Discussion of article "Using Self-Organizing Feature Maps (Kohonen Maps) in MetaTrader 5" - page 5
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
Hello, dear forum members!
Very interesting article! I am trying to use this SOM code too.
Can anyone suggest how to make it easier to calculate the result of the area around the BMU (circled in blue for clarity), taking into account the distance from the BMU ?
Did you mean to say that the code executes much faster in Java? Please attach your sources for comparison, it's interesting to see.
Has anyone found a solution? Just found this article now. I'll join you in thanking me. And to the question.
My comment 5 years after my last comment.....
Why am I not seeing the story about the training set and the control set? The article cites historical analysis using some sort of colour system, but in the phrase "machine learning" the key word is "training" and it is performed to trade on future periods.
In short, why this sophisticated historical analysis? You run optimisation and see which period and which shift are tested better.
Why don't I see the story about the training set and the control set? The article gives a historical analysis using some kind of colour system, but in the phrase "machine learning" the key word is "training" and it is performed to trade on future periods.
I asked a similar question in the discussion of this article https://www.mql5.com/en/articles/5473.
I have studied the material on this issue, it is most likely that Kohonen maps simply perform the task of displaying multidimensional data, and they are not intended for data analysis itself
I asked a similar question in the discussion of this article https://www.mql5.com/en/articles/5473
I've studied the material on this issue, most likely, Kohonen maps just perform the task of displaying multidimensional data, and they are not intended for data analysis itself
Read it, it seems to me you two (three of you?) didn't share a tasty candy, and on it went.
I think any publication has the right to be, but there is no description here of what a neural network does in the usual sense - makes a "new" decision in a "new" situation. There is only historical analysis. It's perplexing.
I'm watching Simon Haykin, there are good examples there. And at the end there are tasks, among the tasks there is forecasting, or what I would call forecasting. I'll write a couple of more sentences on the subject if I get the hang of it.
Read it, I think the two of you (three of you?) didn't share a tasty candy, and on it went.
It was not so, I once again decided to return to NS, and the choice fell on Kohonen maps, having googled the material (this site is very well indexed by search engines), I got acquainted with all the proposed materials of the search engine.
I was interested in these types of NS about 5 years ago, without theoretical training, now the amount of knowledge on NS is quite decent, and re-studying the material on Kohonen maps caused a lot of questions.
I asked a specific question... and then instead of searching for the truth I came across the defence of "author's interests", which for some reason repeat the article from Wiki and are not supported by anything else, except...well, as you said, "it went on and on" - "you are a fool" - "go read"
I watch Simon Haykin, there are good examples. And at the end there are tasks, among the tasks there is forecasting, or what I would call forecasting. If I figure it out, I will write a couple more sentences on the essence.
I have already read it, as the first book on NS it is the best, then how much literature I read - more than half of new literature will be reprints from Haykin.
And at the end there are tasks, among the tasks there is forecasting, well, or what I would call forecasting. If I figure it out, I will write a couple more sentences on the essence.
I would be glad to discuss it, I have been looking for information on this topic for a long time - Kohonen maps are not designed for anything - nothing at all! - They are just beautiful displays of multidimensional data.
the idea of this kind of NS is quite tempting, the principle is like an electronic component decoder - we input a combination of input data and get a ready result at the output.
tested Hamming networks, well, as if what I'm looking for, but ... so far I've abandoned NS - I've started working on simple solutions, here are some ready-made ones https://www.mql5.com/ru/forum/307970/page11#comment_12625353.
To quote S. Osovsky:
"A self-organising network can also be successfully used for forecasting, for example, loads in an electric power system. This subsection will present the details of solving the problem of forecasting hourly loads in an electric power system on a 24-hour interval".
So it's all good. Such forecasting as described next is generally suitable one-to-one for predicting buy, sell or reject action in forex.
I always look at the root, I knew no one would call it a neural network if Kohonen cards couldn't predict.
I always look at the root, I knew no one would call it a neural network if Kohonen's maps couldn't predict.
they can't, training is there to deploy vectors of NS weights over training sets - the result is to cluster the data, but the response of the network itself is absent to other data - or rather it will be, but it will produce random values.
about the root... the name is not Kohonen's network, it's like Self Organising Maps (SOM).
UPD: I don't see the point in continuing the discussion, the second time the discussion is reduced to what is written in Wiki, and now to what a certain "Quoting S. Osovsky" wrote. I agree to remain in the captivity of my reasoning, which is not supported by the phrase "SOM Kohonen" can predict, and the reverse - they can not
they don't know how to do it, there is training to deploy vectors of NS weights over training sets - the result is to cluster the data, but the response of the network itself is absent on other data - or rather it will be, but it will produce random values.
about the root... the name is not Kohonen network, but Self Organising Maps (SOM).
UPD: I don't see the point in continuing the discussion, the second time the discussion is reduced to what is written in Wiki, and now to what is written by someone "Quoting S. Osovsky". I agree to remain in the captivity of my reasoning, which is not supported by the phrase "SOM Kohonen" can predict, and the reverse - they can not