Algorithm Optimisation Championship. - page 104
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Yeah, and?
It is.)
The championship FF consists of 255 such features, a total of 500 parameters, which are jumbled together, and the pitch is mildly less than what you have in the picture.
It's not the pitch, it's the grid (resolution) of the graphic.)
I won't distract you further.))
I have found a way to write the FF so that the global maximum for the referee (but not the participants) is known, it will get this value already when generating the unique sequence for the FF. As a result, it will be possible to compare the results with the actual value of the FF maximum.
I've racked my brains trying to figure out how to achieve this, as a result I've lost time and now I don't have time to provide the source code, sorry, I will provide it tomorrow.
But now I have a possibility to compare results with the real value of FF maximum and it's very important! The function is not so complicated as I want it, but it is complicated enough not to let my algorithm to get more than 40% accuracy for 10000 FF runs.
I have found a way to write the FF so that the global maximum for the referee (but not the participants) is known, it will get this value already when generating the unique sequence for the FF. As a result, it will be possible to compare the results with the actual value of the FF maximum.
I've racked my brains trying to figure out how to achieve this, as a result I've lost time and now I don't have time to provide the source code, sorry, I will provide it tomorrow.
But now I have a possibility to compare results with the real value of FF maximum, and it's very important! The function is not so complicated as I want it, but it is complicated enough to prevent my algorithm from getting more than 40% accuracy after 10,000 FF runs.
Perfect.
It turns out, that in order to find a compromise with participants and to organize the contest properly, you just need to rack your brains a bit...
About the proverbial universality you talk so much about - I've come to the conclusion that it doesn't always produce the best results.
1. The universality of a solution is always relative, because the solution is limited to the specifics of the problem domain - and hence - the solution is never absolutely universal. When expanding the problem domain, a "universal" solution will always fail. It will have to be reworked.
2. No universality appears from scratch, but is a consequence of a long process of development, generalization of problems and adaptation of the solution. Hence, non-universal solution is the first step towards universal solution.
3. The universality of the solution does not mean the effectiveness of the solution. I think that these two notions are not directly connected and do not depend on each other.
Striving for universality makes one adapt the solution to a widening range of problems, which of course may reduce the effectiveness of the solution in each particular case.
My algorithm for text mining is universal enough for text mining, and can absolutely precisely identify any string in the minimum number of accesses to the FF. Perhaps its further development, may lead to finding maxima of unknown analytic functions. But will it still be effective in this case? I'm not sure.
And so, to understand how we can make a universal algorithm, we need to generalize the range of problems and understand the general mechanism of their solution.
Let's start by summarizing the parameters.
The main parameters with which the algorithm works to find the maximal function value and text key:
1. The number of parameters passed to the FF.
2. The range of the values of the parameters passed to the FF.
3. step (minimal difference between the values).
4. The value received from the FF.
In the absence of more basic parameters, the solution, even without any extra efforts, may turn out to be universal enough...
The search mechanism in these two types of these problems can be generalized, which I will try to do.
It turns out that in order to find a compromise with the participants and to organise the competition properly - you just need to rack your brains a bit...
I'm a little confused as to what the championship is about in the first place. They are a fraction of a second by any software. And not just the highs, but the profiles themselves.
Here's an example, I did recently. And this is function processing.
Like, - long before he repeated it?
And when the function is not known to the software, may the software determine the extrema? The point, as I understand it, is that the participants send the ''black box'' a set of parameter values as an array doudle[x1, x2, xn].
They get the value of the function from the "black box" into their algorithm, send the next set and so on until they find the extremum. We need to make as few calls as possible in the MT environment. I think it is a useful thing in the household.
... And all in a MT environment. I think it's the right thing for the household.
Why? What is the practical use? Examples of tasks?
I've written here in relation to trading.