Merged.

Thanks! Also looking at the other merge strategies, I'm even less worried than
before about the performance change, so I didn't look at the runs and think this
is ready to merge.

Sorry for not removing the comments, I intended to clean them up before merging.
I also forgot to post the dfp and rl configurations. I updated all reports to
include the other attributes, so here are all three reports:

http://ai.cs.unibas.ch/_tmp_files/sieverss/2014-10-02-issue483-compare-base-v1-cggl-comp.html
http://ai.cs.unibas.ch/_tmp_files/sieverss/2014-10-02-issue483-compare-base-v1-dfp-comp.html
http://ai.cs.unibas.ch/_tmp_files/sieverss/2014-10-02-issue483-compare-base-v1-rl-comp.html

You can find the runs in
/infai/sieverss/experiments/executed/issue483/2014-10-02-issue483-v1.

Looks like a nice cleanup! :-) I'll add some comments on bitbucket before we merge. 

Also, I'd like to see a bit more about the changing tasks in the experiments.
Can you let me know where the run files are on maia and add the attributes for
expansions etc. to the report? (The score_... ones are already there, which is
great for the summary, but for looking at individual tasks I sometimes like to
see more from the original numbers of expansions etc.)

Sure, https://bitbucket.org/SilvanS/fd-dev/branch/issue483

Hi Silvan, can you give me another link to look at the code? I'm curious what
the change looks like. (The last one kind of got lost in the deadline storm...)

I think this should be ready for merge.

http://ai.cs.unibas.ch/_tmp_files/sieverss/2014-10-02-issue483-compare-base-v1-cggl-comp.html

I thought way too complicated :-) I forgot that "apply_abstraction", given an
equivalence relation of states, takes care of the transitions. So I simply
collect all to be pruned states and apply the abstraction to the transition
system. Experiments will soon be running.

Forgot to mention it: this is part of the meta issue432.

We want to have a separate method discard_states() which removes all unreachable
and irrelevant states from a transition system. Furthermore, this method should
not use a default shrinking strategy to do so. As this is something which could
impact performance, I dedicate an own issue for it.

Malte, do you have a special implementation of the removal in mind? All I can
think off to start with is to iterate over all transitions and to remove those
leading to/starting in unreachable/irrelevant states. The question then is how
to deal with the removal of states. This is something where we need to "rename"
the remaining states, something which the shrinking strategies normally take
care of. Should we manually compute an equivalence relation which we can pass to
"apply_abstraction"? I think I'll just start with that idea.