How to form the input values for the NS correctly. - page 25

 
lna01 писал (а) >>

You just don't quite understand what is being approximated. There is an input vector X of dimension N and an output vector Y of dimension M. NS establishes the relation between them, i.e. it approximates the dependence Y = F(X). Y can be anything, even triple redundant, NS doesn't care, it solves the problem of approximation F(X) on the training set.

It's you who doesn't quite understand that on a training (or as it's also called: "representative") sample it's a barefaced fitting. Normal traders prefer to see on OOS (out of sample), i.e. out of representative sample (forward - tests). No DC will pay for the accuracy of the approximation.

 
Reshetov писал (а) >>

Dear Sir, where did I claim that interpolation is related to the future? Go see an oculist and read the posts carefully instead of throwing around expressions. I have reported and reiterate for the particularly gifted that extrapolation is necessary for the future.

OK, let me explain my point.

For a start, my understanding.

Firstly, it's not interpolation but approximation, but that's beside the point.


What is extrapolation? It is a type of approximation in which a function is approximated outside a given interval.

What is forecasting? It is approximation in its pure form, because a given function approximates the original function over the entire range of values.


Therefore -- extrapolation by polynomials turns into prediction if a given polynomial approximates the function over the entire range of definition.


Based on this, extrapolation with splines and others, as well as redrawing indicators, is pure extrapolation.

And neural network extrapolation is prediction, because a properly designed network sets the original function over the entire range of values.


A neural network is not used to extrapolate data but to make predictions. You have compared a neural network to redrawing indicators which is wrong, at least.

Conclusion - the approximating abilities of neural networks are applicable to the market even if it is non-stationary.

 
TheXpert писал (а) >>

Therefore -- extrapolation by polynomials turns into prediction if a given polynomial approximates a function over the entire domain of definition.


Nonsense. It is obvious that you are a theoretical nerd who is very stubborn in his misconceptions, i.e. a lamer. Ask anybody, who more or less dealt with autotrading and he will confirm that not every adjustment passes a forward test, which in pro pro-natural language means that not every approximation will give an adequate result on extrapolation. The reason is the non-stationarity of financial instruments.


Better get into the practice of trading or auto-trading than to sign here in your utter ignorance.

 
Reshetov писал (а) >>

Mr. Nerd, normal people have their own brains and their own experience, while nerds quote other nerds because they do not and cannot have their own brains.


Heikin most likely trained the network in a stationary environment, hence his conclusions. In a non-stationary environment the network may fail to learn at all if too many patterns are given because, for example in trading, today a pattern points to buy, and the next time it points to sell. Because any entry has some probability of false signals.


Well, that's a rude way of putting it.

I have my own experience and my own brains, which are enough for me to have a good job and a well-deserved reputation in certain circles.

Including a reputation as a skilled specialist in neural networks.


So next time, before you write nonsense, use your brains, if you have any, and don't start a meaningless rant, especially with insults to others.

 
TheXpert писал (а) >>

Well, that's a rough one.

What is and isn't a hit and run is up to the moderators here. And you shouldn't assume their authority.

 
Reshetov писал (а) >>

Stupidity. It's obvious that you are a theoretical nerd, and you are very stubborn in your delusions, i.e. a lamer. Ask anybody, who more or less dealt with autotrading, and he will confirm that not every adjustment passes the forward tests, which in pro-natural language means that not every approximation will give an adequate result in extrapolation. The reason is the non-stationarity of financial instruments.


Better get into the practice of trading or autotrading than to sign here in your utter ignorance.

It all makes sense, LOL.

The reason is crooked hands. Non-stationarity is not difficult to get around, and with a properly chosen network and input data, it won't be necessary either.

Who said anything about fitting? A network that does not pass forward testing is not "properly done" and therefore does not make predictions, but extrapolates or generally pokes dumbly at the sky, which is adjusted on the tester.

 

Guys, stop trying to see who has the longest. Or at least not here.

Just be nice to each other.

Thank you. (chuckles)

 
TheXpert писал (а) >>

It all makes sense, LOL.

The reason is the crooked hands. Non-stationarity is not hard to get around, and with a well chosen network and input data, it won't be necessary either.

And who is talking about fitting? A network that doesn't pass forward testing is not "properly done", which means it's not predicting but extrapolating or just poking stupidly at the sky, which is adjusted on the tester.

Get some rest. Find yourself some other interlocutors.


The subject of fitting to history has been discussed many times on this (and not only this) forum. I see no point in arguing with a lamer about something that has been obvious for a long time.



TheXpert wrote (a) >>

Conclusion -- the approximating capabilities of neural networks are applicable to the market even when it is non-stationary.

If you're so smart, why isn't your face on the cover of Forbes?

 
Reshetov писал (а) >>

Take a rest. Find yourself some other interlocutors.


The subject of fitting in history has been discussed on this (and not just this) forum many times. I see no point in arguing with a lamer about something that has long been obvious.



If you're so smart, why isn't your face on the cover of Forbes?

It's not all that fast. Is yours there?

 
sergeev писал (а) >>

Guys, stop trying to see who has the longest. Or at least not here.

Be mutually respectful.

>> thank you.

Where else is there to banter but in a forum?


Politeness is adequate in the Institute for Noble Maidens