Genetic Optimizer Settings

[Crassius]

Can someone point me to some guidance on selecting proper settings for the GO:

# of Generations
Crossover Rate
Generation Size
Min Performance
Mutation Rate
Mutation Strength
Reset Size %

I'm not really sure what Reset Size % is, but the other terms I have some understanding of what they do.....

what I don't have any idea of is how to select sizes so one gets a thorough search, but not selecting such a high settings that you defeat the purpose of using GO instead of brute force.

[NinjaTrader_Bertrand]

Crassius, all paramters of the GO are explained here in the help guide, including the reset size -

http://www.ninjatrader.com/support/helpGuides/nt7/genetic_algorithm.htm

There can be unfortunately no 'rule of thumb' for the settings, as they would be very individual to the problem solved with the GO. If you find the GO is behind in found optimal solutions compared to the exhaustive search, try first increasing the # of Generations and their size, so a deeper search field would be covered...

[Crassius]

Some Genetic Algorithm Learning Resouces

Vielen Dank, Bertrand.

I will post a few resources that I found helpful in researching this that go beyond NinjaTrader specific support for others that may find this thread through search in the future.

I found these examples on the MathWorks documentation site that were helpful to read through even though it is for different software.

http://www.mathworks.com/help/toolbox/gads/f6691.html

I have attached two scholarly articles that were focused, even if they are less fun to read than a comic book.

Their subjects are obvious from their titles.

Attached as PDF's please find,

Optimal Population Size and the Genetic Algorithm, Gotshall and Rylander

Large Populations or Many Generations for Genetic Algorithms, Vrajitoru

[staycool3_a]

Vielen Dank, Bertrand.

I will post a few resources that I found helpful in researching this that go beyond NinjaTrader specific support for others that may find this thread through search in the future.

I found these examples on the MathWorks documentation site that were helpful to read through even though it is for different software.

http://www.mathworks.com/help/toolbox/gads/f6691.html

I have attached two scholarly articles that were focused, even if they are less fun to read than a comic book.

Their subjects are obvious from their titles.

Attached as PDF's please find,

Optimal Population Size and the Genetic Algorithm, Gotshall and Rylander

Large Populations or Many Generations for Genetic Algorithms, Vrajitoru

Is there anyone here who regularly uses the GO? Can you please share your settings?

[simple_goodoboy]

Is there anyone here who regularly uses the GO? Can you please share your settings?

Hello calhawk01,

Why not just use the default GO settings already set by NT? I use the default settings. The default settings should be robust enough for backtesting a strategy.

Also, you can just compare the GO optimization of your strategy versus the Default Optimization. If they are similar, then use the default settings.

Also, you can change each GO settings by 1 and compare the default settings and see if test make a big difference.

I start by using GO to give me a general idea of profitability of an optimized strategy. Then I nail down to using short span testing (maybe 6 months at a time) with the default optimizations if I see something that looks profitable.

This is just my opinion as I feel GO is too complex for my simple thinking and the default settings is good enough.

[pel11]

I tend to use at least 50 Generations. The typical setting for use of a GA in computer science has always been about 100 generations and 20 chromosomes (particles).