The current optimizers are rather limited for strategy development
Vince
Vince
Rename
Have you tried out the finantic.Optimizers extension?
No, not yet. I am waiting for your built-in optimizer to permit a head-to-head test. As I found out in my tests previously, LenMoz's PSO had improved performance compared WL6 Genetic Optimizer.
Head-to-Head comparisons help me understand the details that make one approach more suitable for my work.
Vince
Head-to-Head comparisons help me understand the details that make one approach more suitable for my work.
Vince
What is a Head-to-Head test?
I will be a while before we deliver any in-house advanced optimizers. I'm up to my neck (brain rather) in NeuroLab development!
Understand Glitch! I will continue to use WL6 in the interim for system development until you get to the Optimizer. WL6 has most (but not all) of the tools I need, which when partnered with my own tools, allows me to do the job.
Vince
Vince
Eugene,
A "head-to-head" test uses a series of more challenging optimization problems to assess performance. Eventually one of the optimizers will be unable to identify the best solutions. Since the process is stochastic, it takes a number of runs to establish. I did spend over a week a couple of years ago when LenMoz introduced his PSO. Those tests were extremely illuminating.
Vince
A "head-to-head" test uses a series of more challenging optimization problems to assess performance. Eventually one of the optimizers will be unable to identify the best solutions. Since the process is stochastic, it takes a number of runs to establish. I did spend over a week a couple of years ago when LenMoz introduced his PSO. Those tests were extremely illuminating.
Vince
Do you have a link or any background info about this PSO optimizer?
DrKoch,
in 2014, I designed and wrote WL6 PSO. It implemented eleven different PSO algorithms. You can find info on that PSO here...
http://www2.wealth-lab.com/WL5Wiki/ParticleSwarmOptimizer.ashx?HL=particle,swarm
As you may have guessed by my questions on this forum, I've been working on a WL7 implementation. I have a rudimentary version working but am unlikely to make it generally available. Exhaustive will not work for my strategies due to millions of permutations..
Also, I saw that one of the extensions to WL7 includes a PSO Optimizer. I don't know anything about that one.
in 2014, I designed and wrote WL6 PSO. It implemented eleven different PSO algorithms. You can find info on that PSO here...
http://www2.wealth-lab.com/WL5Wiki/ParticleSwarmOptimizer.ashx?HL=particle,swarm
As you may have guessed by my questions on this forum, I've been working on a WL7 implementation. I have a rudimentary version working but am unlikely to make it generally available. Exhaustive will not work for my strategies due to millions of permutations..
Also, I saw that one of the extensions to WL7 includes a PSO Optimizer. I don't know anything about that one.
One comment about optimizers...
Stochastic optimizers continue to be an active area of research in academia. As people have continued to identify more difficult highly multi-dimensional real-world problems, the research community has continued to develop better approaches to solve them. For the types of problems that trading strategies involve, there will always be a need to keep pace with these improvements.
There are some (e.g. differential evolution []https://en.wikipedia.org/wiki/Differential_evolution) that have shown promise in complex problem sets (https://www.frontiersin.org/articles/10.3389/fbuil.2020.00102/full). All of these newer approaches have attempted to address many of the complications of previous algorithms (e.g. excessive adjustable parameters) that make setup and operation difficult for problems with expensive computational costs for calculation of the objective function.
Vince
Stochastic optimizers continue to be an active area of research in academia. As people have continued to identify more difficult highly multi-dimensional real-world problems, the research community has continued to develop better approaches to solve them. For the types of problems that trading strategies involve, there will always be a need to keep pace with these improvements.
There are some (e.g. differential evolution []https://en.wikipedia.org/wiki/Differential_evolution) that have shown promise in complex problem sets (https://www.frontiersin.org/articles/10.3389/fbuil.2020.00102/full). All of these newer approaches have attempted to address many of the complications of previous algorithms (e.g. excessive adjustable parameters) that make setup and operation difficult for problems with expensive computational costs for calculation of the objective function.
Vince
Hi LenMoz,
> unlikely to make it generally available
you might contact me by email: rene dot koch at finantic dot de.
Probably we could cooperate in making your optimizers generally available.
> unlikely to make it generally available
you might contact me by email: rene dot koch at finantic dot de.
Probably we could cooperate in making your optimizers generally available.
Your Response
Post
Edit Post
Login is required