Using artificial intelligence at MTS - page 9

New comment
 
Reshetov писал (а):
But there is no solution to the inverse problem, i.e. by naming an object to a neural network, we can find out its coordinates. All that can be learned about the object is on which side of the dividing plane it is located. Therefore, interpolation and extrapolation are out of the question.

Yury, I want to repeat: all your clever arguments about linear separability, flies and cutlets are applicable only to perceptron network with linear activation function. Your publication is old and dates back to the time when neural networks were in oblivion thanks to Minsky's nasty theorem about this primitive neural network. There are no such neural nets now for a long time.

As for the possibility of interpolation, try to interpolate a sinusoidal function (or anything smooth, even MA with a large window) - and see if they are good for it or not. Another thing is that the accuracy of MA interpolation is of no use in predicting the price itself...
 
Here's a crazy idea. We make a 3 layer network, or rather 5 arrays (3 one-dimensional for neurons and 2 two-dimensional for links). Input - Low and High for the last (roughly) 30 bars plus presence/lips of an open position. Exit - Open_Buy, Open_Sell, Close_Buy and Close_Sell. All this is thrown to optimisation.

Of course, it will be quite long, but still....
 
Itso писал (а):
Here's a crazy idea.

I think Rosh did a neural network in MQL4. You should consult him, it's going to take a long time to do it yourself, it's not a big deal.
 
Mathemat:
Itso wrote (a):
Here's a wacky idea.

I think Rosh made a neural net in MQL4. You should consult him, it'll take you a long time to do it yourself, it's a big deal.
I agree with Reshetov to a certain extent - first you learn to identify the setups yourself, and then you teach the network. It makes no sense to make a neural network on the principle that something will turn out. I think so.
 

I have already explained why I don't believe in neural networks. I have dabbled with them enough - mainly Hoffman nets (which are better than perceptron). The idea here is to train the mesh - direct by optimizer. Otherwise the neural network is really a couple of arrays.

 
Itso:

I have already explained why I don't believe in neural networks. I've dabbled with them enough - mainly Hoffman nets (which are better than the perceptron).

Are you sure that Huffman nets are better? I mean the formulation of the problem, when the values of technical oscillators are fed to the inputs?

Most of the modern so-called neural networks solve approximation problems in their main formulation, and identification goes into the background. And the result is the same, i.e. such pseudo-neurons can only be dabbled with.
 
I'm sure no neural network can help in the case of Forex. In fact, neural networks recognize fuzzy patterns. And in this (recognition) the best network is the brain (of course there is a caveat - if there are only a few recognizable states - but in Forex there are 4 (Open Long, Open Short, Close Long, Close Short). But we are not very good at winning either.

And the thing is that there are no patterns in Forex.

There is the so called "classic market model" (Yelder, "Fundamentals of stock trading"). Of course there are a lot of models, but they, IMHO, either repeat the classical model, or are wrong. In brief - prices move in the support and resistance corridor. Sometimes they break through, and new support/resistance lines are established. There are also long-term trends, they move within the corridor and breakthroughs, but usually in one direction.

Let's think of a physical market model. Importers/exporters are always changing money and moving the exchange rate a bit. And for us - randomly. When the market is flat, the crowd of traders (or rather most of them) think: "Well, the Euro seems to have risen too much - we should sell" and vice versa. Someone will say here that it is not so. That the modern trader uses indicator(s). Yes - but all the indicators based on the muwings actually give signals about reaching these very levels.

BUT - there was a breakout. Let's say upwards. Then the same crowd of traders say: "That's it! The big movement has begun. It is necessary to grab while it is still going". And they buy. And from this the rate goes up even more. And so it goes until they start to sell. At this point, the fate of the trend is decided. If too many traders close longs, there will be no trend. And vice versa.

And here is (IMHO) explanation of the phenomenon "small ranges are followed by large ranges" (William Larry, "Long-Term Secrets of Short-Term Trading"). The market does not move for some reason. Traders get more and more nervous. At the end they are already risky. Some movement starts and everyone grasps at it....

So IMHO the main task of a trader is to correctly identify support, resistance and breakouts.

And we don't need a neuron for that :(
 

The best solution tends to be at a junction; it is more constructive not to deny, but to take the best and add to it.

