Author Topic: " All the mutations don't develop interesting beha  (Read 4911 times)

Offline Botsareus

  • Society makes it all backwards - there is a good reason for that
  • Bot God
  • *****
  • Posts: 4483
    • View Profile
" All the mutations don't develop interesting beha
« on: April 10, 2005, 07:55:04 PM »
AND PLEASE MAKE SURE ALL THE MUTATIONS WORK THIS TIME.... sorry caps.....  ;)

*******

(I was thinking: are robots not evolving new sysvars, like turning, because of lack of values? will see soon enough....)

Offline Numsgil

  • Administrator
  • Bot God
  • *****
  • Posts: 7742
    • View Profile
" All the mutations don't develop interesting beha
« Reply #1 on: April 10, 2005, 09:52:50 PM »
I spliced and moved all his new complaints right here, so we don't untrack the other threads.  The first was in Fish School, the other was, well, somewhere equally improper.

Now Bots, you're bots are proabably dying because either:

1.  Big Berthas are developing (bots that don't reproduce, and so just get bigger and bigger)
2.  Your bots can't survive on their own even with no mutations
3.  Your populations are too small, and so random population fluctuations caused by distant vegs not being found and eaten cause the species to die.  (Ymir has this problem.  It will die out without enemy bots stirring the pot, so to speak.)

Here's how you solve those problems:

1.  Make your bots only reproduce when their body is huge.  Also, decrease the waste threshold in the settings screen to like 100 or 50.

2.  Make it better!

3.  Make your bots do something other than spin when there's no food around.  They should go out and actively hunt.  Fish School does very good in large sims with few vegs because the schools allow the bots to cover alot of ground quickly.

As far as mutations not producing something interesting:  That's because bots have no reason to become interesting.  Evolution produces not optimal things, but servicable things.  Evolution only makes things as good as they have to be, not as good as they could be.

Make your simulations bigger and harder.  Make vegs get less energy.  Give high friction levels.  Stressful situations have always been the ones that give interesting evolution.  Our own species had to survive a long time in large deserts in the past.  It forced us to become smarter.  It forced us to plan ahead for times of drought.

The 'problem' with mutations in DB is that they're realistic.  There's nothing telling the bots that a mutation is good or bad except their ability to survive.  In other AL programs the program determines good or bad mutations.

Also, other AL programs only have like 6 or 7 values that can mutate.  And mutate in very fixed ways.  DB is much much more open ended.

Offline shvarz

  • Bot God
  • *****
  • Posts: 1341
    • View Profile
" All the mutations don't develop interesting beha
« Reply #2 on: April 10, 2005, 10:58:00 PM »
Another possibility is that your mutation rate is set too high.

My current bot has been evolving for over 20 million cycles, it got much better than the parent and accumulated quite a lot of mutations.  So don't say that nothing works - it works, you just need to give it time.

P.S: I am using 2.36.1.
"Never underestimate the power of stupid things in big numbers" - Serious Sam

Offline shvarz

  • Bot God
  • *****
  • Posts: 1341
    • View Profile
" All the mutations don't develop interesting beha
« Reply #3 on: April 10, 2005, 11:03:09 PM »
Just to give you an example: there are 22 mutations that got fixed in the population of my bots over 20 million cycles.  So that means that ~a million cycles is needed for a single mutation to appear and spread through population.
"Never underestimate the power of stupid things in big numbers" - Serious Sam

Offline Shen

  • Bot Destroyer
  • ***
  • Posts: 111
    • View Profile
" All the mutations don't develop interesting beha
« Reply #4 on: April 11, 2005, 05:24:22 AM »
Now we can run large arena sims Ive had quite a few decent evo sims, though Im pretty much waiting for the new mutation stuff to be more stable, like you Ive had problems with it crashing after an hour or so, no idea why. The problem seems to be that FirstBot is so simple that even a basic mutation changes a vital gene. Maybe increasing the duplicate gene chance would build up enough junk DNA to produce some results.

Check out My Evolution bot and Settings. I usually get good results.

I was gunna make a post on running evo sims, maybe Ill get round to it now.

Offline Shen

  • Bot Destroyer
  • ***
  • Posts: 111
    • View Profile
" All the mutations don't develop interesting beha
« Reply #5 on: April 11, 2005, 02:13:51 PM »
Ok constructive critisism. Dont use F1 mode for Evo Sims. It sucks. Theres simply way to much energy in to small an area. FirstBot is likely to need a lot of time to evolve properly as all the genes are pretty vital to its survival so if they mutate it dies off, thats why I said increase the duplicate gene chance so it can develop junk dna.

My sims run at ~50 bots in a size 6-8 arena which has the advantage of allowing individual bots to shine if they get a good mutation, killing off weaklings quickly, as well as running at 500 cycles/sec. Ive had bots evolve -6 body shots, changed genes so they flee from everything except alga, turned into spinners and more. Its all about enviroment and F1 sucks for evo.

Offline Botsareus

  • Society makes it all backwards - there is a good reason for that
  • Bot God
  • *****
  • Posts: 4483
    • View Profile
" All the mutations don't develop interesting beha
« Reply #6 on: April 11, 2005, 02:19:38 PM »
Yea you are right guys , deffinitaly somthing is wrong with my thinking here.

Next appointment...
« Last Edit: April 11, 2005, 02:21:32 PM by Botsareus »

Offline Botsareus

  • Society makes it all backwards - there is a good reason for that
  • Bot God
  • *****
  • Posts: 4483
    • View Profile
" All the mutations don't develop interesting beha
« Reply #7 on: April 11, 2005, 02:54:19 PM »
I like to make all my bots "blue color" before I make the simulation run.

