Issues Show Unassigned Show All Search
Login Remember me? Register Lost your login?
Help Roundup docs
Created on 2014-02-10.15:04:47 by silvan, last changed by silvan.
The final experiments with the configurations from the JACM 2014 paper can be found below. The following tables compare the baseline against our new "best" config for label reduction with dfp merging. http://ai.cs.unibas.ch/_tmp_files/sieverss/2014-04-25-compare-jacm-configs-rl-10k-100k-comp.html http://ai.cs.unibas.ch/_tmp_files/sieverss/2014-04-25-compare-jacm-configs-rl-200k-inf-comp.html http://ai.cs.unibas.ch/_tmp_files/sieverss/2014-04-25-compare-jacm-configs-cgl-comp.html The following tables compare the baseline against the implementation of the old label reduction in the new framwork, which makes it a bit slower overall. http://ai.cs.unibas.ch/_tmp_files/sieverss/2014-04-25-compare-jacm-configs-old-rl-10k-100k-comp.html http://ai.cs.unibas.ch/_tmp_files/sieverss/2014-04-25-compare-jacm-configs-old-rl-200k-inf-comp.html http://ai.cs.unibas.ch/_tmp_files/sieverss/2014-04-25-compare-jacm-configs-old-cgl-comp.html The code has been merged, marking this one as resolved. See also follow-up issue432 for more refactoring and improvements planned for the future.
We want to merge the recent improvements of the merge-and-shrink code, in particular the new label reduction and the framework for non-linear merge strategies.