Machine learning in trading: theory, models, practice and algo-trading - page 1763

 
Aleksey Vyazmikin:

What makes you think a public target would be your target?

I brush off that the target should be zig-zag because -

1) ZZ is the equivalent of ideal position direction decisions, it's an ideal to strive for.

2) Anticipate your question, but what about the other targets? The point is that the ZZ is hierarchically above all other targets, they can be called "local", and the ZZ is "global".

What is a local target? that's all you can think of -- buy not buy when sthastic>0.8 or buy not buy when 16:00, ZZ combines all these local targets in itself, and the opposite is not true.

You can use local targets to improve the forecast of the global target, but you should not confuse and even more so, compare them.

If you have a better solution, I would be happy to hear about it.


Aleksey Vyazmikin:

Can you read carefully? Earlier I wrote that the target cannot be taken at every bar, which prevents me from getting intermediate data from CSV file.

Shoot it at least once a day.

This is exactly the sign with a local target.

If it improves the forecast of the global target, then fine, if not, then it's garbage!

 
mytarmailS:

I brush off that the target should be a zig-zag because -

1) ZZ is the equivalent of ideal decisions about the direction of the position, this is the ideal to strive for.

2) I foresee your question, but what about the other targets? The point is that the zeros are hierarchically superior to all other targets, they can be called "local", and the zeros are "global".

What is a local target? that's all you can think of -- buy not buy when sthastic>0.8 or buy not buy when 16:00, ZZ combines all these local targets in itself, and the opposite is not true.

We can use local targets to improve the forecast of the global target, but we should notconfuse and even more so, compare them.

So I explained why ZZ, if you have a better solution I would be happy to listen.

I have not understood anything yet. Write specifically how you define the target. Is it a ZZ vector on each bar? Change of vector on the next bar? Please, be more specific and let's discuss whether it is good or bad.

 
Aleksey Vyazmikin:

I don't understand anything yet. Write specifically how you define the target. Is it a vector ZZ on each bar? Change the vector on the next bar? Please be more specific, and then we will discuss whether it is good or bad.

layout(1:2)
x <- cumsum(rnorm(100))
plot(x,t="l")

zz <- TTR::ZigZag(x,change = 3,percent = F)
lines(zz,col=4,lwd=2)

zz <- c(diff(zz),0)
zz[zz>=0] <- 1 
zz[zz<0] <- 0
plot(zz,t="h",col=2)

The usual classification is "1" buy "0" sell. I'm genuinely surprised by your lack of understanding, have you read any of Vladimir Perervenok's articles on MO?

 
mytarmailS:

The usual classification is "1" buy "0" sell. I am genuinely surprised by your lack of understanding, have you read even one article by Vladimir Perervenok about MO?

The code on R is incomprehensible to me. Let's put it in words for now - do I understand correctly, that you have built ZZ on the history and make a classification by the vector of the segment ZZ? Or is there a shift?

 

You are pushing in vain, if the series was not random - my bot would have pulled the pattern (it pulls any cycles, if there are any)

Otherwise - TS with adjustment to the current trend, no more

 
Aleksey Vyazmikin:

The code on R is not clear to me. Let's put it in words for now - do I understand correctly that you have built ZZ on the history and are making a classification by the vector of the segment ZZ? Or is there a shift?

Well, yes, the slope of ZZ is a sign for the classification, ZZ is shifted back by 1 point, thus, ZZ is predicted one step forward

 
mytarmailS:

Well, yes, the slope of ZZ is a sign for classification, ZZ is shifted 1 point back, it is predicting ZZ one step forward

What kind of ZZ do you use?

I assume that the most correct classification is closer to the middle of the segment?

 
Aleksey Vyazmikin:

1) What kind of ZZ do you use?

2) I guess the most correct classification is closer to the middle of the segment?

1) Yes, the usual ZZ, there is no difference from the point of view of AMO what kind of ZZ to use

2) yes exactly, AMO tries to build a kind of DTF (digital filter) that lags like all filters, therefore in the middle is always better prediction than on the edges

 
Maxim Dmitrievsky:

You're pushing in vain, if the series was not random - my bot would have pulled the pattern (it pulls out any cycles, if there are any)

Otherwise - TS with adjustment to the current trend, no more

If possible, it's possible, but only up to a certain level of accuracy or probability.

 
Rorschach:

According to the volume ratio, the price looks like a fractal


did you get anything out? and a question, and back and compare the series, in places where there was a big change in compression?