selecting the hardware (computer) to use - page 2

New comment
 
Vladimir:

Wondering what developers offer for optimisation.

Do their methods cope with gully functions?

credit )))))
 
Alexey Volchanskiy:
Why optimize during debugging? First of all, make it work without errors, the usual testing will do. And the cloud doesn't work in the tester, only in the optimizer.
I'm not talking about clicking in the debugger, of course. Debugging in the broader sense, tuning of the algorithm. An EA without parameters is a cherished dream, but so far no such thing has been invented )
 
Alexey Volchanskiy:
credit )))))

I once again reread the help on optimising EAs (I haven't used it myself, it's limited to the help). According to it, there is not even the simplest methods of fastest descent, co-ordinate descent, Monte Carlo. Either brute force with its dimensionality curse, or unknown to me genetic algorithm. Both implemented methods are resource intensive, time consuming.

Do I understand the situation correctly?

If yes, then why there are no add-ons over the strategy tester that drastically reduce the number of runs, implementing optimization by long-studied methods; instead the movement is towards increasing the required resources?

 
Vladimir:

I once again reread the help on optimising EAs (I haven't used it myself, it's limited to the help). It shows that there are not even the simplest methods of fastest descent, co-ordinate descent, Monte Carlo. Either brute force with its dimensionality curse, or unknown to me genetic algorithm. Both implemented methods are resource intensive, time consuming.

Do I understand the situation correctly?

If yes, then why are there no add-ons over the strategy tester that drastically reduce the number of runs and implement optimization using well-tested methods?

You cannot technically make an add-in, there is no such API. We just have to write our own tester. There is some opportunity to analyze a run through the

OnTesterPass();


The genetic algorithm is widely used and there are some articles on it if you don't know it. All other questions to the developers.
 
Vladimir:

I wondered what developers offer for optimisation.

Do their methods cope with gully functions?

What's the problem with gully functions?
Vladimir:

I once again reread the help on optimizing Expert Advisors (I haven't used it myself, it's limited to the help). It shows that there are not even the simplest methods of the fastest descent, co-ordinate descent or Monte Carlo. Either brute force with its dimensionality curse, or unknown to me genetic algorithm. Both implemented methods are resource intensive, time consuming.

Do I understand the situation correctly?

If yes, then why don't you find add-ons over the strategy tester that drastically reduce the number of runs, implementing optimization by long-studied methods; instead the movement goes in the direction of increasing required resources?

Do you have examples to prove the superiority of algorithms other than genetic ones? And exactly for complex stepped functions, with wide horizontal sections, with sharp peaks and troughs, such as optimized functions in discrete data programs (Expert Advisors)?
 
Vladimir:

I once again reread the help on optimising EAs (I haven't used it myself, it's limited to the help). It shows that there are not even the simplest methods of fastest descent, co-ordinate descent, Monte Carlo. Either brute force with its dimensionality curse, or unknown to me genetic algorithm. Both implemented methods are resource intensive, time consuming.

Genetic algorithm drastically reduces the number of runs compared to a full search. If function is smooth enough, it leads to optimization much-much faster. If the function is strongly "torn", even a full enumeration is useless, "torn" functions mean algorithm instability, and the "optimal values" found will most likely be random spikes, rather than optimum points.
 
George Merts:
The genetic algorithm drastically reduces the number of runs compared to the full search. If the function is smooth enough, it leads to optimization much, much faster. If the function is strongly "jagged", even a full search is useless - "jagged" function means instability of the algorithm, and the "optimal values" found are likely to be random outliers, and not the optimum points at all.
Please give a definition (you can give your own) of "optimum points" and/or "optimum values".
 
Andrey Dik:
Please give a definition (you can give your own) of "optimum points" and/or "optimum values".
This is the set of parameters giving the highest optimised value (balance, recovery or whatever) that is stable, i.e. a small change in the input parameters does not result in a large change in the optimised value. If this happens, it is not the optimum value, but just a random function outlier.
 
George Merts:
This is the set of parameters that give the highest optimised value (balance, recovery or whatever) which is stable, i.e. a small change in the input parameters does not result in a large change in the optimised value. If this happens, it's not the optimum value, but just a random function outlier.
There, that's exactly the answer I was expecting, because there can be no other, logically related to your previous post, answer.

And I asked to emphasize the contradiction. Think about it, try to link it together. If you can't, I will help. The point is that your post is an example of a widespread misconception.
 
Andrey Dik:
If that doesn't work, I will help. The point is that your post is an example of a widespread misconception.
And where do I have contradictions and misconceptions? Why speak in riddles (let's speak in riddles?). If you have a concrete objection, go ahead and post your thoughts and we'll discuss it.
123456789...14
New comment