Howard just announced another new one Here . So, who's gonna be the guinea pig? :D
Printable View
Howard just announced another new one Here . So, who's gonna be the guinea pig? :D
Just put it on one of my boxes I'll report back after it's been going a while
one of my Windows boxes and my Linux box have been updated - hopefully it removes/lessens the plateau effect :)
let us know if it is "rewarding" or not ;)Quote:
Originally posted by magnav0x
Just put it on one of my boxes I'll report back after it's been going a while
Dunno, but my "plateau effect" is worse, and RMS's seem to be a bit higher as well..... Can't tell if there is any speed difference based on the original beta exe. No razzle dazzle with this one guys :rolleyes:
I'm not sure if anyone else has tryed this new exe, but after letting it run a while, it seems that the time per generation are greater and the the energies are staying between 40 and 50.... I don't realy know if a higher or lower energy is better. I figured I'd throw it out to those that know though. With the previous exe I was getting energies in the 30-40 range.
Windows Test Client #2 - running on a P4 2.6, WinXP. Average time per gen has dropped to around 5 mins from 10 mins (although may just be this run). Plateau effect seems lessened - hovering around the mid 20s for energy.
Linux Test Client #2 - running on a Barton 2800+, Mandrake 9.2. Similar to the Windows client - average time has decreased, plateau effect not noticeable, hovering around the low 30s for energy.
Not sure if it'll produce better RMSs but it is seeming to be more stable on the energy scores...
according to howard on the DF forum, the new exe " now focuses on getting the structure compact first, and then worries about pseudo-energy val" so this could be why the pseudo-energy is higher than before (i think lower is better). i guess the idea is that a compact structure will yield better RMSD than a larger one with a lower pseudo-energy.