What are good examples of genetic algorithms/genet

Question

Genetic algorithms (GA) and genetic programming (GP) are interesting areas of research.

I'd like to know about specific problems you have solved using GA/GP and what libraries/frameworks you used if you didn't roll your own.

Questions:

• What problems have you used GA/GP to solve?
• What libraries/frameworks did you use?

I'm looking for first-hand experiences, so please do not answer unless you have that.

Answer 1

I developed a multithreaded swing based simulation of robot navigation through a set of randomized grid terrain of food sources and mines and developed a genetic algorithm based strategy of exploring the optimization of robotic behavior and survival of fittest genes for a robotic chromosome. This was done using charting and mapping of each iteration cycle.

Since, then I have developed even more game behavior. An example application I built recently for myself was a genetic algorithm for solving the traveling sales man problem in route finding in UK taking into account start and goal states as well as one/multiple connection points, delays, cancellations, construction works, rush hour, public strikes, consideration between fastest vs cheapest routes. Then providing a balanced recommendation for the route to take on a given day.

Generally, my strategy is to use POJO based representaton of genes then I apply specific interface implementations for selection, mutation, crossover strategies, and the criteria point. My fitness function then basically becomes a quite complex based on the strategy and criteria I need to apply as a heuristic measure.

I have also looked into applying genetic algorithm into automated testing within code using systematic mutation cycles where the algorithm understands the logic and tries to ascertain a bug report with recommendations for code fixes. Basically, a way to optimize my code and provide recommendations for improvement as well as a way of automating the discovery of new programmatic code. I have also tried to apply genetic algorithms to music production amongst other applications.

Generally, I find evolutionary strategies like most metaheuristic/global optimization strategies, they are slow to learn at first but start to pick up as the solutions become closer and closer to goal state and as long as your fitness function and heuristics are well aligned to produce that convergence within your search space.

Answer 2

I am part of a team investigating the use of Evolutionary Computation (EC) to automatically fix bugs in existing programs. We have successfully repaired a number of real bugs in real world software projects (see this project's homepage).

We have two applications of this EC repair technique.

• The first (code and reproduction information available through the project page) evolves the abstract syntax trees parsed from existing C programs and is implemented in Ocaml using our own custom EC engine.

• The second (code and reproduction information available through the project page), my personal contribution to the project, evolves the x86 assembly or Java byte code compiled from programs written in a number of programming languages. This application is implemented in Clojure and also uses its own custom built EC engine.

One nice aspect of Evolutionary Computation is the simplicity of the technique makes it possible to write your own custom implementations without too much difficulty. For a good freely available introductory text on Genetic Programming see the Field Guide to Genetic Programming.

Answer 3

It was a while ago, but I rolled a GA to evolve what were in effect image processing kernels to remove cosmic ray traces from Hubble Space Telescope (HST) images. The standard approach is to take multiple exposures with the Hubble and keep only the stuff that is the same in all the images. Since HST time is so valuable, I'm an astronomy buff, and had recently attended the Congress on Evolutionary Computation, I thought about using a GA to clean up single exposures.

The individuals were in the form of trees that took a 3x3 pixel area as input, performed some calculations, and produced a decision about whether and how to modify the center pixel. Fitness was judged by comparing the output with an image cleaned up in the traditional way (i.e. stacking exposures).

It actually sort of worked, but not well enough to warrant foregoing the original approach. If I hadn't been time-constrained by my thesis, I might have expanded the genetic parts bin available to the algorithm. I'm pretty sure I could have improved it significantly.

Libraries used: If I recall correctly, IRAF and cfitsio for astronomical image data processing and I/O.

Answer 4

I built a simple GA for extracting useful patterns out of the frequency spectrum of music as it was being played. The output was used to drive graphical effects in a winamp plugin.

• Input: a few FFT frames (imagine a 2D array of floats)
• Output: single float value (weighted sum of inputs), thresholded to 0.0 or 1.0
• Genes: input weights
• Fitness function: combination of duty cycle, pulse width and BPM within sensible range.

I had a few GAs tuned to different parts of the spectrum as well as different BPM limits, so they didn't tend to converge towards the same pattern. The outputs from the top 4 from each population were sent to the rendering engine.

An interesting side effect was that the average fitness across the population was a good indicator for changes in the music, although it generally took 4-5 seconds to figure it out.

Answer 5

I used genetic algorithms (as well as some related techniques) to determine the best settings for a risk management system that tried to keep gold farmers from using stolen credit cards to pay for MMOs. The system would take in several thousand transactions with "known" values (fraud or not) and figure out what the best combination of settings was to properly identify the fraudulent transactions without having too many false positives.

We had data on several dozen (boolean) characteristics of a transaction, each of which was given a value and totalled up. If the total was higher than a threshold, the transaction was fraud. The GA would create a large number of random sets of values, evaluate them against a corpus of known data, select the ones that scored the best (on both fraud detection and limiting the number of false positives), then cross breed the best few from each generation to produce a new generation of candidates. After a certain number of generations the best scoring set of values was deemed the winner.

Creating the corpus of known data to test against was the Achilles' heel of the system. If you waited for chargebacks, you were several months behind when trying to respond to the fraudsters, so someone would have to manually review large numbers of transactions to build up that corpus of data without having to wait too long.

This ended up identifying the vast majority of the fraud that came in, but couldn't quite get it below 1% on the most fraud-prone items (given that 90% of incoming transactions could be fraud, that was doing pretty well).

I did all this using perl. One run of the software on a fairly old linux box would take 1-2 hours to run (20 minutes to load data over a WAN link, the rest of the time spent crunching). The size of any given generation was limited by available RAM. I'd run it over and over with slight changes to the parameters, looking for an especially good result set.

All in all it avoided some of the gaffes that came with manually trying to tweak the relative values of dozens of fraud indicators, and consistently came up with better solutions than I could create by hand. AFAIK, it's still in use (about 3 years after I wrote it).

Answer 6

Its often difficult to get an exact color combination when you are planning to paint your house. Often, you have some color in mind, but it is not one of the colors, the vendor shows you.

Yesterday, my Prof. who is a GA researcher mentioned about a true story in Germany (sorry, I have no further references, yes, I can find it out if any one requests to). This guy (let's call him the color guy) used to go from door-door to help people to find the exact color code (in RGB) that would be the closet to what the customer had in mind. Here is how he would do it:

The color guy used to carry with him a software program which used GA. He used to start with 4 different colors- each coded as a coded Chromosome (whose decoded value would be a RGB value). The consumer picks 1 of the 4 colors (Which is the closest to which he/she has in mind). The program would then assign the maximum fitness to that individual and move onto the next generation using mutation/crossover. The above steps would be repeated till the consumer had found the exact color and then color guy used to tell him the RGB combination!

By assigning maximum fitness to the color closes to what the consumer have in mind, the color guy's program is increasing the chances to converge to the color, the consumer has in mind exactly. I found it pretty fun!

Now that I have got a -1, if you are planning for more -1's, pls. elucidate the reason for doing so!