Algorithm Optimisation Championship. - page 10

 
Boris:
I was just browsing and came across 2 Igor Volodin, but then saw that he himself paid attention. Hence deleted my empty post. And in terms of participation I don't possess that level in programming. Sorry for the inconvenience! Good luck to everyone in this interesting competition!

No no, feel free to participate. There will be simple examples of codes. Nothing difficult, take an example and tweak it a bit, or don't tweak it at all and leave it as it is and post it on your behalf (the rules don't forbid doing so).

And my level of programming is well below the average on the forum. You are not alone).

The list has already been dealt with, thanks.

 
Dmitry Fedoseev:

Not all functions have noise. But some do, which is why the gradient descent method fails.

It is not for nothing that the name "genetic" appeared; analogies from nature work well: inbreeding, mutation.

At first, I also wanted to at least partially use the gradient descent method, but I gave it up completely.

So the FF array (area of values) is filled with values "generated" by some kind of mathematical function? The kind that "draws" a parabola and a hyperbola on a school notebook graph?
 
Dmitry Fedoseev:

Not all functions have noise. But some do, which is why the gradient descent method fails.

It is not for nothing that the name "genetic" appeared; analogies from nature work well: inbreeding, mutation.

At first, I also wanted to at least partially use the gradient descent method, but gave it up completely.

It was a mistake to refuse it completely. If possible - you can and should use multiple search algorithms together. This gives more searching possibilities. On smooth, continuous sections of FF, Newtonian and gradient descent will take the lead, while stochastic methods will help on noise, chasms and peaks. So the combined algorithms have a better chance of winning.

ZS. I don't have a combined one.

 
Реter Konow:
So the FF array (range of values) is filled with values "generated" by some kind of mathematical function? The ones that "draw" the parabola and hyperbola on the graph in a school notebook?
If so, there is an order after all...
 
Реter Konow:
If so, there is an order after all...
"Maybe there is, maybe there isn't" (c)
 

By the way, yes, if participants were obliged to save the dynamics of finding the best values to a file, then it would be possible to compare graphs later of how the algorithms moved towards their goal.

It is very revealing.

 
Andrey Dik:
"There may or may not be" (c).

At the moment, for myself, I don't see any solution for developing a search strategy applicable to chaos conditions.

Consequently, whether or not order is present in the FF array (value domain), I will still assume it.

The surface created by the function and its topography.

The intricacies of its curves and random noises will certainly get in the way, but I think they can be recognized.

And it is probably possible to avoid noise altogether, if you follow the search strategy?

(I apologize if for people in the topic my reasoning looks like nonsense, I'm just trying to reason).

 
Реter Konow:

At the moment, for myself, I don't see any solution to developing a search strategy in chaos.

Consequently, whether or not order is present in the FF array (value domain), I will still assume it.

The surface created by the function and its topography...

The intricacies of its curves and random noises will certainly get in the way, but I think they can be recognised.

And one can probably avoid noise altogether if one follows a search strategy?

(I apologize if my reasoning looks like nonsense to people in the subject, I'm just trying to reason).

So already suggested - crossbreeding and mutation, culling the bad ones and replacing them with new random ones.
 
Dmitry Fedoseev:
So already suggested - inbreeding and mutation.
Let's think...
 
Andrey Dik:

The Championship is a great opportunity to test your algorithms under tough competitive conditions, which are tougher and more demanding than those encountered in everyday life. It is a chance to make sure that your algorithm can no longer be better, which means confidence in the possibility of solving life's challenges ahead, or to make sure that further improvement and improvement of the algorithm is necessary or possible.

Strange championship, maybe you should first compare your algorithms with those already implemented, e.g. with ALGLIB. And then ...