Discussion of article "Population optimization algorithms: Saplings Sowing and Growing up (SSG)" - page 8

New comment
 
mytarmailS #:
Why?

Would you be able to drive a car on an unfamiliar road using only your rear view mirrors with your windscreen taped up?

 
Nikolai Semko #:

Can you drive a car on an unfamiliar road using only the rear view mirrors with the windscreen taped up?

In an ever-changing, non-stationary environment, operating on constant parameters is naive to say the least....

As an idea.
The optimisation surface changes slowly if you think of it as dynamic..

If we convert it into a time series, we can try to predict this dynamics... So we can know the optimal parameters of the TC for tomorrow.
 
mytarmailS #:
In an ever-changing, non-stationary environment, operating on constant parameters is naive to say the least....

As an idea.
The optimisation surface changes slowly if you imagine it in dynamics...

If we convert it into a time series, we can try to predict this dynamics... Thus we can know the optimal parameters of the TS for tomorrow and yesterday.

In a changing environment (in this case, the optimisation surface) it is not important where you stop at a given point in time, but where this point will go down or up in the next moment of time. Moreover, and this is not important, it is important whether it will move upwards with acceleration or deceleration.
I.e. if you choose a point on a rising hill, but at the next moment the rate of growth starts to slow down, it will be a worse decision than if you choose a point on a falling trough, but if at the next moment the rate of fall slows down.
Without a predictive model with probability > 55% any strategy is nothing.

 

There is a good criterion for a correct working strategy.
The funds line should be above the balance line more than 50% ( better 60%) of the time.
Go to the top signals and look at the bottom chart where these two lines are.
Almost all of them have the green line (funds line) below the balance line most of the time.
I don't understand why people grow lots instead of growing profits.



this is how it should be


 
I will tell you about the standard GA, what it is and why it is not in the tests.
the standard GA is one of the oldest AOs, and at the same time one of the most powerful. it is a binary algorithm, hence all its limitations, the number of opt parameters and their step. the point is that a binary chromosome has a length limit and there is nothing you can do about it.
besides the limitation on the length of the chromosome (and this is not only the number of opt parameters) there are other disadvantages, such as the impossibility to apply dynamic step and others, not to mention the impossibility to apply zero step.
Despite all the disadvantages, it is still one of the most powerful algos today.
tests in the articles are carried out with zero step, so the standard Ga cannot be tested and entered into the table, it simply cannot be used for these tests. however, I have tried to test Ga with the minimum possible step for tests with 2 me earlier, and now with 10 parameters, and it almost completely converges on all functions! but the use of 100 and more parameters as in the tests can not be applicable, there is a limitation on the length of the chromosome.
bottom line. the standard ha (binary) is morally and physically obsolete. there is no aim to offend the developers, it is just a fact.
at a time when ships are exploring the vastness of the big theatre, i.e. all sorts of chat rooms advise how to live and how not to live, it is time to consider the possibility of adding several AOs to MT5, it will expand the opportunities for the user undoubtedly.
 
"I'll say one thing, no offence intended" (the original sounds different).
it makes no difference whether the FF is static or dynamically changing! the point is how fast AO is able to converge. i.e. AO is able to adjust parameters faster than the FF is changing (it should be).
For a long time I was thinking how to apply neural network to increase convergence of AO..... It seems that times have come such that even such a thing is possible.
It makes no difference to AO whether the FF changes or not, it still searches blindly. it is the search strategy that determines the success of a seemingly hopeless mission.
 
Andrey Dik #:
"I'll say one thing, no offence intended" (the original sounds different).
it makes no difference whether the FF is static or dynamically changing! the point is how fast AO is able to converge. i.e. AO is able to adjust parameters faster than the FF is changing (it should be).
For a long time I was thinking how to apply neural network to increase convergence of AO..... It seems that times have come such that even such a thing is possible.
It makes no difference to AO whether the FF changes or not, it still searches blindly. it is the search strategy that determines the success of a seemingly hopeless mission.
Don't confuse the concepts
The FF is a function for calculating error.
And OD is an optimisation surface.

Calling the former the latter is not correct.
 
Nikolai Semko #:

In a changing environment (in this case, an optimisation surface), it doesn't matter where you stop at a given point in time, it matters where that point goes down or up in the next one

What did I write? Don't you read?

 
mytarmailS #:
Don't confuse the concepts
FF is a function for error calculation
And OP is an optimisation surface

Calling the former the latter is not correct

No, you are confused.

FF is a fitness function, i.e. the value of some evaluation criterion, the whole area of FF values is a surface (can be multidimensional).

and what does "function for error calculation" have to do with it? FF is a general concept for any evaluation criteria, not just "function for error calculation"

And "OP" is not a concept I've encountered anywhere at all.

 
Andrey Dik #:

No, you're confused.

FF is a fitness function, i.e. the value of some evaluation criterion, the whole area of FF values is a surface (can be multidimensional).

and what does "function for error calculation" have to do with it? FF is a general concept for any evaluation criteria, not just "function for error counting"

And "OP" is not a concept I've seen anywhere at all.

The fitness function is a subspecies of the target function, aka the fitness function, fitness is the error.



12345678910111213
New comment