NS + indicators. Experiment. - page 2

New comment
 
SK. писал (а):
I second that. It's not an article, it's just a walkabout.
I was wondering why I didn't want to read it. That's why... :-)
 
Alex-Bugalter писал (а):
Dear Rosh & SK, if you know so well what's good and what's bad, and where it's better to walk.
Maybe you can point out to the uninitiated what you think the harm is and what exactly is not stated correctly in this article?
So many people have been misled, so go ahead and point the right way.
Or are you just out for a walk?
Anyone can indiscriminately cast aspersions.
And in this article: "Neural networks and time series analysis", is it rubbish written also?

P.s.: And Rosh, for me personally, if it's not too much trouble, what exactly did you mean by: "Written didactically awful"?



The article clearly shows not a desire to convey information, but to show its awesomeness. Therefore, the clarity of presentation is in the third place (all kinds of phrases like Gazprom below 5.6 [because only a dumb person would ask where it is visible] - I will not now return to the article to indicate exactly, fuzzy pictures and so on). Further, the authors marvel at how accurately the network predicts the boundaries of maximum and minimum values of future bar prices and say that only a complete fool would not make money on it - it is a complete profanity. So let them make money on it, if it is so simple. A lot of clever words, you can write something like that after reading a couple of books. I repeat - the purpose of this article is to show how clever the authors are and how well they know the spells when working with these programs and to put the reader down as much as possible who will not understand anything from it and will understand how cool everything is there.
 
This seems to be an article from The Currency Speculator. I agree with at least one point there: Close is predicted worst. Only the explanation accepted by the authors I don't like, I don't like it at all. The reason, most likely, is that Close is much more informative compared to other prices - and, as a consequence, less predictable.
 
Mathemat:
This seems to be an article from The Currency Speculator. I agree with at least one point there: Close is predicted worst. Only the explanation accepted by the authors I don't like, I don't like it at all. The reason is likely to be that Close is much more informative compared to other prices - and, as a consequence, less predictable.
That Close is the worst predictor is known without this article.
 
Rosh: That Close is the worst predictor is known without this article.

Just this article is old, I read it 2 or even 3 years ago and it was dated 1999 or 2000, although it was dated 17.12.2007 - which is very strange. The only correct idea in this article is to predict not Close (and not even High or Low) - but the direction of instrument movement, which is much easier to do and more effective in work, which, by the way, was proved by Batter......
 

This is a great article from the point of view of the merketologists at the firm who want to sell the software. It's advertising in a beautiful package, which gives the false impression that everything is so simple and easy. Just buy it and put it on the chart and you're in luck. And the magnificent phrase "two nets with one outlet are better than one" is brilliant. Gives an absolutely clear idea of who wrote it. IHMO, bullshit, absolute ignorance of the NS purpose. NS is a tool, like a surgeon's scalpel, and here it is absolutely clear that the one who wrote this does not understand the idea that the developers have put into this program, just pokes this scalpel (tool) in different places, maybe that will work.

 
njel:
Prival:
njel:
gave
for input 
         double c1 = MathLog (  iHigh(Symbol(),Period(),i) / iClose(Symbol(),Period(),i+1)) * 100.0;
         double c2 = MathLog (  iLow(Symbol(),Period(),i) / iClose(Symbol(),Period(),i+1)) * 100.0;
and the net was timelagged with a trajectory of 15 . but the quality of the forecast was not happy.

The natural logarithm of the ratio of two numbers ? Is this the percentage of price change ?
I'm not. I used to take the % of price change as a proxy for 100%. But recently I've seen such measurements of price movements. And at the moment I'm thinking about how to feed the next NS model. and in general, whether it is worth it. hmmm.

I would encourage you to think about what you are doing. If you want to feed NS simply by searching through indicators, it's almost the same as if you just tried all known combinations of indicators and see if you succeed. Everything is determined by the architecture inside. And there is only one correct input, and that is the quote flow. Everything else is a transformation from this stream.
 
Prival:
I would recommend you to think about what you are doing. After all, feeding NS simply by trying different indicators is almost the same as trying all known combinations of indicators, just to see what happens. Everything is determined by the architecture inside. And there is only one correct input, and that is the quote flow. All the rest is a transformation of this flux.

And you would first try to feed the NS with the quote stream ... ... and then write about what feeds the network. You always need to convert the quote stream, otherwise you get gibberish.
 
Prival:

And the magnificent phrase "two networks with one output are better than one" is brilliant. Gives an absolutely clear indication of who
wrote it. IHMO bullshit, a complete lack of understanding of the purpose of NS.



If you had tried this statement, namely the use of multiple neural networks instead of trying to cram everything into one, this post would not have happened. Took the phrase out of context and interpreted it in your own way.
Not a successful backing vocal.


After all, feeding the NS just by trying different indicators is almost the same as trying all known combinations of indicators, just to see what happens. Everything is determined by the architecture inside. The only correct input is a quote flow, everything else is a transformation of this flow.
That doesn't make any sense. I'd rather agree with your second statement than the first.
 
Alex-Bugalter писал (а):
Rosh & SK, if you know so well what benefits and what harms, and where better to walk.
Maybe you can point out to the uninitiated what you think the harm is and what exactly is not stated correctly in this article?
So many people have been misled, so go ahead and point the right way.
Or are you just out for a walk?
Anyone can indiscriminately cast aspersions.
And in this article: "Neural networks and time series analysis", is it rubbish written also?

P.s.: And Rosh, for me personally, if it's not too much trouble, what exactly did you mean by: "Written didactically awful"?

There's no need to be so forward-looking. There really aren't that many useful articles and books. The normal path of the average specialist in a field of knowledge usually runs from a textbook (where basic definitions are given and numerous examples are discussed to reinforce new concepts to the level of free use), through studying and publishing articles (usually engaging in a debate [in a good way] on a single applied issue) and on into a creative activity where the only source of information and inspiration is one's own intellect (working in an unexplored field).

It is easy to see that articles, by definition, are a source of information, which point out new aspects on an issue with which the reader is already familiar at the level of a textbook ("like articles" in low-brow multi-coloured, entertaining magazines do not count). For a first introduction to the concept of "neural networks" this article may be all right (although a correctly constructed textbook explains this concept better). So in my opinion Prival is right in saying: "It's advertising in a pretty package that gives the false sense that everything there is so simple and easy".

The best way, in my opinion, is to walk from a specialized library to a desk and back.

PS. I just remembered that in the book The Altist Danilov, Vladimir Orlov uses a wonderful term, which the author uses in a pejorative sense - "young demons with a television education". Danilov was written 30 years ago, at that time there were no computers. But in essence Orlov is right, it's just that instead of televisions now there are computers. .

123456789...12
New comment