Genetic algorithm basin question

For discussions about programming, programming questions/advice, and projects that don't really have anything to do with Puppy.
Post Reply
Message
Author
User avatar
mahaju
Posts: 487
Joined: Mon 11 Oct 2010, 07:11
Location: between the keyboard and the chair

Genetic algorithm basin question

#1 Post by mahaju »

Hello
Does genetic algorithm always converge towards and optimal solution? Are there any papers or articles that have proved this mathematically or empirically?
The process or crossover preserving genes that are closer to optimal sounds somewhat logical, but couldn't an improper mutation completely diverge the solution away from optimal?

Thanks in advance.

User avatar
Flash
Official Dog Handler
Posts: 13071
Joined: Wed 04 May 2005, 16:04
Location: Arizona USA

#2 Post by Flash »

By 'genetic algorithm' do you mean random mutation followed by natural selection? If so, what is optimal?

bugman

#3 Post by bugman »

i don't personally believe that mathematics is particularly useful in the life or social sciences, apart from trends and probabilities

bugman

#4 Post by bugman »

i also don't think trends or probabilities are particularly useful

who cares if smoking is likely to cause cancer or heart disease if i get cancer or heart disease and i don't smoke?

and i see the 98-year-old on tv who smokes a pack a day?

:roll:

User avatar
rcrsn51
Posts: 13096
Joined: Tue 05 Sep 2006, 13:50
Location: Stratford, Ontario

#5 Post by rcrsn51 »

bugman wrote:who cares if smoking is likely to cause cancer or heart disease if i get cancer or heart disease and i don't smoke?
The people whose tax dollars are used to treat those cancer patients.

bugman

#6 Post by bugman »

rcrsn51 wrote:
bugman wrote:who cares if smoking is likely to cause cancer or heart disease if i get cancer or heart disease and i don't smoke?
The people whose tax dollars are used to treat those cancer patients.
i live in the u$, no worries, eh?

but now you're discussing economics rather than biology

User avatar
Flash
Official Dog Handler
Posts: 13071
Joined: Wed 04 May 2005, 16:04
Location: Arizona USA

#7 Post by Flash »

bugman wrote:... now you're discussing economics rather than biology
No, he's discussing whether the government has a legitimate interest in regulating smoking. Since the government winds up paying for the care of many people who have illnesses strongly associated with smoking, the answer is clearly yes. (Unless you like your tax dollars going to subsidize tobacco companies, which don't pay for the damage they cause.)

Just as the government has a good reason for telling motorcycle riders they have to wear protective headgear. Guess who winds up paying for the care of many of the people whose brains were addled when they wrecked a motorcycle while not wearing a helmet?

And I might extend the argument to a nation that is contemplating building nuclear power plants. When the damned things blow up, the contamination is not confined within the borders of the plant, or even the borders of the nation responsible.

bugman

#8 Post by bugman »

Flash wrote:
bugman wrote:... now you're discussing economics rather than biology
No, he's discussing whether the government has a legitimate interest in regulating smoking. Since the government winds up paying for the care of many people who have illnesses strongly associated with smoking, the answer is clearly yes. (Unless you like your tax dollars going to subsidize tobacco companies, which don't pay for the damage they cause.)
the government pays for the care of people? in the u$?

really? i'm quite ill and had no idea . . .

User avatar
Flash
Official Dog Handler
Posts: 13071
Joined: Wed 04 May 2005, 16:04
Location: Arizona USA

#9 Post by Flash »

Are you forgetting about state governments?

nooby
Posts: 10369
Joined: Sun 29 Jun 2008, 19:05
Location: SwedenEurope

#10 Post by nooby »

Guys don't get me wrong now. I easily derail any thread even my own.

What about the code that learn from it's own mistakes? Such a neat idea.

They have shown Robots that learn to walk by endlessly falling and falling for weeks or months on end and not giving up and in the end they run like a Sprinter and will one day beat any athlete on the Olympic Games :)

Go back derailing I don't mind. I am not the Mod either. *Friendly Smile*

But the thread is really intersting. Could one set up a kind of software that learned to be a very good assembler Puppy that was optimised for every hardware it came upon? Some 3MB big and could do what Puppy do now :)

I go to sleep now and in my dreams I teach such a program to make a lean and mean small puppy with assembler code. Fast and strong. :)
I use Google Search on Puppy Forum
not an ideal solution though

User avatar
mahaju
Posts: 487
Joined: Mon 11 Oct 2010, 07:11
Location: between the keyboard and the chair

Hi

#11 Post by mahaju »

Hello friends
This thread is going completely off topic
I need to give a report on implementing network intrusion detection system based on genetic algorithms, so that the system can learn from a database of network intrusion patterns, whether a given activity at a given time is a possible intrusion or not
The work is still in it's initial phase and I am studying the basics of genetic algorithms for this. It was stated in a book (which only had a sort of introductory chapter on Genetic algorithms) that due to the genetic operators used in the algorithm, the optimal solution is always preserved through the generations as the possible solution evolves. The book didn't mention any references for it's proof. I was hoping I would get an answer to this.
Don't get me wrong, I have read all the replys to my question and it's a very interesting discussion but I need to write my report as well
:D

bugman

#12 Post by bugman »

Flash wrote:Are you forgetting about state governments?
i suppose that depends on the state you live in, here we're derailing, but not as quickly as others

i would amend my previous statement [whatever it was] to boot

mathematics does not belong in economics either

currently reading about john nash, the nash equilibrium, and its contribution to the current mess

nooby
Posts: 10369
Joined: Sun 29 Jun 2008, 19:05
Location: SwedenEurope

#13 Post by nooby »

mahaju , I guess you ahve to find a forum for Genetic programming. Where they do such things daily and them would know how you could proceed.

