Genetic Algorithms

[GilaManster]

http://gah.stanford.edu/index.html

Check out this distributed computing project. It may help you out in your research.

GilaM

[RobRich]

One the better resources you can find on GA tech is provided via the Navy:

http://www.aic.nrl.navy.mil/galist/

The site offers tons of research examples, links, and most important, source code for linear/non-linear operations, numerical optimization, recombination operators, etc. Test suites are even provided for exploration of GA code paths, plus some apps even provide a standardized path for many popular GA operations with a scripted interface model for inputing and processing your desired datasets.

Robert Richmond

[qball]

Quote:

qball--Just give it a rest man. I don't know exactly wtf he is talking about either, but that doesn't mean it is invalid because I have never done what he asks.

ok, maybe, I have.

In principle, EAs can compute any computable function, i.e. everything a normal digital computer can do.
But EAs are especially badly suited for problems where efficient ways of solving them are already known

just texted from:

http://www-2.cs.cmu.edu/Groups/AI/html/faqs/ai/genetic/

google it, right there

if you want to do it, you need to, well, do it.

[baby_ja]

Quote:

Originally Posted by qball

Genetics? or Algorithms?



what's problem? how does "Genetic Algorithms" solve, or got to do?

answer me

[hrev0]

Your description of what youre tryin to do is a bit vague but i dont think a genetic algorithm is going to be your best solution. Can you be more specific on what you mean by "optimize what options are ordered"?

Id look into Adaptive Resonance Theory.

Let me know more about the project, Ive been reading alot about NNs lately maybe I can help.

[Rootstonian]

As a computer programmer, that's a tough one; lots of variables. Almost needs to be done with historical data and a statistician I won't be much help here, but just had to offer my $0.02

For example: I have a Chevy truck. Probably can get the same options on a Chevy Impala. Do I want leather, heated seats and an electric sunroof on my truck? Probably not. So, you have to look at the market you're selling to. I use my truck as a "work" truck by someone MIGHT want the leather seats etc. So, what quantitiy of truck do you pre-make with the leather seats???

And trends can change that a program could never pick-up on. Like the recent surge in gas prices; would the program know to lessen the number of trucks ordered?

I would go back to a basic data base and some online analytical processing (OLAP) tools for this type of problem. And throw in the gut insticnt of the dealer ordering the vehicles. Some things just can't be programmed. I'm not saying this can't be done, I just don't think it's going to be as accurate as tracking historical sales and the gut feel of the dealer.

[Ruler2112]

I'm very sorry that this thread got dredged up again.

After much research, I did create a functional and efficient GA for this app. (I added a few very simple tweaks to the standard GA structure, just because I thought they'd increase efficiency. They seem to have improved results. ) My program worked basically like this:

The person running the program provides one number standard - how many vehicles they want to order. The program first diagnoses how many of each model should be ordered. Then for each model, it has a list of options available. I analyze past sales and use a GA to predict what options would be best to order, encoding all available options into a bit-stream.

It has been working relatively well. The GM uses it in conjunction with the 'gut feeling' developed over the past 20+ years. I know if I could do better if I had an advanced math degree, but I don't and from what I remember of Calc, I don't want one either.

Thank you both hrev0 and Rootstonian for your input. (I wish you guys would've been around a few years earlier when I was struggling with this project. )

[Rootstonian]

Glad it "mostly" works. What you were trying to build was a JIT (Just In Time) inventory system. That is, you want to run out of a product the day your delivery comes in; tough to do in car sales.

I bet your program and algorithm would work great in a restaurant environment where weekly sales are more stable.