April 4th, 2004, 03:27 PM
|
|
Guest
|
|
Posts: n/a
Time spent in forums:
Reputation Power:
|
|
Quote: | Originally Posted by swanepoeljan I wanna learn some more about AI and neural networks but
have been told to understand genetic algorithms first.
Well I've browsed about and got a site(URL) that tells you how to implement it but I don't get the logic about it. |
Why? Neural Networks are just as easy to implement as GAs.
Quote: | If you wanna solve the problem of finding a certain target number and can only use 0 - 9 and +,-,* and / and you use binary 4bit genes to encode it to a chromosome why do u have to do all that crossover and mutation stuff. Why can't you just go through all the possible chromosomes till you find the answer? |
Say you have a chromosome of length 8 genes. How many bits is that? 32 right? That's 4 billion possible solutions. 16 genes is 16 billion billion. You can't search it one by one. 16 billion billion is just too big. GAs partition the search space implicitly by seeking the best approximations to the right answer and merging them in attempt to find a better answer. Mutation is necessary due to the potential loss of interesting genes from the working set and because the population may get stuck in a local minimum.
Quote: | Can somebody explain more clearly the point of genetic algorithms and what kind of problems you solve with them. |
I believe you can solve the TSP with GA. Although solve is really approximate the optimal solution. You might actually find the optimal route, but you'll never know for anything other than trivial examples - if you knew the answer why search for it? The TSP is an example of a constraint optimisation problem; you want to find a solution which is as best as possible that matches certain constraints.
Other more interesting problems are possible candidates for GAs. Google might help uncover some interesting uses of GAs...
|
Dev Shed Forums Sponsor:
Del.icio.us
Digg
Google
Spurl
Blink
Furl
Simpy
Y! MyWeb
Linear Mode
