Go Premium for a chance to win a PS4. Enter to Win

x
  • Status: Solved
  • Priority: Medium
  • Security: Public
  • Views: 363
  • Last Modified:

Genetic Algorithm to Control Sub GA's at Different Time Frames

Is it possible/appropriate to use a master genetic algorithm (boolean) to crunch output from sub algorithms. The sub algorithms may generate boolean, integer, or float value outputs. The other major consideration is that these sub algorithms operate at different time scales. The proposal is to use ABC style algorithms to minimise pitfalls and limitations of standard Genetic Algorithms. Queen Bee Evolution which is significantly more efficient and advanced over standard Artificial Bee Colony Algorithms.
0
XGIS
Asked:
XGIS
  • 2
1 Solution
 
ozoCommented:
it may be, depending on the problem
0
 
XGISAuthor Commented:
The master algorithm is required to make buy and sell decisions on a financial time series. The
Sub algorithms are designed to learn a number of tasks;

eg One GA to determine optimal entry and exits without conventional indicators.
eg One GA may be assigned to determining the best time to enter a trade at the 15 minute time-frame, that dynamically assesses the most probable entry at smaller time frames.
eg A larger scale GA may look at 24 hourly or fortnightly cycles.  Longer cycles may feed a basic buy/sell signal.
eg Another GA may be used to manage a neural network to classify data based on time period.

Artificial Bee Colonies can use multiple unrelated datasets without the need for rigid structure and pre treatment like ANN's. If the newer Queen Bee Evolution algorithms are the most powerful (up to 200x faster than normal GA) then should ALL of the algorithms be Queen Bees.  Weighting may need to be considered also as data at longer time frames results in more significant moves than at smaller timeframes.

Thankyou
Aaron

This should help define the problem a bit more.

0
 
derduffCommented:
If I understand you correctly, you ask if it is possible to use genetic algorithms inside the fitness function of another genetic algorithm. Yes, this is possible and has been already done under the name meta-evolution in order to optimize parameters of genetic algorithms (e.g. population size, selection strategies, crossover operators..).

Having said this, you should really start with an exact formulation of you fitness function. What are the variables you want to optimize, how do you evaluate success, etc What you are describing in you second post is a really complicated strategy, which will most probably lead to a system you can not really trust.

And lastly: There are really few differences between all the different kinds of bioinspired optimization. This research field produces a plethora of algorithms with fancy names, but there is really not much innovation. I suggest you start with the simple ones, e.g. Differential Evolution, a standard Genetic Algorithms and maybe Evolution Strategies.

It is far more important that you understand the problem, and choose a good formulation.

0
 
XGISAuthor Commented:
Thanks for the advice. After more research I should be able to compile a more workable concept.
0

Featured Post

Ask an Anonymous Question!

Don't feel intimidated by what you don't know. Ask your question anonymously. It's easy! Learn more and upgrade.

  • 2
Tackle projects and never again get stuck behind a technical roadblock.
Join Now