Upcoming Planner submission and test domains

[Mark "mak" Roberts]

Hi folks,

I highlight a few things I need answers to bold ... 

This is a friendly reminder that your planner submission is upcoming.  The deadline will be the end of the day anywhere on earth on Sunday 30 March 23:59 UTC-12 (for which most of us is Monday morning).  Please do your best to meet this deadline, because I cannot release the training distributions until I have received the final version of all the planners.  

I've selected the domains/generators, but I'm still refining parameter ranges on some of them.  I'm probably being too picky, but I want to give the instance-based learners a reasonable range while maintaining fairness.  For some domains, this is non-trivial and has required extensive validation experiments.   I remain hopeful that I'll have all the ranges ready in time.  However, I may need to refine some of them over the following week as the experiments guide this.  

As a back up plan, I propose we pipeline the ones I know are fine (you can start immediately on these problems), and provide a wider range for the ones I think may change (you can delay these a week or so until you have tighter bounds).  I need a response from every team on this:  Getting initial ranges for all domains on March 31 and refined ranges for some domains on or before 7 April [ will | will not ] hurt our approach (so long as we know which ones will not change).

Concerning planner setup.  Many of you followed the setup instructions for EC2 and sent a revision or other notes to handle issues.  That was much appreciated!!  I am planning to continue iterating on any setup issues going forward.  Two teams elected to stay on a 64bit platform despite the memory hit they will take -- I am allowing this but encourage everyone to try their best to use a 32bit version.

We haven't yet heard from Amazon EC2 yet.  If we hear from them in time, I may start the "No DCK" part of the runs as you develop your learning models.  Any objections to me running the "No DCK" experiments during the learning phase (assuming I'll rerun some if we spot a problem)?

Thanks,

mak

[Mark "mak" Roberts]

Hello IPCL Competitors,

I only heard from four of fourteen competitors concerning your final planner submission due a few hours ago.  If you intend for me to use the previous version you sent, I need to know.  Just respond to this message.  If I don't hear from you by the end of the day, I'll send individual emails.

Once I receive confirmation that I have your final planner version, I will email the tarball of training problems to your team.  For those of you who have already done so, you should be getting the tarball shortly.

I was able to settle on what I hope will be adequate ranges for each of the domains.  However, please let me know immediately if you spot a problem with the proposed ranges.

mak