tag:blogger.com,1999:blog-107568321062020427.post6078877665892275814..comments2023-09-19T00:48:06.643-07:00Comments on Intelligent Trading: Genetic Algorithm Systematic Trading Development -- Part 3 (Python/VBA)Intelligent Tradinghttp://www.blogger.com/profile/17765336450326139518noreply@blogger.comBlogger6125tag:blogger.com,1999:blog-107568321062020427.post-41579329148869845542011-09-08T22:54:21.918-07:002011-09-08T22:54:21.918-07:00It seems like a reasonable approach assuming you a...It seems like a reasonable approach assuming you are walking forward the data sets. I.e. the new culled population is evolved, but the results of that evolution are tested on new out of sample data. <br /><br />One could argue that any type of GA is curve fitting; but if the OOS data performs comparably well to IS, and you have some understanding as to what behavior it's capturing, then it should give you sufficient confidence to continue live. You can also monitor conditions and see if they are behaving as expected (given your back-testing information).<br /><br />Cheers,<br />ITIntelligent Tradinghttps://www.blogger.com/profile/17765336450326139518noreply@blogger.comtag:blogger.com,1999:blog-107568321062020427.post-25961826688262189492011-09-08T02:53:56.297-07:002011-09-08T02:53:56.297-07:00Hi,
I know this is an old post, but it's very...Hi,<br /><br />I know this is an old post, but it's very relevant to the challenges I face at the moment.<br /><br />I've been working on a high-frequency intra-day model for some time now. I use neuro-evolution (NEAT to be specific) as a vehicle. <br /><br />To combat curve fit, I've adopted a strategy whereby I back-test evolved candidates over my OOS data as soon as they are born. When a newborn candidate performs similarly over the OOS(compared to in sample) data, the candidate is set aside for later use. When the evolution process stagnates, I use these candidates which I set aside to seed a new population, and start the evolutionary process again.<br /><br />My question is : Am I curve fitting to all of my data(OOS and in sample combined) here? <br /><br />I'm sure the answer is not that simple, but I'd appreciate any thoughts.WaShierhttps://www.blogger.com/profile/05472384825203654857noreply@blogger.comtag:blogger.com,1999:blog-107568321062020427.post-47685917034472062482010-02-23T18:52:04.887-08:002010-02-23T18:52:04.887-08:00Hi Brad,
I wrote my own custom scripts for this e...Hi Brad,<br /><br />I wrote my own custom scripts for this example. There are a few Python GA algorithms available on the web; pygene comes to mind. However, you'll have to write your own custom scripts for the fitness function as there aren't many examples with this type of application.Intelligent Tradinghttps://www.blogger.com/profile/17765336450326139518noreply@blogger.comtag:blogger.com,1999:blog-107568321062020427.post-38350226730927500912010-02-23T18:02:37.992-08:002010-02-23T18:02:37.992-08:00What python package did you use to create the outp...What python package did you use to create the outputs in this example? I couldn't find a package called "Python GA" with a quick Google search. Thank you!Bradnoreply@blogger.comtag:blogger.com,1999:blog-107568321062020427.post-45407087791495264682010-02-20T15:22:05.457-08:002010-02-20T15:22:05.457-08:00Thank You,
Yes, it is curve fit, and I was planni...Thank You,<br /><br />Yes, it is curve fit, and I was planning to discuss that before this segment is finished. That does not preclude it from being a useless exercise, however. <br /><br />It's not so easy, however, to build and back-test systems=) Hopefully, you understand the scope of the site is to show you how to augment successful system building using modern techniques.<br /><br />Thanks for your observations.Intelligent Tradinghttps://www.blogger.com/profile/17765336450326139518noreply@blogger.comtag:blogger.com,1999:blog-107568321062020427.post-38065130522128569642010-02-20T12:58:50.208-08:002010-02-20T12:58:50.208-08:00What you're getting here is curve-fit system. ...What you're getting here is curve-fit system. Try it in out-of-sample data and it will fail miserably. Curve fitting is very easy.Anonymousnoreply@blogger.com