the number of boxes.
   6 Describing the goal state for $n$ boxes needs a formula of $n!$ size. The size
   7 of the encoding also increases with the number of free blocks. When a screen is
   8 circular the parser also parses the blocks on the sides to be free spaces
   9 whereas they can never be reached. By improving the parser to disregard those
  10 positions we can increase the performance too.
  11
  12 The coordinate-based approach revealed that the following factors affect the
  13 feasibility of solving a particular \emph{sokoban} field:
  14 \begin{itemize}
  15   \item Overall size of a field
  16   \item Proximity of boxes to the man
  17   \item Number of boxes (to a lesser extent)
  18 \end{itemize}
  19 A possible improvement for the coordinate-based approach is the elimination of
  20 outer walls, which can reduce overall field sizes.\\
  21 In total, the material obtained by implementing two different ways to encode a
  22 \emph{sokoban} solver might give the us the ability to approach the next
  23 assignment in a more elaborate way.