Okay, I'll wait anyways until you have updated the pull request. I just wanted
to look at the code out of curiosity, since I also implemented this for the IPC.
Here are the results for an experiment evaluating the current patch.
http://ai.cs.unibas.ch/_tmp_files/seipp/issue773-v1-opt-issue773-base-issue773-v1-compare.html
The left comparison compares plain simple stubborn sets and simple stubborn sets with
min_pruning_ratio=0.2, i.e., pruning is diabled after 1000 expansions if not at least 0.2 of
the successors have been pruned (for simplicity the two configs use the same config and the
parser of the base version just ignores the unknown min_pruning_ratio argument). The right
comparison checks that the plain simple stubborn sets (without min_pruning_ratio) are not
affected by the patch.
Using min_pruning_ratio=0.2 raises the coverage of LM-cut by 11 tasks. 8 domains are affected
positively by the change while coverage decreases by 1 in Rovers. The patch doesn't seem to
affect the code path taken without min_pruning_ratio.