The genetic algorithm and its possible applications - page 10

 
Реter Konow:
In order to create, you need an idea. A concept of creation. You argue that it's not necessary. That it's enough to shake the "jar" with particles more violently and everything will build itself - the universe and living beings... Well, shake it, then...

I would work on concepts first. Of course, there is much more to "shake things up" quantitatively than there is to think about qualitatively... Well, create a suitable GA and let it bubble with entities and we'll observe.

By the way, I created a thread called "algorithmic centrifuge". I expressed similar views there. I was thinking of creating a "parameter mixer".

Yes, shaking is a universal way to create something new. For instance, if a man and a woman shake the bed a bit, a new life will be born.

 
khorosh:

Yes, shaking is a universal way to create something new. For example, if a man and a woman shake the bed a little, a new life is born.)))

There's probably a law in the universe - everything starts with an Idea and ends with a mindless shaking... Art, for example, has proved it. Now the Market is proving it too... Next, biotechnology will prove...

 
Реter Konow:
And what is the complexity of systems created by "shaking"? This method is similar to the Rorschach test, where blots give birth to pictures. By the way, the real approach of modern artists is to splash paint on canvases, and "connoisseurs" will find something to admire themselves.)

Are there any serious results? Are working systems in place?

Who cares, you don't believe me anyway.)

Any AO is just a certain way of sorting (search strategy), but the search is always in a LOCAL direction in the search space! If one can find the best result non-randomly, then one can construct an analytical formula and directly calculate the best value of the function, why do we need AO then?

Certainly there are serious results, I have told, optimization of a form at maximization of strength and minimization of volume, search of new materials with the set properties with construction of multidimensional crystal lattices, but it so, in a flash.

 
Andrey Dik:

Any AO is just a certain way of sorting (search strategy), but the search always takes place in a LOCAL direction in the search space!

gradient descent method
 
Andrey Dik:

all it takes is poisoning the air and water to make the entire food base extinct, which man has been doing remarkably well lately. man can put on, roughly speaking, a gas mask, but birds and fish cannot.

viruses are rubbish! man is clever enough to make himself immune to infections and with regeneration and immortality like jellyfish, the only problem is man himself - by accidentally destroying all life on the planet man will become extinct (formally of course, in reality simply move to another suitable planet and start shitting there).

the funniest thing is that man can make his own food from almost anything, petroleum being the best, of course. so he does not particularly care about living food, non-living food is enough for him.

a perfect being is one who can modify his/her genome and create food from inanimate materials. humans are close to perfection. whether a perfect being needs morality is the question of a million.

Morality is paramount, only then survival, morality distinguishes us from animals.

 
TheXpert:
gradient descent method

gradient descent is also a random direction search that does not guarantee finding the global maximum, the "gradient descent" search strategy is based on the assumption that the surface will continue to change in the same direction accordingly, so it is extremely bad for functions with sharp transitions, kinks and holes.

there is no point in using any AO, albeit gradient descent, if the extremum can be calculated analytically rather than iteratively.

 
Кеша Рутов:

Morality is paramount, only after survival, morality distinguishes us from animals.

Man, with his lauded "morality", is in fact worse than animals. look at your wife - no wolf would kill for a rabbit's skin, nor a rabbit for a wolf, no matter how much they hate each other or are afraid of each other.

No one in the animal world eats their own kind just for the sake of looking younger.

 
Andrey Dik:
clearly
 
Реter Konow:
And what is the complexity of the systems created by "shaking"? This method is similar to the Rorschach test, where blotches give birth to paintings. By the way, the real approach of modern artists is to splash paint on canvases, and "connoisseurs" will find something to admire themselves.)

Are there any serious results? Are working systems in place?
 
AZAT KHALITOV:
One of the results is tested on my demo account - I have a signal in my profile, the second result reacts to very rare signals in a certain market condition - when the sentix index is above about 8 (ideally above 20 as in 2017 and 2018, but in early March there were 3 such events due to oil news - sentix reflects overall volatility of the entire European exchange, in 2019 3 events, while 2017 and 2018 every day 1 event, out of about 1000 events only 2 unprofitable). In general the results are like looking for onamalias in a price chart - when they occur price tries to return to normal. Somebody wrote here that it makes sense to use algorythmic approach, it does - you can use daily strategies but with free parameters and free coefficients for them and of course the result depends heavily on the function used in ontester() + inclusion of testerstop() in the algorithm for rejecting some results. Of course, the results are not perfect, for example, the Expert Advisor that is being tested as a signal now opens only sell orders due to limitations of the initial strategy (so to say, the inherent meaning) but I have understood how the strategy works and I will write a new algorithm with corrections for enabling Bay positions within a month. Immediately notice the tested result still can not be used - it's just a bare idea without improvement, without risk management + wrong stop loss (all bugs checked and need fixing, but so far not included in the working code) plus we need treatment of the news factor. So far the errors have not affected testing due to low volatility of the entire European exchange (this does not apply to local volatility, which is now observed). In general, everyone gets different results due to different implied meaning and different approaches to the optimization, implemented by the ontester() handler. The result is also affected by how you describe the algorithm of the initial strategy (small details - mutations for the final result) - in one result, an error in the description (not noticed before optimization) allowed for opening additional orders in the final result. Using genetic algorithm in this case is similar to using it in neural networks (a neural network is a complex function with a large number of parameters and one of the ways of training it is a genetic algorithm)
Reason: