Article: Price forecasting with neural networks - page 8

 
Aleku:
So, it seems that I have disturbed the debate on high matter after all.


No, it's just that these questions have always been and always will be. Even those who flee from them, still somehow solve them for themselves, but do not realize it. And this does not make his decisions any better.

And the task you wrote about - building a city using MT4 tools - is really interesting. I would have had a hand in it long ago, if I knew the subject. But alas.

And what was published here under different smart names even a dilettante in this area looks ridiculous to me.

 
Yurixx:
Aleku:
So, it looks like I've interrupted the debate on high matter after all.


No, it's just that these questions have always been and always will be. Even the one who brushes them aside, still somehow solves them for himself, just without realizing it. And this does not make his decisions any better.

- So why don't we start a separate thread for that? And discuss thermodynamics and kinetics there, too,
and chromodynamics and relativity?
 
Yurixx:

And the task you wrote about - building a city using MT4 tools - is really interesting. I would have had a hand in it long ago, if I knew the subject. But alas.

'Using artificial intelligence in MTS' - there seems to be an implementation in maveric's post. But the author himself says it's interesting just as a code in itself, not as an expert.
 
Aleku:
VBAG:
Aleku:

As a consequence, the Achilles' heel of ANN is over-optimisation.

This is true when it comes to time series prediction. But this is only a small part of the huge field of application of ANN.

- But in this particular thread and in the particular article under discussion, that's exactly what it's about.
Or am I wrong?

VBAG wrote (a).

A properly formulated problem is solved perfectly with these methods.

- I'm not going to argue against it - the problem is how to correctly
The problem is how to formulate the task correctly or how to prepare data for the grid.

This is the main problem - first formulate for yourself what you want. Make sure there is a black cat in a dark room before you start looking for it. The main misconception, in my opinion, is to expect results from ANN that it cannot give in principle.
Read the proceedings of an ANN pro conference (I posted it at the beginning of this thread) There you may find approaches and ways how to correctly
to formulate the problem. Good luck to all.
 
Mak:
"If the parts do not possess some property in principle, even in embryo, then the system of those parts will not possess that property. " - this is not true.
Ultimately everything is made up of elementary particles, which are made up of quarks, and at the level of quarks there are no concepts of the macrocosm except the laws of conservation of energy, momentum, momentum ... And even then, at the level of elementary particles, even these fundamental laws are approximately satisfied (due to the uncertainty principle). There are plenty of things that arise only through collective interaction, and are completely absent in single instances. There is even a science called synergetics that studies all this.

This is correct. There is such a thing as "emergent properties" of an object. An object can have a property that its constituent parts do not have. For example, neither Na nor Cl separately has the property of "salinity", but NaCl does. Separate computer parts are just parts, but when connected together in a certain way, they form a new object - a computer that has the property "to calculate and display information". It is the same with aeroplanes: a single engine, fuselage and wings may not fly for long (just as long as it takes to reach the ground), but an aeroplane as a whole has the emergent property of flying in the right direction, of gliding.

The reason for this state of affairs is some Law perceived by ordinary people as a law of nature (like the law of gravitation, conservation of momentum, etc.).

Mak:
So far, no one has been able to determine what is the difference between living and non-living things? How does the intelligent differ from the unintelligent, etc.


Life is a form of existence in the universe. Our ideas about what is life are limited by the level of current consciousness. Under FS (physical consciousness) one identifies life with the period of active action of the physical body. In MS (mental) time is perceived spatial-like, so one's own immortality becomes self-evident. In reality there is nothing but life in the universe :)

Quantitatively, the degree of "reasonableness" for people can be determined by the number of concepts that a person is fluent in. At that, decisions made by a more developed person are not always correctly perceived by less developed people; it may appear to them that the decision is erroneous or even absurd.

Mak:
The name Artificial Intelligence arose more out of emotion than anything else.
To define it, one must first define natural intelligence,
and define the differences - how artificial intelligence differs from natural intelligence.


Natural intelligence is the ability to unconsciously receive and process information by means that are present in man but are not part of the physical body, the brain, for instance. The human brain, with all its neurons and nerve connections, is only an executive mechanism, like that of a printer. If a printer prints, for example, a textbook on programming, he may become proud and think that he is so smart, when in fact the information (which is available in nature independently of his will) has only been given to him to process. The level of development of intellect may be measured by the ability of a person to get the information he/she needs (from a source, which is unknown to a person in FS). In SF information is a property of the environment, so any intellectual truth is perceived by man directly and directly. For the same reason truth cannot be arrived at intellectually.

Artificial intelligence, at this level of science and technology development, is simply a term for all sorts of computing devices that process information. Living beings can create their own laws within the limits of their competence, and importantly, have the full right to dispose of their creations. For instance, a human being can create a computer and is the full and undivided owner of it. A human being perceives a computer adequately and fully, while a computer does not perceive a human being in any way. A man can only put into a computer the conscious part of information as well as conscious methods of processing, while the unconscious part cannot, because of its unconsciousness. Since natural intelligence is a capability based on unconscious properties of reality, artificial intelligence is out of the question at this stage of human development. It is simply a term.

