Discussing the article: "Role of random number generator quality in the efficiency of optimization algorithms" - page 15

[Deleted]  

Now that the dog is wagging its tail (empiricism by optimisation) and not the other way around, we can consider any optimisation algorithm for a conditionally stationary process.

In this case we can use the terminology of finding global and local minima.

But not for optimising unknowns and fitting to abstract minima or maxima.

But even in this case, AO tends to overtraining (pre-fitting), then validation techniques are used to determine the robustness of certain parameters from learning theory.

 
С
Yuriy Bykov #:
Unfortunately, it became even less clear what it was about.
fxsaber #:

Obtuse language+forum format = misunderstanding with high probability.

Those who wish to participate in constructive discussion of the problem of searching for robust solutions can write to me in private messages. We will organise private chat with participants by invitation.

And participation in conversations that do not imply constructive dialogue is not on my current task list.

[Deleted]  
If all this were written down somewhere, we wouldn't have to brainwash people with maxima and plateaus and other bullshit that has no meaning outside the context of the stationarity of the process.
[Deleted]  
Even when the conditions are met, brute force Monte Carlo works as well as the whole bunch of algorithms. That is, just choose random values of parameters n times and validate.
 
Nikolai Semko #:
I’m known on this forum as a staunch opponent of trading systems with numerous input parameters (apart from risk management parameters) that require optimisation. I consider this to be just another covert form of backtesting. Haha.

It would be interesting to have a discussion with you, but I’m not particularly keen on getting bogged down in theory. The results should speak for themselves :)

In practical terms, this debate ultimately comes down to which market model you subscribe to. If you’re convinced that the market is a ‘monolith’, where the rules of ‘profitable trading’ don’t change over time (or, for that matter, from one instrument to another or from one time frame to another), then you’ll stick to the idea of minimal optimisation, as you wrote. But if, like me, you were to subscribe to the idea that the market is a constantly changing entity, where rules appear and disappear, sometimes quickly, sometimes more slowly – allowing traders to exploit ‘windows of inefficiency’ which, by definition, are not static – then you would lean towards intelligent, dynamic self-optimisation, as I do. I hope the general principle is clear