Machine learning in trading: theory, models, practice and algo-trading - page 1788

 
Valeriy Yastremskiy:

It's easy to see, you select the sections and see how the AFR of these sections differ at all TFs and you need to catch the differences.

Cap ... )))

 
Hello, are there any results? Can I see the code?
 
mytarmailS:

Cap ... )))

There are areas of sufficient length with different efficiency of various EA logics, and there are short drain zones, if you look at the reports, the rate of drainage is always higher than the deposit increase, and usually after a big drain there is no data. There are zones of small plums and usually BP returns to the zone of satisfactory work of owl. That is, the ACH can also be late.

Damn, I'm slicing in circles)))))

 
mytarmailS:


I do not know, think, try, as for me the entry point and easier and more objective but it is imho

What can I give you specifically to test your hypothesis?

 
Aleksey Vyazmikin:

What exactly should I give you to test your hypothesis?

Just a price, and a marked target - what is the TS working, what is the system screwing up in the form of a vector 00001111100


You can simply draw the phase by the balance like this.

But this is trivial, it is better to consider the angle of inclination.


If you do not know the price and the balance chart, then I'll try to change it

 

Some justification for applying amateur radio methods is the Karunen-Loeve theorem for decomposing a random process into a series of orthogonal functions. This is partially similar to the Fourier decomposition, but differs significantly from it in that the coefficients are random variables rather than numbers.

The problems of applying this decomposition for us are that a) prices in the zero approximation are similar to SB, for which this decomposition has no predictive value b) in the more exact approximation, prices are essentially and complexly non-stationary, which leads to the fundamental inapplicability of this theorem (unknown ACF in the future).

 
mytarmailS:

Just the price, and the marked target - what is the TS working that is the system screws up in the form of a vector 00001111100

Look, the easiest variant - if we opened and there is a profit during the time interval we put "1", and if there is a loss, we put "0".

What should we do with those sectors that were not involved in trading? It turns out that there will be 3 values of the target?

 
Aleksey Nikolayev:

Some justification for applying amateur radio methods is the Karunen-Loeve theorem for decomposing a random process into a series of orthogonal functions. This is partially similar to the Fourier decomposition, but differs significantly from it in that the coefficients are random variables rather than numbers.

The problems of applying this decomposition for us are that a) prices in the zero approximation are similar to SB, for which this decomposition has no predictive value b) in the more exact approximation, prices are essentially and complexly nonstationary, which leads to the fundamental inapplicability of this theorem (unknown ACF in the future).

What can be the BP state cast? Which parameters to consider in the first approximation, averages, descriptive behavior, how to gradate, current, how to relate the scale from month to minute and to tick.

mytarmailS:

This is a simple price, and the marked target - what is the TP working, what is the system screwing up in the form of the vector 00001111100

To be honest I did not understand at once that at every price you should put 0 or 1)))))

 
mytarmailS:


Karoch throw me a price and the balance chart, and I'll spin it

For what period discount?

The standard report will do?
 
Aleksey Vyazmikin:

For what period discount?

Standard report fit?

For what you want but within reasonable limits, but please just the price and balance, nothing extra, so as not to torment my laptop)