Using artificial intelligence at MTS - page 19

New comment
 
maveric:

For classification tasks, it is our interpretation of the data that is taken into account. Roughly speaking, in a letter recognition task, in a set of examples of the letter A, the letters X and U and others should not appear :)

I want to do both. Maybe two grids, maybe one as it turns out. The first step the grid classifies the current situation. If it gives a sufficiently clear signal of the beginning of a trend, then the second step is an attempt to look into the future in order to estimate how much money may be made on this trend.

I assume that approximation of financial series is worse than classification.

Any classification, I am not talking about clustering, requires images. Say to classify images of faces based on a set of features such as distances between some specific parts of the face (nose, mouth, eyes) ... there are variants of classification based on frequency characteristics ... Why am I saying all this, it's simple. What exactly would be
a characteristic against which to classify a market condition? Yes, and so as not to step on the rake, I want to draw your attention that this characteristic belongs to the type of hidden ... ...i.e. it cannot be obtained using conventional TA methods. The signal of start of a trade ... The grail, the dream of all traders ;) Any network can be trained to very reliably detect class affiliation, an example would be text recognition software ... which most people work with all the time. The main thing is to set the task correctly.
 
Reshetov:
Multi-layer neural network interpolation can be done like two fingers on the pavement.
But such interpolation is as useful for real unsmoothed data as a goat's milk... The mouwing from a high or low (best of all LWMA, by the way) interpolates very well, but the price recovery from a forecast mouwing makes everything clear: the mouwing forecast error is multiplied by an appropriate coefficient.

Somewhere in a neighboring thread about scalping based trading systems there is a message about the existence of FCs with highly smoothed data (exactly in those FCs scalping is not limited in any way, because it is useless). Maybe their interpolation results will be better...
 
Reshetov:
The price series, like any other continuous series, can be approximated without any problem. Interpolation with a multi-layer neural network can be done like two fingers on the pavement. Extrapolation of non-periodic series by neural networks is a waste of time.
I disagree only with the last point, it is possible to extrapolate a stochastic series, but one must impose boundary conditions... Say, in my opinion, there could be indicative answers to questions such as: What are the trends of the reference levels? Maybe another goal more practical, say the direction of price movement and the angle of the asymptote to the MA over a certain period.
 
Mathemat:
Reshetov:
Interpolation by a multi-layer neuron can be done like two fingers on the pavement.
But such interpolation is more useful for real unsmoothed data than a goat's milk... The mouwing from high or low (by the way, LWMA is best) interpolates very well, but when price is restored from the forecast mouwing everything becomes clear: the error of forecast mouwing is multiplied by an appropriate coefficient.
I may be wrong, I came across an article somewhere, where the author advocated using interpolation based on NS as a filter ... interesting idea, but it has to be tested experimentally ... I doubt it's that simple, but it's worth experimenting with the idea.
 
rip:
maveric:

For classification tasks, it is our interpretation of the data that is taken into account. Roughly speaking, in a letter recognition task, in a set of examples of the letter A, the letters X and U and others should not appear :)

I want to do both. Maybe two grids, maybe one as it turns out. The first step the grid classifies the current situation. If it gives a sufficiently clear signal of the beginning of a trend, then the second step is an attempt to look into the future in order to estimate how much money may be made on this trend.

I assume that approximation of financial series is worse than classification.

Any classification, I am not talking about clustering, requires images. Say to classify images of faces based on a set of features such as distances between some specific parts of the face (nose, mouth, eyes) ... there are variants of classification based on frequency characteristics ... Why am I saying all this, it's simple. What exactly would be
a characteristic against which to classify a market condition? Yes, and so as not to step on the rake, I want to draw your attention that this characteristic belongs to the type of hidden ... ...i.e. it cannot be obtained using conventional TA methods. The signal of start of a trade ... The grail, the dream of all traders ;) Any network can be trained to very reliably detect class affiliation, an example would be text recognition software ... which most people work with all the time. The main thing is to set the task correctly.
I don't know yet what exactly I am going to give as input to the networks.
I think I will select different price indicators and something else.
If there are some dependencies, I hope to find them with sufficient accuracy :)
And I am far from thinking of writing a grail, 6-7 guesses out of 10 is already a good result :)
 
maveric:

I don't know exactly what I'm going to give the nets as input yet.
I think I will pick up different price indicators something else.
If there are any dependencies, I hope to find them with sufficient accuracy :)
And I am far from thinking of writing a grail, 6-7 guesses out of 10 is already a good result :)

70% ... it's practically a grail ;)
 
rip:
Reshetov:
The price series, like any other continuous series, can be approximated without any problem. Interpolation with a multi-layer neural network can be done like two fingers on the pavement. Extrapolation of non-periodic series by neural networks is a waste of time.
I disagree only with the last point, extrapolation of stochastic series is possible, but boundary conditions have to be imposed...
In principle, nobody forbids to hammer nails with a TV set. Although normal people have long ago learned to correctly and unmistakably choose the length of a position: http://gzt.ru/business/2007/03/04/220019.html
 
Reshetov:
rip:
I would only disagree with the last point, extrapolating a stochastic series is possible, but you have to impose boundary conditions ...
Yes, in principle, no one forbids hammering nails with a TV set. Although, normal people have long ago learned how to correctly and unmistakably choose the length of a position: http://gzt.ru/business/2007/03/04/220019.html

I'm not quite sure what nailing has to do with it... and what does an insider conspiracy have to do with the length of a position?!
 
rip:
Reshetov:
rip:
I would only disagree with the last point, extrapolating a stochastic series is possible, but you have to impose boundary conditions ...
No one forbids hammering nails with a TV set, though normal people have long ago learned how to correctly and unmistakably choose the length of a position: http://gzt.ru/business/2007/03/04/220019.html

I'm not quite sure what nailing has to do with it... and what does an insider conspiracy have to do with the length of a position?!
By the time you figure it out, it'll be too late.
 
Reshetov:
rip:
Reshetov:
rip:
I would only disagree with the last point, extrapolating a stochastic series is possible, but you have to impose boundary conditions ...
No one forbids hammering nails with a TV set, though normal people have long ago learned how to correctly and unmistakably choose the length of a position: http://gzt.ru/business/2007/03/04/220019.html

I'm not quite sure what nailing has to do with it... and what does an insider's conspiracy have to do with the length of a position!
By the time you figure it out, it'll be too late.
Philosophy, religion... they are designed to create a solid ground for faith. But it's up to everyone what to believe in...

2Reshetov: I perfectly understand your position and think it is correct, but progress must be everywhere ...
1...121314151617181920212223242526...28
New comment