Posted by Phil M on September 02, 2003 at 03:07:48:
In Reply to: Re: Evolution Pressure: Variation posted by stmx3 on August 29, 2003 at 09:28:33:
: --I assumed that culling of the least-fit occurs by their receiving few or no votes. I haven't been here long enough to have been involved in discussions as to how poems "died"...I thought there might be some limited lifespan weighted based on the number of votes received, if any. So, if that were correct, then it seems my first case (above) is being used. Whether or not lifespan is being modified, the strongest are conferred a greater likelihood of breeding.
First of all, I must admit that I don't actually speak for David, the progenitor of this project, or have any input other than whispering the odd crazy idea in his vicinity (which, if he's sensible, he should probably ignore) but I think I know the mechanisms. In certain facts I'm plain wrong... :)
Having said that, I think the 'least fit' measurement occurs from loss of votes (votes against) so that (by chance) rarely presented poems don't lose out just because chance dictates that they haven't been in many contests. The ones that are culled (as I understand) pass some negative threshold, which random-walk theory dictates will only happen if they consitently fail to match up against their competitors. (Chance may mean their fights are skewed towards stronger competitors, but there's still an advantage from that in that upper-quality poems good scoring.
Anyway, lifespan isn't set in stone (or ROM). It will depend, as you say, partially on the votes accumulated. As long as an entitity does well against competitors, it stays alive. There's a penalty for breeding, but a still good poem can replenish its expenditure of goodwill.
: --In my second case, the voter has a direct input in the breeding likelihood of "abominations". I.e. there are votes to "neuter" the poem and let it die naturally without producing offspring.
Ultimately, abominations are culled (the more universally they are considered to be this, the sooner it will occur) and it dies, which is the same as leaving the breeding cycle. (Poems are labelled "Dead", but they could as easily be labelled "Sterile" or "Packed off to the Old Folks' Home".) The voter's input is vote for or vote against (one of each at for every pair compared).
: --I think these 2 cases are very equal, although I'll have to give more consideration to you gaussian distribution effects. My misconceptions lie in not fully understanding the program. That is to say, I don't really know what happens to those poems that, for whatever reason, don't get an opportunity to be voted on. Or perhaps they statistically come up less frequently than other poems. And I don't know what their "lifespan" is.
As mentioned above (sorry, I take a linear/tangential approach to writing) voteless points sit around until chance comes round and gives them a kick, mroe or less. Or if they find themselves chosen for breeding (not sure if there's a threshold for that, apart from the Dead/alive separation, but I believe there's a slightly skewed probability towards high-scorers) they may go into arrears without a vote cast against them, but by then they have offspring, which should even things out (given they are untried and statistically as likely to be bad as good).
: --We are averaging about 60 votes/hr (over the past 17 days). So, that means 2880 poems per day are being looked at...a little over twice the database size (1200 poems, right?) I think you would want the number of voting opportunities to be at least 5x the database size to ensure 95% of the poems get an opportunity to be looked at. So, a poems lifespan should be 2.5 days. If lifespans are longer than this, then I don't think it matters if you vote for or against a poem. But I think it would matter if lifespans are less.
David? Average living lifespan figures, please? And/or average dead lifespans, I suppose. What a true statistics-geek would to do is to analyse (for given points in time/number of votes cast) how many once living poems are dead at any particular point in time, how long living poems at that point in time (but dead now) ended up living and see how if that extrapolates into the number of currently living poems (now) that were already alive at that point. From that you may be able to get a flowing mortality rate by age and hence see if there's a theoretical mortaility curve for the latest generation. Sounds more complicated than (I think) it is... :)
: --Thanks for the discussion. I'll give the gauss distribution more thought, and look back at the archives to see what discussions I've missed on the programs internals.
Concentrate on David's words, for they are the Word Of God and (usually) precise and correct. Everything else is just our observation of the universe and attempts to make sense of The Great Unknown by way of superstring theory, dark matter and the prevalence (or otherwise) of a universal ether. Of course, in this universe the rules occasionally change, but David is a Communicative God and usually mentions when he does that (often with a reason) though there are also probably things he considers a trade secret... :)