Machine learning in trading: theory, models, practice and algo-trading - page 2060

 
Evgeny Dyuka:
All charts in this thread start in the lower left corner and end in the upper right corner)
I wish I had lived that way...

no not all)

 
Rorschach:
I have doubts.

I did that, I even described it here...

Only I went even further...

I made forecasts for all indicators, then I used the obtained forecasts as new signs, and then I made more forecasts based on the signs and their forecasts, and so on, 7 times, until the error decreased...

This is all based on MGUA self-organization or meta traits from modern...


Thanks for the link, great kid, fun to read ))

 

When you need to describe an extremum in a more human form than a clear and not very realistic zigzag.

Why is that? It's just an alternative with a more human layout of chart extrema. Why not ZZ, ZZ has only one extremum, one point, it does not take into account inevitable shifts in the non-stationary market...

The solution is this: we make a pattern of extrema and look at the correlation between the current price and the pattern...

code in the world's best language )

y <- c(1:10,9:0)
x <- cumsum(rnorm(length(y)))

ccor <- cor(y,x)

layout(1:2)
plot(x,t="l",main = paste("corelation",round(ccor,2)))
plot(y,t="l",main ="шаблон екстремума")

The pattern can be anything, a trend or whatever, you can dream it up...


This is what it looks like on a sliding window correlation calculation chart with signal lines -0.7 and 0.7


It seems to me that it defines extrema very humanly and filters the unimportant from the significant ones

Then you can try to train the regression model and see what will happen

xx <- cumsum(rnorm(100))
layout(1:2)
plot(xx,t="l")
cor.vec <- rep(NA, length(xx))
for(i in 10:(length(xx)-10)){
  ii <- (i-9):(i+10)
  cor.vec[i] <- cor(xx[ii] , y)}
plot(cor.vec,t="l",col=4)
abline(h=c(-0.7,0.7),col=2)
 
mytarmailS:

When you need to describe an extremum in a more human form than a clear and not very realistic zigzag.

Why is that? It's just an alternative with a more human layout of chart extrema. Why not ZZ , ZZ has only one extremum, one point, it does not take into account the inevitable shifts in the non-stationary market...


For a zigzag, it would be enough to teach the neural network to predict the length of the shoulder - longer than the previous one or shorter (as a target we need long ones).

Files:
 
Evgeniy Chumakov:

For a zigzag, it would be enough to teach the neural network to predict the length of the shoulder - longer than the previous one or shorter (as the target needs long ones).

Hmmm... That's also an interesting idea.

Have you tried it?

What's in the file?

 
mytarmailS:

Hmmm... That's also an interesting idea.

Have you tried it?

What's in the file?


I haven't tried it, because I don't know anything about neural networks and I'm as dumb as a samovar.... I was stupidly looking for similar sections, but it doesn't always repeat like in the past.


In the file:

First column - event result (positive sign = long arm, negative sign short)

Second column - variant of " long arm " event

Third column - the event variant "short leverage


Generally speaking, the two outermost columns are possible events.

 
Evgeniy Chumakov:

In the file:

Ah, I see... Now I'll try but I will do mine, because you have not thrown prices or signs, and it is not clear how to build a target

 
mytarmailS:

and it's not clear how the target builds


Patterns.

 
mytarmailS:

When you need to describe an extremum in a more human form than a clear and not very realistic zigzag...

...

...

....

Then you can try to train a regression model and see what happens

Trained...

Gray is the original data, blue is from the model

I also inserted a trace for illustration how the model fades in the new data