Machine learning in trading: theory, models, practice and algo-trading - page 457
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
And let's define right away, I'm not talking about Reshetov's RNN, but about his optimizer... Not to be confused, these are completely different products.... that is... if anything....
Reshetov's profile has so much stuff that it is impossible to find anything).
By the way, on Habra recently saw an article where random forests are simulated. Pretty simple realization with examples and instances. If you want, you can implement it in MQL, as well as in neuron. In contrast to neuron, nonlinear solutions appear in the forest.
And let's define right away, I'm not talking about Reshetov's RNN, but about his optimizer... Not to be confused, these are completely different products.... so... if anything....
+1
There is Reshetov's RNN, but despite the name it has nothing in common with neurons, just one formula with some complicated rules from statistics. Saw only one and a half examples with this thing.
And there is Reshetov's jPrediction, which is essentially a full-fledged neuron with some special method of model training. This is a separate program written in Java and ready model can be exported by generated code into mql. This is what Michail uses.
+1
There is Reshetov's RNN, but in spite of the name it has nothing in common with neurons, just one formula with some complicated rules from statistics. I've only seen one and a half examples with this thing.
And there is Reshetov's jPrediction, which is essentially a full-fledged neuron with some special method of training the model. This is a separate program written in Java and ready model can be exported by generated code into mql. This is what Mikhail uses.
All right.... What is the difference between AI and NS????
NS is just an artificial neural network, which is part of any AI, and AI is a complex of blocks and functions serving the NS. Sample preparation, preliminary reduction of inputs, control of overtraining, rationer, tester, etc. This is actually the difference. AI is a set of functions and procedures for preparing and training and testing a trained neuron.... Something like this...
hilarious)))
Misha, you'll be nagging the shit out of me, not getting into the bottom of it. Understand that a professor who does everything by the book can
can fuck up a fifth grader's parrots pretty good... You know that neither neural networks nor scaffolding
They can't and a lot of them were just created out of the blue, tried it - oh, it works.)
What we're talking about you know, what others are talking about - I do not give a shit and I know that he went on to rivet them and sped up
you wanted to get it for 100)))
Exactly when it was about the third version of his product, I then suggested him to parallele the calculations and after that he did paralleling and accelerated the program calculations by times, if not dozens of times. I judge it by the size of the trained file, if before 6-7 columns were trained 3-5 hours, it now takes 15 minutes, not to mention the paralleling of calculations. And after that he managed to release as many as 14 releases, finishing in each error and implementing new ideas. So you don't know what product we're talking about. Yes, maybe the early versions were crude and did not meet your requirements, but now the product is quite serviceable. IMHO. In any case, it has only one drawback, with more than a hundred columns and more than a hundred rows, it takes much longer to train....
I still hope to get good computing power in order to build a long enough and adequate model of the market.... there and we'll see....
This is a fun place to be. There's also a team of trayDoors and programmers in a neighboring thread, and it's a hoot.
12 and 14 last youzalu, promised to look at 20, but he disappeared. Misha, the problem is that he didn't take into account the recommendations and
the load of the first versions dragged ...)) Don't worry, it's ok, and 18 Yusufov had many fans, although I told him at the beginning
I told him at the beginning of his branch to get rid of it))) Later, maybe some other god will turn up and you start believing in him))).
But matstat forums where they posted their rattles (and Reshetov and Yusf) - were ignored)))
Well in the end what is the load of previous versions???? What are the mistakes, can you tell in two words????
Repetition is the mother of learning, the more so that the old branches are blacked out, it is not a sin to update the information. Otherwise your invective becomes unfounded.....
+1
There is Reshetov's RNN, but despite the name it has nothing in common with neurons, just one formula with some complicated rules from statistics. I've only seen one and a half examples with this thing.
And there is Reshetov's jPrediction, which is essentially a full-fledged neuron with some special method of training the model. This is a separate program written in Java and ready model can be exported by generated code into mql. This is what Mikhail uses.
Jpredictor outputs exactly the same weights as RNN. The main difference is how to train: in optimizer or by embedded logic via kernel machine and committee from regular mlp + svm
And if you consider that jpred. uses polynomial transforms, it's the toughest fitting, imho... just fitting tulle to tulle. And, as you know, you can't predict markets with polynomials, but it's easy to draw beautifully in the past.
As a matter of fact if to train RF to give weights for RNN it will be the same but much faster, almost instantly and without any paralleling. And instead of nuclear machine as a part of SVM you can use digital filter generator.
https://sites.google.com/site/libvmr/
https://en.wikipedia.org/wiki/Kernel_method
Sensei, I had enough laughs yesterday. One doesn't know what he's talking about and cuts in.)
The other fucking saw it six times, but asked for the seventh.) What are you going to do with this information?
You know it won't do you any good, but people with a normal level will compare it in 5 minutes.
a rattlesnake with a little bit adequate models and throw it away. Thank you for telling me)))
How to get it under the plinth was announced yesterday, but of course you do not understand...
Let's do it again. Especially for me. I'm trying for you..... Make me laugh, etc. Come on... tell me how to get Predictor under the plinth, because I missed it.....
And, as we know, by polynomials it is impossible to predict markets
I do not agree with it, all depends on how this polynomial is received, in consequence of what kind of optimization and if reception of a polynomial means absence of overtraining or undertraining effect, such polynomial can be quite suitable for some time.... IMHO.
And here I do not agree with it, all depends on how this polynomial is received, in consequence of what kind of optimization and if reception of a polynomial means absence of overtraining or undertraining effect, such polynomial can be quite suitable for some time.... IMHO.