But no , But no ,  I like "blue color"/...


Never mutated bots to win.

Never mutated bots to win.

No way F1 mode , BIG SCREEN SIMS BIG SCREENS SIMS.

NO WAY ,PLANTS TOO FAR ON BIG SCREEN , THATS WHY BOTS LOSE..

NO WAY MAKE IT BETTER....

NO WAY NO TURNING , A BOT MUST HAVE turning PROGRAMED BY THE you S E R not the M A S H I N E..

NO WAY MAN < NO WAY...

I look blue color lets go punks: bring your constructive critisimssiss spelling...

Offline PurpleYouko

  • Bot God
  • *****
  • Posts: 2556
    • View Profile
" All the mutations don't develop interesting beha
« Reply #8 on: April 11, 2005, 03:51:34 PM »
Quote
NO WAY NO TURNING , A BOT MUST HAVE turning PROGRAMED BY THE you S E R not the M A S H I N E..

Are you trying to say that there is no way that the mutations can ever come up with .aimsx or .aimdx in the actions section of the gene?

That is ridiculous. How hard is it to generate a random value of 5 (.aimdx) or 6(.aimsx)?

It is actually 2 chances in 1000 or 1 in 500 that when a new value is randomly generated, it will be one of these two.
The chances of it being preceded by a number that makes it useful is considerably less and the chance of it being followed by a store command is also somewhat less.

The chance of evolving a command such as 150 .aimsx store would be

1 in 3 (to select the command option from the sliders)

multiplied by 1 in 500 to get the aimsx (or aimdx)

Multiplied by 1 in 3 (to select pure number from the sliders)

Multiplied by 1 in 32000 to get the 150

Multiplied by 1 in 3 (to select the type of variable from the sliders)

Multiplied by 1 in 11 for type of operator



Total chance = 1 in 4,752,000,000

It really isn't all that surprising that they don't develop turning abilities now is it?

Now off you go to run your sim for 4 billion mutations.
Set your scripts to detect .aimdx.
Now tell me that you don't get a turning gene randomly created and we will discuss it further.
There are 10 kinds of people in the world
Those who understand binary.
and those who don't

:D PY :D

Offline PurpleYouko

  • Bot God
  • *****
  • Posts: 2556
    • View Profile
" All the mutations don't develop interesting beha
« Reply #9 on: April 11, 2005, 03:53:55 PM »
There actually may be one issue with what I just wrote.
That is that scripts are not (at present) set up to detect numbers in genes, only the sysvars so if your robot develops a gene like

150 5 store

The script won't catch it.
There are 10 kinds of people in the world
Those who understand binary.
and those who don't

:D PY :D

Offline Botsareus

  • Society makes it all backwards - there is a good reason for that
  • Bot God
  • *****
  • Posts: 4483
    • View Profile
" All the mutations don't develop interesting beha
« Reply #10 on: April 11, 2005, 04:08:30 PM »
FOR A PERSON READING THE CODE IT IS HARD TO SEE STUFF LIKE: 5 store , it looks like a memory location storage; a noob wont even think that it is some kind of sysvar , forget about bothering to looking it up on the sysvars file...

Therefore its important that the actualy sysvars mutate.

5 store

must be desplayed as:

.aimsx store

so its easy on the user!

I like to keep sysvar mutation seporate from number mutation. storage to memory locations is also important.


OR JUST MAKE MUTATIONS FOR 'INSERT A NEW VALUE' WITCH YOU SAYED YOU FIXED BUT NEVER RELEASED THE VIRSION OF THE PROGRAM WERE ITS FIXED.
« Last Edit: April 11, 2005, 04:13:56 PM by Botsareus »

Offline Numsgil

  • Administrator
  • Bot God
  • *****
  • Posts: 7742
    • View Profile
" All the mutations don't develop interesting beha
« Reply #11 on: April 11, 2005, 04:20:30 PM »
We should probably set up some kind of number to sysvar conversion for mutations.  If you have a number followed by a store, check to see if that number is a sysvar.  If so, convert it to a sysvar.

Offline shvarz

  • Bot God
  • *****
  • Posts: 1341
    • View Profile
" All the mutations don't develop interesting beha
« Reply #12 on: April 11, 2005, 04:32:10 PM »
Quote
Total chance = 1 in 4,752,000,000


Now PY, don't say that - this sounds like one of anti-evolution people talking.  Of course, the chances that a gene like that would be created from scratch are very low.  Now calculate the chances in the following scenario:

1. It does not have to be 150 - can be any number from 100 to 200 (or even from 1 to 1000).
2. There are already "store" commands in the DNA, so that mutation simply has to appear right in front of the "store" command.  It does not have to generate that command at all.
"Never underestimate the power of stupid things in big numbers" - Serious Sam

Offline Botsareus

  • Society makes it all backwards - there is a good reason for that
  • Bot God
  • *****
  • Posts: 4483
    • View Profile
" All the mutations don't develop interesting beha
« Reply #13 on: April 11, 2005, 04:33:52 PM »
ok , thats a good solution Num.


 read "AAA Here is post"  for more ideas!!!, If you dont understand somthing in there:

*The Part that you dont understand: Quote it in your own reply inside that thread, and please explain how you dont understand it.

*I still think the 'mutation rates of mutation rates' when set to 1 are currently changing the 'other rates' too slow for the fact that: its the Fastest possible rate change.
« Last Edit: April 11, 2005, 04:35:10 PM by Botsareus »