I'm pretty new with genetic optimizations,
If some members can share their experiences with us, that would be really cool.
What parameters are the most important to have optimizations with quality?
Let's take an example,
1. I have 480 000 tests to do in "exhaustive" mode. In genetic mode, the population size is only 109 (automatic). It seems really small. What's a good ratio between the total number of tests and the population size to have good results? 1/10, 1/5, 1/3, 1/2?
2. Is the crossover probability/mutation probability significant to modify? I do not think so but?
3. The maximum of population is set to 383 (for the 480 000 tests), how do I can optimize/change it? It's a ratio with the population size? For example I want to increase the population size to 10000. What Maximum Number of generation should I take?
4. Is the Genetic Algorithm Subtype important for the quality of the results?
Best Regards,
Laurent
Genetic Optimizations
-
- Posts: 159
- Joined: 20 Nov 2010
- Location: France
- Has thanked: 76 times
- Been thanked: 34 times
Re: Genetic Optimizations
I'm surprised, for me Genetic Optimizations is a key feature of MultiCharts, and nobody seems to know/use it correctly!
Nobody with experience to answer to my questions??
Regards,
Laurent
Nobody with experience to answer to my questions??
Regards,
Laurent
- TJ
- Posts: 7740
- Joined: 29 Aug 2006
- Location: Global Citizen
- Has thanked: 1033 times
- Been thanked: 2221 times
-
- Posts: 159
- Joined: 20 Nov 2010
- Location: France
- Has thanked: 76 times
- Been thanked: 34 times
Re: Genetic Optimizations
Nope, I haven't read a book about it, only the MultiCharts documentation.
And it does not answer to my previous questions (Question 1 and 3 are the most important in my opinion =)).
I hope I do not need to read a book to understand how to calculate 3 or 4 parameters in MultiCharts. (I hope that not so difficult to calculate).
Would be just cool to find users that could share their experience on the subject!
I'm the only one who want to use that nice feature!!!???
Regards,
Laurent
And it does not answer to my previous questions (Question 1 and 3 are the most important in my opinion =)).
I hope I do not need to read a book to understand how to calculate 3 or 4 parameters in MultiCharts. (I hope that not so difficult to calculate).
Would be just cool to find users that could share their experience on the subject!
I'm the only one who want to use that nice feature!!!???
Regards,
Laurent
-
- Posts: 38
- Joined: 21 Aug 2011
- Has thanked: 14 times
- Been thanked: 16 times
Re: Genetic Optimizations
Genetic Optimization is one of sampling method.
Nobody can answer your question because the performance of Genetic Optimization rely on your criteria, data, your strategy. Try to plot it with 3D Optimizer.
If the combination above produce only 1 peak of result, then 109 samples out of 480k is enough. The reality is your may have several peaks result. If you have says 10 peaks then at the same ratio you should have 10*109 sampling. Then you will get 10 values of 10 peaks to compare in final step.
The other setting is linked to sampling method and probability significant that is the result of Z-distribution.
I do not use it since i find it is too rough for my strategy.
Nobody can answer your question because the performance of Genetic Optimization rely on your criteria, data, your strategy. Try to plot it with 3D Optimizer.
If the combination above produce only 1 peak of result, then 109 samples out of 480k is enough. The reality is your may have several peaks result. If you have says 10 peaks then at the same ratio you should have 10*109 sampling. Then you will get 10 values of 10 peaks to compare in final step.
The other setting is linked to sampling method and probability significant that is the result of Z-distribution.
I do not use it since i find it is too rough for my strategy.