Machine learning in trading: theory, models, practice and algo-trading - page 296

[mytarmailS]

Has anyone tried to work with recurrence plots? You can read it herehttps://habrahabr.ru/post/145805/, in particular, to substitute MO instead of raw BP? might work as a variant

x <- cumsum(rnorm(100))
ox <- outer(x, x, function (a, b) abs(a-b))


par(mfrow=c(1,3))
plot(x,t="l")
plot(ox,t="l")
image(ox)

and more to readhttp://geo.phys.spbu.ru/Problems_of_geophysics/2005/20_Zolotova_38_2005.pdf

[fxsaber]

Help in finding the R-analog.

Forum on trading, automated trading systems and trading strategies testing

Libraries: Cross-platform library of original mathematical functions

Automated-Trading, 2017.03.27 11:39

Cross-platformlibrary of original mathematical functions:

Author: fxsaber

[Andrey Dik]

fxsaber:

Please help with finding the R-analog.

What's wrong with R?

[fxsaber]

Andrey Dik:
What's wrong with R?

In the description of the function there is a dash where you have to specify the R-analog. I doubt that R has no such stat function. To remove the dash, you need a name.

[Andrey Dik]

http://www.scilab.org/

[anonymous]

fxsaber:
In the description of the function, there is a dash where you have to specify the R-analog. I doubt that there is no such stat function in R. To remove the dash, you need a name.

cor(x, y, method = 'pearson')

[fxsaber]

anonymous:

cor(x, y, method = 'pearson')

This is completely different. There is no pattern.

Description	MQL5	R
Calculates Pearson, Spearman and Kendall correlation coefficients	boolMathCorrelationPearson(constdouble&array1[],constdouble&array2[],double&r) boolMathCorrelationPearson(constint&array1[],constint&array2[],double&r) boolMathCorrelationSpearman(constdouble&array1[],constdouble&array2[],double&r) boolMathCorrelationSpearman(constint&array1[],constint&array2[],double&r) boolMathCorrelationKendall(constdouble&array1[],constdouble&array2[],double&tau) boolMathCorrelationKendall(constint&array1[],constint&array2[],double&tau)	corr()

[toxic]

fxsaber:

Please help me find the R-analog.

"analog" - is a multifaceted thing, indeed, from the banal Pearson correlation, that is, the normalized scalar product of two vectors as a proximity metric, to the rest of the arsenal of machine learning, with the search for representative features and non-linear classification.

IMHO what is not so good for fans of R(r-astes), matlab, "math", etc. is that there is a habit and dependence on high-level complex functions with a simple interface and a wrong understanding of what they do, or rather what they CAN do, if you have access to all the parameters and guts, with an understanding of what and how, not just the interface and an article on hubra that there is such crap.

I call this process DE-ORTOGONALIZATION, or "mosaic consciousness", when a person is forced to fill his head not with the essence of algorithms, but with thousands of names of functions and parameters from some libraries. But given that neither 100k nor a yard of high-level functions won't solve all the issues of real-world engineering problems, since there will always be something to be "tweaked" with a soldering iron, this direction of development is risky.

[fxsaber]

toxic:

"analog" - thing is multifaceted, indeed, from the trivial Pearson correlation, that is, the normalized scalar product of two vectors as a proximity metric

Benchmark shows that R-corr is slower by orders of magnitude, since the algorithmic complexities are quite different. Therefore there is hardly any analogue in R showing normal speed characteristics.

[anonymous]

fxsaber:

This is different. There is no pattern.

...

apply(embed(pattern, length(signal)), 1, cor, y = signal, method = 'pearson')