This Puppy forum and very few are doing genetic programming at all.
Maybe no a single person of all users. I doubt that there is one doing it.
I use Google Search on Puppy Forum
not an ideal solution though

User avatar
Flash
Official Dog Handler
Posts: 13071
Joined: Wed 04 May 2005, 16:04
Location: Arizona USA

#14 Post by Flash »

bugman wrote:...mathematics does not belong in economics either
I would agree.
...currently reading about john nash, the nash equilibrium, and its contribution to the current mess
Do you have a link to what you're reading about the Nash equilibrium?

Nooby, if Mahaju had replied immediately to my first post, perhaps the thread would have stayed on track. What a difference a day makes, eh? :)

nooby
Posts: 10369
Joined: Sun 29 Jun 2008, 19:05
Location: SwedenEurope

#15 Post by nooby »

He had too high expectations based on the Term Programming over the part of teh forum. He thought us experts on programming on the University level like Teachers at a Doctorate level?
I use Google Search on Puppy Forum
not an ideal solution though

bugman

#16 Post by bugman »

Flash wrote:[
Do you have a link to what you're reading about the Nash equilibrium?
no, it's very loose and undisciplined reading, still getting over the flu

that fucking a beautiful mind book followed up by wiki pages on nobel economists and other internets dreck

still can't decide if i want to learn more and really depress myself or not

edit: i expect some of my antipathy came from dredged-up memories of my ex-brother-in-law, this guy:

http://home.earthlink.net/~hipbone/

it took me a long time to figure out a lot of game theory [and the game he invented] was really crap, it wasn't just me being thick

and it's crap that's killing people, yippee!

User avatar
mahaju
Posts: 487
Joined: Mon 11 Oct 2010, 07:11
Location: between the keyboard and the chair

#17 Post by mahaju »

nooby wrote:He had too high expectations based on the Term Programming over the part of teh forum. He thought us experts on programming on the University level like Teachers at a Doctorate level?
It's not that
I just find the people in this forum more helpful
Other forums are mostly full of trolls
I am just hoping that if somebody has any idea about this thing he/she will be able to help

User avatar
Moose On The Loose
Posts: 965
Joined: Thu 24 Feb 2011, 14:54

#18 Post by Moose On The Loose »

mahaju wrote:
nooby wrote:He had too high expectations based on the Term Programming over the part of teh forum. He thought us experts on programming on the University level like Teachers at a Doctorate level?
It's not that
I just find the people in this forum more helpful
Other forums are mostly full of trolls
I am just hoping that if somebody has any idea about this thing he/she will be able to help
As I see it, there are two main problems with any system that attempts to optimize something.

(1) How you define "better" can be tricky. A classic example of a very bad definition is the case of a city dumping its sewage into the river. The sewage is taken away and is no longer a problem for the city so from one point of view it is a very goo answer to what to do with sewage. If you are in the city down stream you would disagree.

(2) Even with a good definition of "better" you can fall into a "better" but not "best" dip in the curve. If the curve happens to look like a hill with a well at the peak. Sliding down hill doesn't get you to the best result. Unless your random mutation happens to hit the well exactly, you won't ever get there.

User avatar
Dougal
Posts: 2502
Joined: Wed 19 Oct 2005, 13:06
Location: Hell more grotesque than any medieval woodcut

#19 Post by Dougal »

bugman wrote:mathematics does not belong in economics either
Mathematicians don't have a particularly high opinion of economists...

A friend at uni went to do a PhD in game theory (under the supervision of a Nobel winner).
The reason he chose game theory: because it's easy... (I think of it this way: it's a new discipline, so they're doing the basic, easy parts now.)

He's now doing his post-doc in Cambridge (or is it Oxford?).
When it was heard he was accepted, he'd meet old professors in the hall and they'd say:
"I heard you're going to do a post-doc in Cambridge"
He'd reply:
"Yes, in the economics department"
"Oh", they'd say and walk on, having lost all interest.
What's the ugliest part of your body?
Some say your nose
Some say your toes
But I think it's your mind

User avatar
mahaju
Posts: 487
Joined: Mon 11 Oct 2010, 07:11
Location: between the keyboard and the chair

#20 Post by mahaju »

As I see it, there are two main problems with any system that attempts to optimize something.

(1) How you define "better" can be tricky. A classic example of a very bad definition is the case of a city dumping its sewage into the river. The sewage is taken away and is no longer a problem for the city so from one point of view it is a very goo answer to what to do with sewage. If you are in the city down stream you would disagree.

(2) Even with a good definition of "better" you can fall into a "better" but not "best" dip in the curve. If the curve happens to look like a hill with a well at the peak. Sliding down hill doesn't get you to the best result. Unless your random mutation happens to hit the well exactly, you won't ever get there.
Genetic Algorithm doesn't guarantee to give the best answer
That is the reason it needs a large solution space and a lot of training time to converge towards the most optimal solution (may or may not be the best answer) but I have read that because of the way genetic algorithms treat it's parameters, the algorithm usually converges to the best solution in a large pool of possible solutions.
I need a reference where this has been proved (mathematically or experimentally or in any other way)

Post Reply