Genetic Vs Default Optimizer Reliability

[MisterGee]

Few questions about the optimizers.

To preface this post: when optimizing strategies with lots of variables or really large variable ranges, using the default optimizer takes an obscene amount of time to complete, however the genetic optimizer completes the task in, at most, minutes. So..

1) I've noticed with the GO, I can run a 5-100 generation optimization on any parameter and get a different group of top performing variables every time. Is this normal?

2) Will a completed default optimization give more or less reliable results than the GO?

3) Is there a # of generations for the GO that will give the most accurate results? Or are the default setting for the GO the ideal combination?

4) What, besides reducing the # of variables or data ranges, can I do to increase the speed of the default optimizer? (computer hardware?)

5) How reliable are the results for optimizing on Max Sharpe Ratio? (I've optimized a few strategies and gotten SR's of up to 700..)

6) Not a reliability question but what does optimize on max probability do? Results usually return really terrible results even on strategies back test/optimize well on other factors.

Thanks

[NinjaTrader_ChelseaB]

Hello MisterGee,

From the help guide:
"Genetic algorithm (GA) is another optimization technique that can be used to optimize your NinjaScript strategy parameters with. The Default optimization algorithm tests every single combination of parameters within the optimization range to find the best result possible, but since it is testing every combination possible, this requires a lot of computing power and time to complete. The GA can be more suited for these extensive optimizations because it is able to approximate best solutions without testing every combination through the evolutionary theory borrowed from biology."
...
"The GA determines its solution through the following steps:
1. Begin with an initial population size consisting of randomly selected individuals (parameter setting combinations)..."
http://ninjatrader.com/support/helpG..._algorithm.htm

An optimization will run through all possibilities. A Genetic Optimization will not run through all possibilities but will instead use randomization to spot test different parameters looking the best results then generations are spawn off of those results to further try and find the best results. Think of the first pass as random seeds.

Yes, the Genetic Optimizer will give different results as the process introduces randomization to get a wide range of scatter points.

http://ninjatrader.com/support/forum/showthread.php?p=297676#post297676

2. I cannot comment on if using the Optimizer vs the Genetic Optimizer is more reliable as these work in very different ways. Through testing you may find which is more suitable for you, or if you may want to do both. This thread will remain open for any community members that would like to offer their opinion.

3. I also cannot comment on what values to use or how to get better results in any way, however, this thread will remain open for any community members that would like to share their experience.

4. The CPU single threaded performance is the typically the largest bottleneck when it comes to speed of backtesting. However, the motherboard bus speed, and memory will come into play. Also, the speed at which the hard drive is read and written will impact performance as well.

5. I cannot comment about reliability, however, this thread will remain open for any community members that would like to share their experience.

6. Below is a link to a thread where developer Brett has a few notes on Max Probability. The gist is that this is the probability that the trade will be a positive trade.
http://ninjatrader.com/support/forum...794#post257794