Genetic Optimizer and memory usage

[neb1998]

Ok a few things i have learned in this short 2 months with ninjatrader.

Optimizing is a much different animal than with a 32 bit program and multicharts.

Generation Size is directly proportionate to the # of cores and the amount of memory that the GO will use.

If you set the Generation size to 1, it will only use one core.

As you change to 2, it will use 2 cores, 4 or more will use all cores.

Then as you increase to much larger values, watch your ram usage also increase drastically.

I mention this because coming from MC and the GRAIL, we would typically specify a gen size of around 40, generations around 50 and about 2000 iterations...the NT GO uses a type of convergence instead of max iterations although it should not affect performance of the optimizer to much once the right settings are dialed in.

I am starting this thread to see what feedback others have noticed with complex systems, the best / gen size versus # of generations and other settings to achieve good results without turning the GO into a brute force optimizer.

Hope this is clear

[Trader.Jon]

neb,

Welcome to the optimization opportunities in NT!

Personally, I use the PHGenetic GO (3rd party freeware) script in NT7, of the ones available:

http://www.bigmiketrading.com/ninjatrader-programming/4273-index-genetic-optimizers-ninjatrader.html

I found the following settings to be the most useful in terms of results (screen shot to be attached)

Jon

[neb1998]

Thanks Jon. I will try these and let you know. I am trying to setup with a 3rd party ninjascript consultant right now to make some changes to how the data is stored in text files for walkforward runs. I am looking to do anchored walkforwards, say 5 years of data and about 70-80% insample, and the rest out of sample. And have a fitness function that is calculated by both IS and OOS. This kind of testing is similar to what the GGO did in The Grail System...although it only works with multicharts and TD recently bought it.

So having to go the custom route. It accelerates system development 10 fold.