Genetic optimization settings similar to TradeStation

[peterfin]

Hello,

I am trying compare some of my workflows using genetic optimization between my Amibroker and Tradestation.

I am not any expert on genetic optimization.

Does anyone have tips what would be the closes solution for Amibroker setup so I get similar tool like TradeStation offers? That means number of Generations, Population size, mutation rate etc:

I have read https://www.amibroker.com/guide/h_optimization.html and understand I can setup Runs and MaxEval in Amibroker’s optimizers. .

Anyone tried to compare genetic optimizations TradeStation/Amibroker and can share tips how to setup Amibroker so I have similar genetic optimization tool like in TradeStation?

Thank you.

[Steve]

AmiBroker does not include a genetic optimizer, but if you have the software development skills, you could write a genetic optimizer plugin with the AmiBroker Development Kit.

AmiBroker includes Covariance Matrix Adaptation and Particle Swarm optimizers, but they use different optimization approaches.

[Milosz]

@peterfin you might find this Tomasz's reply interesting:

Smart optimizer already exists and is multithreaded, was: Multi-Threaded Genetic Optimizations - #2 by Tomasz

... and this one:

Backtesting Two (or more) Strategies Simultaneously

AmiBroker can do ALL the things that Tradestation can PLUS zillions of others. And Tradestation has only SINGLE SECURITY backtester. TS does not even have true portfolio-level backtesting so TS backtesting is very limited. AmiBroker is in different league, orders of magnitude better and provides features and speed not available anywhere else. In fact Tradestation guys wanted to buy AmiBroker's technology for lots of $$$ but we said NO, because we cared about our customers and did not want them …

[Tomasz]

AmiBroker uses many, not one, smart evolutionary optimizer engines like for example Particle Swarm Optimization, Tribes and CMA-ES. Those are modern cutting-edge alternatives to pretty old and obsolete stuff that TS uses and are far superior to genetic, converging much faster and requiring much less parameters so you don't really need all that params you have in TS. In most cases you are fine with defaults and even if you need to modify parameters it is only number of evaluation/runs that needs to be modified, nothing else.

An absolute MUST READ is the manual: http://www.amibroker.com/guide/h_optimization.html everything you need to know is written there.

[PWFores]

Just try. There is risk for curve fitting.
Try optimizing on more similar tickers (portfolio) and create a system
and parameter settings that works on different tickers and timeframes
to have the best OOS chance.