(in a later stage mankind will switch to MS en masse; as a result new directions in science will appear, and finally man will be able to comprehend and transfer to machine qualitatively other abilities based on properties of astral and mental worlds; but it will not happen soon yet)

 
"In fact, there is nothing but life in the universe :)" - and yet, I have yet to encounter a definition anywhere.
Often a clearly living thing behaves like a non-living thing, and vice versa...
Whichever definition I came across, I always found an example that contradicted it.

"Quantitatively, the degree of "intelligence" for humans can be quantified by the number of concepts a person is fluent in" - the definition is anthropomorphic :), not applicable to non-humans :)

"Natural intelligence is the ability to unconsciously receive and process information by means available to man but not an accessory of the physical body, the brain, for example" - again an anthropomorphic definition. But it is only human beings that are intelligent. If you take out the word "human", then any computer or calculator fits the definition - "...the ability to unconsciously receive and process information. ."

In general Artificial Intelligence (AI, AI) is just a once fashionable topic.
They are solely concerned with the IMITATION of individual human functions.
Intelligence is out of the question until there is an answer to the questions - what is life, mind, intellect, consciousness.
Until then, it will be more or less good "prosthetic devices" - tools without their own brains (the user must supply the brains :).

Neural networks are a fashionable topic in AI, including now.
Constructed, by the way, also on IMITATION .
But it is only a tool, it does not contain brains, one must apply one's own.


Getting back to the topic of the thread.
NS is just a "shovel" (tool) which allows (sometimes) faster digging ... if of course you know how to dig.
If you can't, it won't help you.

The NS is a very flexible "shovel", so it usually gets bent in such a way that you don't understand,
whether you're digging a hole or making a mound, and it takes a lot of brainwork
to work out what you've actually got?
Is it what you wanted, or something else entirely?
(more often the latter).

In general, the moral is that NS is certainly an interesting topic,
But for ordinary people (like us) is absolutely meaningless.

All of course, IMHO, and at will ...
But that said, people will still be looking for grails in the NS ... :)
 
Mak:
But no matter what, people will still look for grails in the NS ... :)
Yes, Mak, they will anyway, but they will find only a fraction of percent. Majority will give up almost at once, not getting what they want with dumb application of NS (by feeding raw data to inputs and doing interpolation). Majority of others come back to NS after some time and fail again. And only small part of them start using them properly, apply them adequately and get good results. The adequate area of application of NS (regardless of what they simulate) is pattern recognition and classification.
 
Mathemat:
Mak:
But no matter what, people will still look for grails in the NS ... :)
Yes, Mak, they will anyway, but they will find only a fraction of a percent. The vast majority give up almost immediately, not getting what they want from NS when using them stupidly (by feeding raw data to inputs and doing interpolation). And only small part of them start to use them properly, apply them adequately and get acceptable results. The adequate area of application of NS (regardless of what they simulate) is pattern recognition and classification.
Dear Mathemat and Mak, I join you and completely share your point of view!
Alexey. I still can't get out of my head your elaboration on tick analysis. I am interested in this subject. But I only think about it, so I have nothing to share. I hope you haven't abandoned it. I think a simple 2-3 layer grid may be useful for improving results of analysis of tick information rate. It's similar to the targeting correction in our missiles back in the 70s, which used a simple perceptron, and was hardwired into the hardware. But this is speculation so far.
Regards,
Vladimir
 
It's far from straightforward with ticks, VBAG. See further 'Tics: amplitude and delay distributions'.
 
The key idea proposed by Reshetov - to limit it to two classes: "Price is likely to go
- Price will most probably go up" and "Price ...- Down" is a very good idea but its implementation using the iAC indicator is a mess.
I propose a very simple Expert Advisor, which uses candlestick patterns / instead of iAC /, at the same time
I'm pasting the test results of this EA on EURUSD and H4 from 01.07 to 15.11. 2007. ,

where profitability was = 9, expectation = +160 pips, profit = +1480 pips, total = 9 trades.
These results were obtained out of sample , with fixed lot.
So, the Expert Advisor itself:
1. it uses candlestick patterns, i.e. instead of iAC we write:
double a1 = High[1]-Low[2]; double a2 = High[3]-Low[2];
double a3 = Low[1]-High[2]; double a4 = Low[3]-High[2];

2. let's trade EURUSD on H4 / all the rest is bpc ! / ,
3. This MTS is a reverse one, i.e. if buy is closing, then we immediately buy shorts, and vice versa...
4. weights x1,x2,x3,x4 from 1 to 200, in steps of 1, and losses bs,ss from 50 to 90, in steps of 5 - optimized
weights,bs,ss,for the previous 6 months are optimized /fitted/, then they are inserted into the next month, as soon as this month
will pass, the optimization window is shifted one month forward, etc... SHOW MUST GO ON! ..
5. additional condition: after 5 bars / i.e. after 20 hours / moose in an open position is either
transfer to break-even / if possible/, or the position is closed with a small loss
/the trend has not come true/. By the way, this condition is not in my EA, but I manually checked
the above period, the results were even better, the profitability is = 30,
expectation = + 190 pips, profit = +1600 pips.
6. we put the obtained weights x1,x2,x3,x4 and losses bs,ss for each month into the switch() variant operator ,
DiRoLnoDoLgo.mq4 file - recompile, set ALL external variable to true
/to see results in few months/.
I wonder how the "neuron" in this EA is trending, or else,
... or maybe there are some mistakes in this EA ? :)
Files: