Machine learning in trading: theory, models, practice and algo-trading - page 1337
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
In short, I said it would be the same thing, only in reverse
the power of theory
And I did not understand that it will be/still is the opposite :)
What's not the bruteforce method, playing with samples?And this is how you can combine two models from the last and penultimate experiment - though the separation was amplified to 0.55.
In general it is necessary to automate the process with the selection of sets of different models.
at a minimum, auto trait enumeration, at a maximum, target traits, too.
sample sizes have swapped places, and the test results are similar in both cases.
So everything is on automatic, we cut the sample and twist models around it - there is not much manual work here, once the process is set up.
No, the sample size remained, that is, if you agree that 30% here and in past experiments is the best indicator. The breakdown itself has changed, if conventionally in the first case studied for 2014-2016, in the second case for 2015-2017. There is an assumption that 2014 can be safely thrown out - we should try, then the difference is one year to study, but it is interesting that in the first case the validation was for 2017, and in the second case for 2014 (in fact still 2015 until April)!
Although in name I may be associated with the main character of your tale, but not in essence, because I just suggest to take into account the maximum of additional information from the experience of a trader, for example in my topic with templates -https://www.mql5.com/ru/forum/270216
You have a slightly different approach. I'm not suggesting that we impose specific deals on the MO and train on them. I'm suggesting that preliminary information should limit the search area. Like, Mr. Big is either in New York or the Bahamas, there are arrangements with the CIA - they will help, and finding and killing is your job.
There were many versions, at least what I remember, in one of the dummies was the MSUA-type enumeration, and there build polynomial signs
About SVM on the site he even wrote himself (for vector machine is the method of reference vectors). All right, SVM is linear, but due to polyfunctional features it is non-linear, "nuclear".
here is the site https://sites.google.com/site/libvmr/
This thing on mt4 has nothing in common with that oneWell maybe, you can't say for sure until you get into the code yourself, are you saying this based on your own analysis of its code? I personally trust the dude who dug in the guts of this "vector machine", it hardly makes sense to mislead me, but if you have parsed Recht's code yourself and clearly saw there a nuclear SVM, then you need to understand who is wrong...
PS "nuclear SVM" cannot be made with polyfics, non-linearity can be added, yes, but for "nuclearity" or "nuclearity", if you like... you need to count set of linear classifier responses at each point in vicinity, and multiply results by kernel from distance to point.
Moreover, the signs are not enumerated from a pile of some, but transformed by polynomials of several existing onesI would try this method, but I have no idea how to implement it.
I've already written to you twice about the book, but you ignored it :)
Ivakhnenko, "MSUA" (method of group consideration of arguments)
Yes the search engine does not know such a book - I have read the essence, the link to the lectures, which describes a similar approach (there below). I read it, but the question is in the implementation of cunning formulas.
are you hilarious?
Maybe the title of the book is different? There is no such literature by his authorship.
As I understand it, you propose to transform the sample and reduce the number of predictors, replacing them with polynomials, well, you can certainly try. Can I throw you the sample, if you have it implemented? And if not, we can think about a joint implementation.Well, at the bottom there is a link to MSUA, and in it a link to the site
(Can you teach me how to use the internet? )
What is the bottom - teach me of course :) You can just give a link, why complicate things ...
http://www.gmdh.net/articles/
3rd book from the top
there is both about the enumeration of features and about sample balancing, model evaluation criteria and a lot of other interesting things
More precisely: "Inductive method of self-organization of models of complex systems"?