New Mutation Paradigm

[Numsgil]

Shvarz:

Yeah, it's hard to decide what will be beneficial to mutations and what will be harmful.  I'll just try to figure out as many mutation possibilities and we can play around with them to see which are beneficial and which are harmful.

If you check out the mutations details for a bot several mutated generations down the line, and it has a disproportionate amount of a certain mutation type missing, you can probably draw some conclusions that that isn't a good mutation type.

Deletions, for instance, seem to generally be more destructive than useful.  I've had alot more successful sims that had deletions all set to 0.

[Greven]

This topic seems really interesting! The last few day I have thought about my DNA system, and found it rather dull!!??  :ph43r:  :sleep:
And I understand what Num and Carlo are saying! So it is totally fine that it will be dropped/forgotten etc....

Regarding Carlo's proposal about writing the limits the sysvar can have into the sysvar file, is a stroke of genius!!!!  :clap:  :clap:  ;)  B)  :P  B)  :P  ;)  B)  :P  ;)  :lol:  :bigginangel:  :boing:  :boing:  :boing:  :boing:

But how are we going to implement it? I can think of a few ways to do that:

If the number is bigger or lower than the limits:

•  We can use the limits.
  Example: limit 0-100, number = 204, then it will be scaled down to 100 etc.
• Start over with the limits: (ie. making them circular)
  Example: limit 0-100, number = 101, then number will be 0, if it is 200 it will be 100 etc.
• Or it could be a different version of number 2, counting up/down:
  Example: limit 0-100, number 101, the number will then be 99, at 200 it will be 0, at 201 it will be 1 etc.

I am for no. 2 or 3, (mostly 3). In this way the numbers can be mutating like hell, but we still get usefull DNA/genes whatever.

About making a one gene cycle, I see the point, but I dont think it will be super to do it. It will make survival much more random that it is now, and I think the old topic Num started about non-linear gene execution is better.... But I am open for new a excting ideas. :clap:

[Greven]

Deletions, for instance, seem to generally be more destructive than useful.  I've had alot more successful sims that had deletions all set to 0.

Deletious mutations will always be harmfull when you put a skinny clean little bot into the world of DB! When a designed bot is mutating, deletion will always make a gene less or non-functional. But if you have a bot with 10000000000 of lines with junk DNA, which have mutated for fun to be the fitteste for millions of generations, a deletion will not be has harmfull.... ;)

[shvarz]

If we put the limits for sysvar into the sysvar file, does that mean that by editing it we can affect the mutation rates?  If normally .repro is limited to 1<...<100 and I replace that to 49<...<51, does that mean that bots will not be able to mutate the .repro as easily as before and that it will always be set to 0?  That would be pretty cool!

[Carlo]

I'd make it this way:

every time a value is loaded from memory, the location's range is read and the value is normalized into a 0-100 value.

when a value is written down, it is retransformed back using the ranges of the destination location.

It seems to me it's the only way to implement this. But it would be a major change, completely messing all the existing dnas.

[Zelos]

wouldnt it be possible to make it so that the program code down commands and variables and values into a sequans of 1/0 or a/t/c/g (for more dna like) and then change a random base pair or 1/0 to something and then recode it so it becomes understandeble for the program when it reads it?

or maybe even make so that the program decode it into this system and read from it all the time and change it and then when we use something like "save dna" it decode it so we understand it.

[Numsgil]

The DNA is already parsed into a type-value block, which is in many regards identical to binary or quadrary representation.

Which is my fancy way of saying that DNA is already stored as strings of bits (ie: numbers), and then unparsed when you look at it again.

[Botsareus]

As you can see the system being talked about is flowless , it includes everything and anything, I espasialy like the idea of sections of dna being duplicated. A few qustions about that:

*That includes conditions too right?

*Can Dna from one gene be duplicated into another gene? (I currently modified my DB to do that for conditions, hopeing that it will produce better flow between genes.)


I am gessing there is a way to create empty genes and then fill them up too right? (simuler to split gene)

[Numsgil]

New mutations aren't respectful of gene boundaries, so yes, conditions, parts of genes, whole genes, all depending on your settings.

The weakness of the current system is it tries to be too smart.  The new system just treats the DNA as a solid string of stuff that can get fiddled with without worrying about what's where.

Basically, it's like making the DNA a bunch of building blocks and giving them to a baby to play with.  Much less inteligent design.