|
|
| (One intermediate revision by one other user not shown) |
| Line 1: |
Line 1: |
| − | Note to all folks editing these wiki pages:<br>
| |
| − | if you don't see the changes you make appear ...<br>
| |
| − | you need to clear your browser cache ... and then refresh the page ...<br>
| |
| − | this seems to happen at most wikis ... but i notice it here a lot.
| |
| − | ----
| |
| | | | |
| − |
| |
| − | Use this talk page to discuss various ideas of the implimentation of DB ...
| |
| − | and provide feedback for Bau.
| |
| − | ----
| |
| − |
| |
| − | ===Evo===
| |
| − | '''Here is how the 'evo' part works in my program:
| |
| − |
| |
| − | *A.) It uses femily based periodically changing mutation rates to avoid mullers rachet.
| |
| − |
| |
| − | *B.) Saving '''best robot''' at the end of a simulation or after n time, will result in saving a '''nuke problem''' robot or '''geek loophole''' robot, therefor the following must be done.
| |
| − |
| |
| − | #) Each robot must have its skill factor calculated to know its '''true''' best value. This is done using <b>".good"</b> in each robots substructure.
| |
| − | #) Troughout the simulation the higest peaking point of a robots skill is determind. Lets call it J
| |
| − | #) J is checked agenst the J of the three previos 'runs' (each run lests 8000 cycles (although I have "/end" display something else, I am talking about my current work right now) this is done by using an external "saecheck.txt" file.
| |
| − | ::a) If J is higher then the lest three 'runs', then 'save robots dna' is called. Wait for this 'run' to complete just incase J goes even higher.
| |
| − |
| |
| − | *C.) The true 'best robot' is now set as the robot to restart/resume the simulation with.
| |
| − |
| |
| − | i like these ideas, Bau ...<br>
| |
| − | makes a lot of sense to me. i will play with this prog this week and see how<br>
| |
| − | things go ... what questions are raised, etc ... and perhaps then know enough<br>
| |
| − | about these new features to ask relevent questions. thanks. {{User:Griz/sig}} 12:16, 2 Nov 2005 (MST)
| |
| − |
| |
| − | edit by bots , summary:
| |
| − |
| |
| − | '''Geek loophole problem and Nuke problem'''
| |
| − | http://forum.darwinbots.com/index.php?showtopic=781
| |
| − | http://forum.darwinbots.com/index.php?showtopic=703
| |
| − |
| |
| − | My definition of 'mullers rachet' is different from the Ensyclopidia it turns out, so here it is:
| |
| − |
| |
| − | '''Bau's rachet''' happens when an organism is changing at the same rate and time as all the other organisms (the changes in this case are making the organism worse) Since the saturation of worse organisms are the same for the entire feild, there is no selective presure for killing off the worse robots. Causing overall effiency to go down.
| |
| − |
| |
| − | : Not to beat a dead horse or anything, but what you are describing only occurs if Muller's ratchet is, uh, ratcheting. Basically if you don't give evolution time to sort good and bad mutations out, why should you expect to find good mutations accumulating at all? What sort of mutations/generation ratio are you running with anyway? --[[User:216.249.144.14|216.249.144.14]] 18:34, 5 Nov 2005 (MST)
| |
