From marbles-isi@mailman.isi.edu Tue Jan 15 20:52:21 2002 From: marbles-isi@mailman.isi.edu (Martin Frank) Date: Tue, 15 Jan 2002 12:52:21 -0800 Subject: [Marbles-isi] Fwd: Re: did this paper get published? Message-ID: <5.1.0.14.2.20020115125049.00b26de0@tnt.isi.edu> What Wayne talked about was actually Marbles 1 Prime - same as Marbles 1 but assuming all tasks are of the same value. This is the problem that I suggested Seongbin and Alejandro submit a AAAI-2002 paper about. Cheers, yours, Martin >Date: Tue, 15 Jan 2002 14:48:56 -0600 >From: Weixiong Zhang >X-Mailer: Mozilla 4.76 [en] (Windows NT 5.0; U) >X-Accept-Language: zh-CN,en,zh-TW,zh >To: Martin Frank >CC: zhang@cs.wustl.edu >Subject: Re: did this paper get published? >X-AntiVirus: scanned by AMaViS 0.2.1 > > >Hi, Martin, > >How are you. Glad to hear from you. > >Yes, BERAP is just Marbles. I found Marbles is not informative so >... The paper can also be found at the CP01 workshop website. > >http://www.lsi.upc.es/~larrosa/Soft01.html > >We are still working on it, trying to solve large problem instances. I >will send you a copy of the paper that we are writing for your >comments. Please send me a copy of your paper. Thanks. > >-weixiong (wayne) zhang >Computer Science Department Tel: (314)935-8788 >Washington University Fax: (314)935-7302 >St. Louis, MO 63130 http://www.cs.wustl.edu/~zhang > > >Martin Frank wrote: > > > I would like to cite it. > > > > Also, is BERAP just a re-naming of our "Marbles 1" problem, or is it a > > different problem? > > > > Thanks Wayne. Cheers, yours, Martin > > > > > -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- > > Name: cp2001.pdf > > cp2001.pdf Type: Acrobat (application/pdf) > > Encoding: base64 From marbles-isi@mailman.isi.edu Tue Jan 15 21:07:24 2002 From: marbles-isi@mailman.isi.edu (Martin Frank) Date: Tue, 15 Jan 2002 13:07:24 -0800 Subject: [Marbles-isi] checked in: no more DCE_HOME_PCSYNTAX - _mostly_ good news Message-ID: <5.1.0.14.2.20020115130003.00ae8290@tnt.isi.edu> The current cygwin utilities version now recognizes PC path names such as "C:/code" so that it is no longer necessary to also enter a "//C/code" variant, the former suffices. I have changed the installation instructions and checked in numerous changes in Makefiles and shell files renaming DCE_HOME_PCSYNTAX to DCE_HOME. It is no longer necessary to define the DCE_HOME_PCSYNTAX and TMP_DIR variables to install our software. The good news: easier and less error-prone fresh installations. The bad news: ***** for existing installations, you may have to upgrade your CygWin installation after your next checkout if you have a CygWin version installed that is more than about 3 months old *****, so you may want to choose the time of your next cvs update carefully in this case... in this case do this: (1) un-set your DCE_HOME_PCSYNTAX environment variable (2) set your DCE_HOME environment variable to forward-slash PC syntax, such as C:/code (3) un-set your TMP_DIR environment variable I apologize for any pain... but the simpler set-up is a definite long-term benefit... Cheers, yours, Martin From marbles-isi@mailman.isi.edu Tue Jan 15 21:33:12 2002 From: marbles-isi@mailman.isi.edu (Seongbin Park) Date: Tue, 15 Jan 2002 13:33:12 -0800 Subject: [Marbles-isi] Fwd: Re: did this paper get published? In-Reply-To: Your message of "Tue, 15 Jan 2002 12:52:21 PST." <5.1.0.14.2.20020115125049.00b26de0@tnt.isi.edu> Message-ID: <200201152133.g0FLXCg23548@tnt.isi.edu> > What Wayne talked about was actually Marbles 1 Prime - same as Marbles 1 > but assuming all tasks are of the same value. This is the problem that I > suggested Seongbin and Alejandro submit a AAAI-2002 paper about. > > Cheers, yours, Martin > > Hi Martin, Thank you for forwarding the message. As I mentioned in your office, the only difference that I found is the inclusion of the penalty function in the formulation (i.e., penalty for unfilled tasks), but the problem structure is almost same as the Marbles 1. By formulating it into BERAP, they try to minimize the cost incurred by the penalties for unfilled tasks and they show that the decision problem version of BERAP (instead of the optimization problem) is NP-complete by reducing the known NP-complete problem(i.e., set packing problem) to the decision problem. As for the AAAI submission, (although, we don't have much time -- deadline is just several days away) we still have something to say about since the prediction formula gets us very precise numbers of tasks that can be schedulable and these estimated numbers are well backed up by experiments. Also, as you pointed out, if we assume that all tasks have same values, then we can even predict the "solution quality" from the predicted numbers right away since it will be just multiplication of the estimated number of task and the value for a task. Again, it would be nice if we could explicitly argue how that (the predictions) can help us design algorithms for the current problem since then it will be an example of "phase-transition aware" solvers. The situation seems a bit similar to "inverse optimization" in that we have an optimal solution and try to figure out what makes it optimal. Also, we do have some "more" information than the problem itself, so there might exist a way how we could exploit the information -- in other words, we are trying to solve a problem (Marbles problem) with the information saying that certain number of tasks can be "safely" schedulable (i.e., beyond that numbers are the regions of very difficult problem instances). I'm not sure whether this makes sense, but we might even consider the predictions as constraints in the sense that the problem now becomes "solve a Marbles problem subject to the constraint that the number of tasks <= predicted number of tasks". Or, we might want to cast the problem in such a way that "multi-resolution" type of solving might be possible; i.e., start the search space discretized initially and as we approach the number of predicted number of tasks, the algorithm explores a "finer" space than the original one. Well, I'm afraid that it sounds hand-waiving, but as we discussed in December, detecting the existence of phase transition is not sufficient and hope we can find a way to "embed" the phenomenon into algorithms so that they act sensitively to the phenomenon. Seongbin From marbles-isi@mailman.isi.edu Thu Jan 17 02:19:10 2002 From: marbles-isi@mailman.isi.edu (Martin Frank) Date: Wed, 16 Jan 2002 18:19:10 -0800 Subject: [Marbles-isi] checked in: marbles1 Message-ID: <5.1.0.14.2.20020116181435.00b11008@tnt.isi.edu> The experimentation setup now works correctly if all of the solvers exceed the time limits and if some of them exceed the time limits sometimes. (Previously threw exceptions or hung forever.) I'm nearly 100% happy with the experimentation setup for Marbles1 now, all that's missing is having the problem file name carried throuhg to the Excel output. I'll then work on introducing common APIs for the Marbles1 and Marbles2 problem definition, solution definition, and solver invocation so that Min can benefit from the new experimentation setup as well -- without having to copy and paste lots of code... Cheers, yours, Martin