 
Itso:


So IMHO the main task of a trader is to correctly identify support, resistance and breakdowns.
The trader's main task is to increase equity and systematically release this equity into the balance. Anything that does not contribute to the main task is bullshit and empty rhetoric.
 
Reshetov писал (а):
Mathemat wrote (a):
Reshetov wrote (a):
Minsky, M and Papert, S (1969) The PERCEPTRON; an Introduction to Computational Geometry, MIT Press, Massachusets

translation available:

Minsky M., Papert S. The Perseptron: Translated from English: Mir, 1971. - с. 261

My advice, children, before fooling around, and before making public much-important conclusions based on the results of fooling around, try first to study the materials on the subject. Firstly, it won't make any harm, and secondly, it will allow avoiding stepping on rake, about which everybody knows for a long time already.
Thanks for pointing out the source. Well, as for the matter, I actually got acquainted - by publications on the subject of neuroprediction.

Articles articles, but the geometrical sense can't be taken away. And it's that a linear filter allows to separate flies from cutlets, if coordinates (values of features) of these very objects are known using linear plane under condition of linear separability. But there is no way to solve the inverse problem, i.e. having namedan object to find out its coordinates. Everything there is to know about an object But we can only find out on which side of the separating plane it is located. Therefore, interpolation and extrapolation are out of the question.




FACT - The brain is never capable of understanding circumstances that are not bound by space and time. This is what makes it impossible for a person to be objective. This is the algorithm of the brain. I explain - Space and time, create only flat panoramas of the seen. I.e. panoramas LIMITED by their outline. Yes, - these are our OBJECTS, no matter how we support them with formulas, they will always correspond to the algorithm of our thinking as a kind of HOMO! And formulas, as a product of the same brain, cannot have a possibility - to MEET THE REQUIREMENTS, OF THE WHOLE UNIVERSITY. It is this circumstance that makes man - not only LIMITED, but also DEFINITELY in the essence of its limitations. I.E. Errors in Forex are guaranteed exactly until the trader - will not trust the EXPERT, with a predetermined PRINCIPLE in it. This is the WAY to avoid such errors with EXPERT - and gives us a different approach, as I said - EXPLOITATION of the objective. Such exploitation does not know how to make errors. And that, in turn, will be in line with the SUTIe of success, the EXPERT. Only if we abide by certain principles of the world. THOSE PRINCIPLES that do not even fit in the minds of the GENIENS. EINSTEIN - would agree with the following statement: Opinions, feelings, results of research, the Highest Principles of the UNIVERSE (descriptions of the Laws of the Universe) - are not only VARIABLE, but they are also subjective . And it is a FACT - the Descriptions made, even by geniuses, and even with formulas, like it or not, are SUBJECTIVE. Including the formulas of EINSTEIN himself, for they contain the concept of time... FACT - ARCHIMED asked for a POINT OF REFERENCE and then he would have leveraged the WORLD as well!!! Only a POINT has the properties of an OBJECT. It has no time and space. More precisely speaking, a point with potency is EXTREME, i.e. it is a method of escape from the three-dimensional world. I can show by examples the way how to get us useful RESOURCES from it. It's not even TRIZ! FACT - All Manifestations of the visible and invisible, - are LOCATIONS OF FLAT panoramas, so even if they have a scientific description or interpretation of them. That, and in geniuses - such SOUL, = MORE insignificant and not essential, even when applied in the description of the WORLD, the HIGHEST MATHEMATICS, with formulas. Even when it is the Theory of Relativity with time and space - EINSTEIN.

FACT - Time already shows that we are not dealing with the whole, but with a part of the whole. Any part in isolation from the object is a SUBJECT. THEN.

Objectivity and Multidimensionality of the Universe - these words are also synonymous with FOREX. Profit, PURPOSE, Result, is the UNIVERSITY of the Inner UNDIMENSITY of Events, where there is a GENERAL RESOURCE. THE ABILITY to distinguish the OBJECTIVE from the OBJECTIVE for EXPERT, for programming Neuro Shel inputs - is lost because the parties do not understand the OBJECTIVE from the SUBJECT - there are hundreds of examples of how TO TAKE AND EXPLOIT - OBJECTIVE!!!

12345678910111213141516...28
New comment