 |
| Herb[Romulus2] 2004-05-08 04:01:18 | Though we haven't even maxed out the last breed, the latest top value 10.116106 (1438.5 Mpts) [v4.34] represents a new breed with a wide variation from what we had before. It would make sense to run an isolated results.dat with this value only. Go cracking  |
| excaliber[Free-DC.org] 2004-05-08 07:02:46 | Ooohhh, I like doing that kind of stuff. I'm on it  |
| Herb[Romulus2] 2004-05-09 04:44:30 | I just send in a 10.169632 which is an ascent of the new breed, but again diverted far from the previous one to form possibly an own line. |
| Stephen Brooks 2004-05-10 01:01:59 | What we may have is the point where the new one first appeared, and all the old results, and some of these additional new ones that have been triggered are interpolations between those two regions. |
| Herb[Romulus2] 2004-05-10 07:18:41 | Yes, them interpolations create good new values. The really interesting thing at the moment is we have 3 (not yet sure about a 4th) successful breeds climbing fast. Since I'm watching the variations (about 7 month or so) we had always one dominating breed, which pushed the previous one away just by the time. This time is much different, there are so many new combinations of breeds possible at once. I'm really curious when this will settle and at what level A lot of fun after a bit lame period now |
| kitsura 2004-05-10 09:28:25 | What I'm curious to know is where these new breeds are coming from. Do they branch of from those top yields or are newly evolved? And would you suggest that everyone run the client with empty result.dat or with the sample one? |
| Herb[Romulus2] 2004-05-10 10:24:16 | The new top value 10.176758 differs with 141 points from the new breed in the top post. Formerly I would have said , that's a new breed, but look: - my 10.169632 differs by 251 points - I suspect an own breed at 10.130496 with a difference of 969 points - I suspect another breed at 10.152853 with a difference of 1015 points You may think, hey them last 2 are pretty close together, in fact they differ each other by 328 points. - an "oldie" already with 9.933922 and 1401 points difference and the closest difference at 1017 points to anything else in the rechecked lists - another "oldie" with 9.890808 and 1070 points difference and 1030 points difference to anything else The last two don't seem to improve anymore, but may still serve as interpolation trickle base. In most cases you can see where values are coming from, but that one from the first post, just suddenly appeared and still is in power. Clear as mud, eh |
| kitsura 2004-05-11 02:42:22 | Very weird, I don't know how to analyse the results even though I've seen your Excel file before. Anyway I'll just continue evolving my results to see if I hit anything interesting. |
| Stephen Brooks 2004-05-12 08:40:59 | I think I can see your selection of new breeds appearing on this multi-headed comet-like thing: |
| [TA]z 2004-05-12 09:03:12 | whoa, very nice |
| Herb[Romulus2] 2004-05-12 21:11:17 | Yeah cool, and it looks like as really several hot spots started up there |
| kitsura 2004-05-13 00:18:10 | Sorry how do you interpret the graph? The colors represent the muon % right? but how do you tell if its a new breed or not? |
| Herb[Romulus2] 2004-05-13 01:37:28 | The colours represent the number of results in the relation of percentage to particles. These new hot spots (like an infrared picture) in the upper scale will hopefully generate a growing stream like a comet, as the original from 0 was. A bit speculation, but it looks like as if these newer breeds have always been there and derived already before the 2% from the normal stream of results. Based on this chart, I would set up a sample results.dat with all results from the envelope of this cloud, also these odd values with low particles but high percentages. This sample should only be changed where the envelop grows. This would ensure continuous wide testing of the whole parameter field with some kind of success. It would be an interesting thing to automate a selection routine to pick these distributed values out of the envelope ebenly. |
| Stephen Brooks 2004-05-13 02:41:12 | That graph is a rather quick way of showing how the optimisation looks - the number of Mpts is just being used as "another variable" to spread out the different sorts of result at the same percentage. |
