Message6121

Author florian
Recipients florian, gabi, jendrik, malte, salome, silvan
Date 2017-01-14.11:08:13
Content
Unfortunately, this was not the easy fix, I hoped it would be.

(still all results in one report, sorry)
http://ai.cs.unibas.ch/_tmp_files/pommeren/issue383-v1.html
http://ai.cs.unibas.ch/_tmp_files/pommeren/issue383-v1-total_time-lama-first-issue383-base-issue383-v1.png
http://ai.cs.unibas.ch/_tmp_files/pommeren/issue383-v1-total_time-seq-sat-lama-2011-issue383-base-issue383-v1.png

Coverage goes down by a lot, especially in miconic-fulladl, parcprinter and sokoban.
Interestingly, miconic-fulladl is the domain that looks really good in the
total_time
plots. It seems that a lot of tasks are now detected as unsolvable (see for example
http://ai.cs.unibas.ch/_tmp_files/pommeren/issue383-v1.html#error-miconic-fulladl).

Looks like this is the same issue mentioned in msg2522 and msg2543.

Another thing that surprised me in the results was that lama-first has the same
coverage as seq-sat-lama-2011. I thought lama-first is the first step of 
seq-sat-lama-2011. Are the other steps not helping, or just not showing up in
the report?

Maybe we have to let this issue rest until we can look at the landmark code as a
whole.
Let's hope its not another 3.5 years.
History
Date User Action Args
2017-01-14 11:08:13floriansetmessageid: <1484388493.88.0.911155167959.issue383@unibas.ch>
2017-01-14 11:08:13floriansetrecipients: + florian, malte, gabi, jendrik, silvan, salome
2017-01-14 11:08:13florianlinkissue383 messages
2017-01-14 11:08:13floriancreate