Code center > Suggestions
Mutation Protection method- in voting
jknilinux:
Ok, Nums and Eric (and Bacillus' first idea, if you count that B changed his vote to his new idea) are tied...
If we ultimately end up with a tie/one vote difference between ideas, maybe we should combine them. Then again, people might like the hybrid less...
We could implement both, but that might take too long...
If this continues, then maybe we should count bacillus' old idea as being a variation of eric's, since it basically is just eric's idea with costs. So, eric's supporters are really the ones who voted for eric AND the ones who voted for B's old idea. In that case, we'll make B's old idea the winner, but make the costs optional.
So far, this is what it looks like:
---Settings------
[*] Turn on Mutation protection?
[*] Turn on Mutation protection costs?
-------------------
0: 10 .protect store '(protects self)
...12345678... '(protects next nine BPs)
9: 100 .mkprotenzyme store '(make the most recent .protect instruction give 100% mut protection, which costs 100/50 or 2 nrg per cycle per BP (10 nrg/cycle in this case), and whose cost increases exponentially over time OR ...and whose mutation protection decreases by 1 per cycle)
--AND--
---Settings------
[*] Turn on Mutation protection?
[] Turn on Mutation protection costs?
-------------------
0: 10 .protect store '(protects self)
...12345678... '(protects next nine BPs)
9: 100 .mkprotenzyme store '(make the most recent .protect instruction give 100% mut protection, with no costs/degradation)
This should make a little under 2/3 of the users happy, judging by the polls. The first case is good for evosims, the second for antbot or NNbot evolution/leagues-with-NN-weight-adjustment.
jknilinux:
OK, apparently the vote's over- no one has voted to change anything since my last post in mid-october, so numsgil:
It says the suggestions forum is led by both you and ericl- can you implement the mutation protection as described in my last post, since eric has disappeared?
By the way, to the new users (peterb, de-evo, etc...), feel free to vote!
peterb:
--- Quote from: Numsgil ---It's all pretty moot if Eric never shows. Remember he's the one that ultimately implements things, and he seems to be AWOL.
--- End quote ---
Is only one person devolping DarwinBots ?
I have some knowledge on programming languages.
Well that is I'll easily pickup such languages often.
Dough I never went deep into them ( these days I get headaches from looking long at code)
Anyway I went from :
MSX assembler / Pascal / QuickBasic / VB6 / VB2003.net / C# (pocketpc) / C2003.net / Perl (little) / phyton (little) / LSL / vbscrip / VB2008.net
And maybe some more lost track of it a bit.
Dough I'm not a big fan of C languages, takes lots of time to write and read while VB.net is equaly fastand better readable.
Time might be the biggest issue dough.
I hope a project like this is structured / devided ?? >> I'm not sure how it is organised but it might be an idea to put source code in a special forum folder (so people could reply with improved code.. (and discuss their distributed programing efforts).
peterb:
Different idea.
Well different, I'm not sure I think DNA rampage by mutation often results in something supricingly
Yesterday one of my zerobot sudenly became a better hunter dough sadly also a canibal what I had tried to not to become.
Thiking about it a bot described by 1 gene, errors would happen and effect everything equaly inside that gene.
If a bot has multiple genes, like most bots; Then what genes would most likely mutate?
> maybe the smaller (so to say specialized)ones
> or making it depending on the seize of the cond fields (to give to verry advanced super eye commands some penalties?)
> or making it depending on the start stop seize
> or depending on the general gene size lengthe
> or based on how often genes fire
On multibots perhaps something that preserves body-bot specification (based on how often genes fire ?)
On zero bots... hmmm I got no idea ....
just a side to note that bots can change type from zerobot to 1 gene bot to multiple gene bot to multibot > all diffferent structures > different rules?
Dough on the otherhand sometimes you create a bot and wonder only what if the numbers in that Cond part would change.
It might be programmable over generations
Then perhaps just have some kind of flag commands
like :
1 .flagcommanderrors store
1 .flagvaleuerrors store
1 .flagallerrors store
=========>> or .DNAerrortype: -1 for numbers, 0 for both, 1 for commands
5 .flagpreserve store (preserves the next 5 code lines for mutation)
5 .flagmutate store (the oposite)
*.flagprotectedbits (how much bits are protected)
*.refprotectedbits (as ref command so you like to preserve?, then also others can identify you (or rather that is your famliy) by it
*.refunprotectedbits (maybe thats to much).
When using flagmutate Then it would impley that the rest could mutate.
Maybe put a cost to it too per cycle; the cost would depend on the number of bits protected by it seams fair
Or we would all make zero bots; close to oneliners
Maybe have something that more advanced creatures (more genes, more connected bots) have a better effect of it.
Would be like an evolution upward step and so have less error risk? (or maybe the oposite?)
DNA lesserrorriskfactors, overal bot gene:
*.numties *.numgenes sum *.shell sum
A shell also has UV protection from light so less errors son only in pond mode add *.depth)
*.numties *.numgenes sum *.shell sum sum *.depth sum
~maybe minus the amount of poison inside as that is dangerous stuff to handle, even for the creature itself.
Wow I'm a long speaker I notice, dough I think those are nice ideas concluding it would result like:
So the bigger the more advanced the more protection dough also a higher cost. (and can be dependant of pond height).
I dont know if this command would be a problem for the stack ?
As it could stand before a gene [cond] but also inside a gene [cond] or in the [start] part.
jknilinux:
10 .flagpreserve store is pretty much the exact same thing as 10 .protect store. I think you're basically saying the same thing as the Ericl and Bacillus hybrid (which became the final draft of the idea), you're saying to make an instruction to protect DNA as opposed to metadata, plus costs. If you read through the posts, you'll see this was bacillus's original idea.
However, the refprotectedbits instruction seems like it might be able to address some of the problems the metadata camp was having with eric's idea- namely, treating the protected DNA as a single object vs. a wad of instructions.
Nums- what do you think?
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version