Machine learning in trading: theory, models, practice and algo-trading - page 3498

New comment
 

Has anyone tried to classify two price series in one sample, "0" - rows with predictors from EURUSD sample and "1" - AUDCAD?

I think that if the separation is based on the available predictors after training, it may indicate fitting to a particular data structure - identify such predictors and throw them away. Then we build the model on a sample of one of the two and evaluate the change in the result.

Such cleaning, it seems, should allow to improve generalisation of price behaviour on different market instruments.

 

just like that



or just like this.



Does anyone realise that the amplitude, phase, frequency of the main harmonics directly determine the optimal settings of a trading robot?

 
mytarmailS #:

just like that



or so



Does anyone realise that optimal settings of a trading robot directly depend on the amplitude, phase and frequency of the main harmonics?

О! Haken Ashi indicator

 
Ivan Butko #:

О! The Haken Ashi Indicator

It doesn' t even smell like Haken Ashi.

 
Aleksey Vyazmikin #:

You know, between the closing and opening prices. It's like it's illiquid.

I don't have an answer to that question.


I think that you should distort ticks and then convert them into candlesticks, then you won't have this problem, but how to distort candlesticks directly is a question.

 
mytarmailS #:

I think it is necessary to distort ticks and then convert to candlesticks, then this problem will not occur, but how to directly distort candlesticks is a question

Yes, it is, if you change the trend on ticks then everything is fine.

original series


changed trend



But it's difficult to work with ticks, we need to think how to work with candlesticks directly.


The density of ticks in a candle also affects, if the density is high in ticks, the picture is better.


 
mytarmailS #:

original range

modified trend

Man, this could really be a good tool against overfit

 
mytarmailS #:

Man, this could really be a good tool against overfitting

And flipping, for class balancing.

interpolation smooths out the jumps on big moves, adds more points.

extrapolation adds points into the future, development options.

GAN, VAE add more pattern variations.

 
Maxim Dmitrievsky #:

more flipping, to balance the classes.

interpolation smooths out jumps in strong movements, adds more points

extrapolation adds points into the future, development options

GAN, VAE add more pattern variations

I look more towards decompositions, everything is clear and controllable there. You can decompose what you want and how you want, select the component you want, delete/replace/change it as you want.





I can't believe that the whole analysis is 7 lines of code (data generation, decomposition, visualisation).

What the hell is python, R is the best tool for analysis...

library(quantmod)
library(Rssa)
n <- 5000
p <- rnorm(n, mean = 0.02) |> cumsum() |> xts(Sys.time()+1:n) |> to.minutes(name = NULL)
r <- reconstruct(ssa(p))

par(mfrow=c(3,3))
chart_Series(p, name = "ORIGINAL")

for(i in 1:8) r[[i]] |> chart_Series(name = paste0("Comp",i)) |> print()
 
Unfortunately, no one gives a crap about your python and R, no matter how hard we try, this is still being handled by 3 cripples and activity is on the rise.

A difficult topic for most. If you look at articles on other MO and trading resources, they are doing some entry level crap. They start to get into it, then quit. There are no professional articles. So don't rely on crowdsourcing.

The most they can do is to quote Prado's articles.
1...349134923493349434953496349734983499350035013502350335043505...3529
New comment
Reason: