gogomaus 2002-06-02 22:55:26 | I noticed a strange behaviour, when getting started under version 4.11 . After 33 runs (just sent in manually) I found 6 runs with identical parameter values (at least integers are the same), but differing results from 1.247% to 1.475% !!! I cannot believe that same environment conditions will lead to some 20% deviation in muon yield. Did anyone notice comparable surprising results at the very beginning of his evolutionary runs ? What in the hell is varied, if not the 12 parameter values ??? |
Stephen Brooks 2002-06-03 10:22:35 | There are random effects in the decay of the muons into pions. Since we are dealing with a very small number of particles, the statistical variations are quite large. When I get the best design back again I simulate it several times to find out a more exact percentage. "As every 11-year-old kid knows, if you concentrate enough Van-der-Graff generators and expensive special effects in one place, you create a spiral space-time whirly thing, AND an interesting plotline" |
Michal Hajicek 2002-06-06 11:40:43 | I have same strange numbers in first ten v4.11 results. It was already explained, so I have another question: Why some parameters are simulated several times? And I want to know whether it will happen only at the beginning of evolution, or if this time wasting will continue. Really strange. I think it must be easy to control if some parameters were already used or not. |
Jwb52z 2002-06-06 13:57:41 | It's not "time wasting." When one is working with such complex calculations and such, it is almost impossible for the same result to happen twice even with the same variables because we're working with what I would consider a part of particle physics. What do you think Stephen? |
gogomaus 2002-06-07 01:31:14 | That was not the question of Michal. I think, we all agree that statistical variations will occur under these model simulation (I was just surprised about the largeness of it). Problem seems to be the genetic algorithm design chosen by Stephen. That tends to stick on the first fairly good result. Mutation, cross-over or linear combinations of parameter sets should lead to higher diversity. But it looks like if "parents" are selected from identical results, so evolution doesn´t really make good progress. Meanwhile I have nearly 200 runs, but the bunch is hopping around 3 lines (in terms of zrsc/rotrsc) again and again. (I do recommend the 3 D result viewer of Chris Johnson, which is a smart analyzing tool of your evolution.) As long as users have to repeat mediocre results too often, I consider it indeed as waste of time. |
Sokeo 2002-06-07 07:44:43 | Where can I get that 3-D viewer? Ya got a link? Thanks |
Sokeo 2002-06-07 10:20:00 | Never Mind I found it in the fourm. |
Stephen Brooks 2002-06-07 10:49:11 | quote: Oh. That's just reminded me that I forgot to stop the muon program from choosing two identical parents for a crossover or linear combination. I think I'll fix that in a minute. quote: Yes but you are only viewing 2 out of the 14 available dimensions. Maybe the parameter you are looking at isn't being as important as some of the 12 others that are not displayed? "As every 11-year-old kid knows, if you concentrate enough Van-der-Graff generators and expensive special effects in one place, you create a spiral space-time whirly thing, AND an interesting plotline" |
Michal Hajicek 2002-06-07 10:51:47 | Gogomaus is right about my question and also about very low diversity at the beginning.... And now, my second trial is comming: Why are exactly the SAME parameters used several times??? I think it's easy to check whether these parameters were already used or not. |
gogomaus 2002-06-10 00:19:12 | Stephen, does it mean I have to download muonv411.zip again ? And this time it will process a real evolution ? It´s a pity to have spent so much time on a single result optimizer, but at least we have learnt much about result variance... Unfortunately my home telephone device crashed on Friday and I have to wait for service ... |
Michal Hajicek 2002-06-11 09:54:26 | Gogomaus, it's a pitty in this case, but version 4.11 is version 4.11. No difference between old and new download, I compared it byte by byte. |
gogomaus 2002-06-11 10:21:21 | @ Michal and all others, seems like Stephen is too busy with exams, so his announcement of fixing that bug has not been realized up til now. Should be necessary to have a new version 4.12 or so... For the time being we have to live with a lot of repetitions... |