Machine learning in trading: theory, models, practice and algo-trading - page 178

 
Mihail Marchukajtes:

Well, I even began to wonder.

I turn to R-torturers, directly toAlexey Burnakov.


Thank you for the idea. I will keep it in mind, but I will postpone it for later. I have no time. I can't finish my own project, plus I have a full-time job.
 
Zhenya:

Potentially interested, depending on what exactly we're talking about. Email me privately or directly here, whichever is more convenient.

Sorry Zhenya, but as far as I know you do not know R, and I do not want you to write code for me and in a language I do not know, I just wanted to write code with someone experienced in ML in the language of R, that he would help me in something and guide me and in passing to protect me from any unpleasant and not obvious mistakes...

Those who I was counting on did not respond, so this proposal is not interesting to them, or their own business, or, or... This is normal, everyone has his own way ...

So I'll have to write myself unfortunately...

Mihail Marchukajtes:
I think that ifmytarmailS :told about his MEGA idea about identifying patterns, it would be interesting, but so.... I know something, but I won't tell you... It is not interesting to communicate like that, I've disclosed a couple of my secrets, who wanted to hear them, heard them, the rest ignored and in vain .....

That's the point is that I'm not interested in communicating, it's interesting to implement and see what happens...

And to explain to the full essence, all the advantages before the other algo, I need quite a lot of text and draw pictures, really a lot, and it needs to strain, and judging by the reaction and the interest, then after I write and draw that all the same thing beyond just one conversation will not go, and I already have these empty talk gag reflex developed

It's just that with this algorithm it will be possible to extract the clear patterns from the data, instead of adding hundreds of predictors to ML, it's the same greedy noise cleaning, which has been discussed 1000 times but no one got anything useful

But this algorithm may not work either, since there may be no statistical regularities, but I'm sure that if this algorithm doesn't work, no ML will work ...

 

The fourth and final part of the article about RNeat is out...

I dare say at least one person here would be interested in reading it

http://gekkoquant.com/

 
Gianni:

If it is not inconvenient then me too in a private whisper about what you Combinator asked.


Thanks.

Pardon, Zhenya, but I do not know anything about you, and respected Combinator well-known figure here and more than once proved his ingenuity, so I answered him without further questioning, although he probably knew it all himself. But if you tell me about your experience in algorithotrading and it appears that you are a virtuoso in juggling with more than 10 correct chips than the local "grailers", and you even have worked withHFT, at least at the local FORTS, I'm sure we will become friends))) In the meantime, I suggest you take the Turing test and atthe Winton contest on cagle and get into the first at least twenty, the contest passed, but to see what place YOU could be in, I recommend it.

 
mytarmailS:

not to shove hundreds of predictors into ML, that's the greedy noise cleanup that has already been discussed 1000 times here but nobody got anything right.

You're wrong - even the code was posted by Dr.Trade
 

interesting publication

"Experience in building trading robots".

http://www.kamynin.ru/archives/category/torgovye-sistemy/page/122

"Spectral Analysis in the Stock Market"

http://www.kamynin.ru/archives/1560

 

You're absolutely right, bros!!!!! It's not what model or optimization method we use, it's what data we use for our system that's important. There is a philosophy of constructing exactly the output function, with which you can increase the generalizability, that is, to fit the output to the inputs. I really thought of another way to bring the data, but I will not tell you about it yet.

And by the way, maybe let's write all the articles????? And publish them at the same time.

Details are not necessary, the main thing is to convey the meaning of the method.

 
mytarmailS:

The fourth and final part of the article about RNeat is out...

Thanks, I'm curious too. I had RNeat retrained and I didn't get anything working with the package, I'll try the result from the article again.


SanSanych Fomenko:
You're wrong - even the code was posted by Dr.Trade

Are you talking about PCA component analysis, or something else? I don't remember all examples, I've posted here :)

If you're talking about PCA, you won't make a sweet out of garbage anyway. It's necessary to have quite good predictors mixed with bad ones, then PCA can sift out the bad from the good.

 
Dr.Trader:

Thanks, I'm curious too. I have RNeat retrained and I have nothing with the package and did not work, I will try the result from the article to repeat.

I posted it for you in the first place :)

Mihail Marchukajtes:

You're absolutely right, bros!!!!!.......................

what are you talking about? :) what are you right about? what article? what do we write at the same time? i don't understand it at all :)

 
mytarmailS:

I posted it for you in the first place :)

what are you all talking about? :) what's right? what article? what do we write at the same time? i don't understand it at all :)

It happens......