Paweł Wieczorek  committed 431a7a3


  • Participants
  • Parent commits bc3f651

Comments (0)

Files changed (1)

File doc/final/index.tex

 \subsection{Comparison within the lazy splitting group - Graph 4.}
+The full focusing prover always beats its less deterministic version, the weak focusing prover, as we expected,
+there is nothing to deep analysis.
+More interesting are differences to naive prover, which we suppose to be much worst than other provers.
+As mentioned in previous sections the tests Tensor with zeros and With with zeros are hard for our
+algorithms with lazy splitting due to non-deterministic choices.
+However, this disadvantage of basic lazy splitting does not explain why the Naive prover is almost twice
+better than the FullLS prover at the test Tensor commutativity. It shows defect of our implementation.
+The FullLS (and the WeakLS) prover tries to make possible inversions after each recursive call, but
+context is fully decomposed into atoms at beginning. It means that after each step whole context would be
+analysed for possible inversions. We can solve this problem by splitting input resources into two sets,
+one for already inverted hypotheses and second for not.