Machine learning in trading: theory, models, practice and algo-trading - page 2311

[Valeriy Yastremskiy]

Maxim Dmitrievsky:

Then their presence here is strange, since they all learned the Tao a long time ago.

I do not see a strong difference between csos, matstat econometrics and any other metrics))) everything starts with averaging))))

[Maxim Dmitrievsky]

Valeriy Yastremskiy:
I do not see a strong difference between csos, matstat econometrics and any other metrics))) everything starts with averaging))))

Then there is no point in swapping the shoe for a shoe

[Valeriy Yastremskiy]

Maxim Dmitrievsky:

Then there is no point in swapping the shill for the soap.

There may be a point, but it is random and expensive). In the sense of solving this range of problems is to identify something or to simplify calculations. Decomposition into stationary functions, for the sake of identifying cycles makes sense if these cycles exist.) In nature they definitely exist, and of course in the results of vital activity they are simply obligatory)))). But to compare these stationary functions with the phenomena that generated them... well that's probably not today....

[Rorschach]

Thoughts about 2 ways. 1 - look for the characteristics of the rows on which you can earn. It turned out to be not so easy, you look at the plot where people were able to earn, and the statistics do not show anything.

2 - Fitting the system to the series. The simplest case is multiplication of the original series by +-1 on some condition. If we still cannot detect regularities, why bother, take random parameters as a condition or change the direction of the deal at some time interval. As an example of owls in the trailer.

[Rorschach]

Experiments to normalize the distribution. Two files with euro and 2 with different psc.

[Forester]

Maxim Dmitrievsky:

Then there's no point in swapping the shoe for a shoe.

Maxim, you seem to have figured out the alglib MGC https://www.mql5.com/ru/forum/36408/page17#comment_9620369

pcabuildbasis(
double[,] x,       // матрица цен инструментов 
int npoints,       // количество цен для каждого инструмента
int nvars,         // количество инструментов
out int info,      // результат операции, любое положительное число - все ок
out double[] s2,   // массив разбросов / дисперсий для всех найденных векторов
out double[,] v)   // массив векторов, каждый вектор и есть искомые весы для выравнивая наборов вокруг нуля 

How to get e.g. 2 columns of principal components from s2 and v.
I assume x has to be multiplied/divided with these coefficients?
Do you have a formula?

The s2 and v arrays seem to be sorted, are the main ones at the beginning or at the end?

[Maxim Dmitrievsky]

elibrarius:

Maxim, you seem to have figured out the Alglib MGC https://www.mql5.com/ru/forum/36408/page17#comment_9620369

How to get e.g. 2 columns of principal components from s2 and v.
I assume that x has to be multiplied/divided with these coefficients?
Is there a formula?

Arrays s2 and v seem to be sorted, are the main ones at the beginning or at the end?

Did pca and lda, but I don't remember anymore, unfortunately it was a long time ago. Didn't get anything useful, so it's forgotten

[Forester]

Maybe someone else knows?

Here step 4, there is this code to create component columns, but so far I don't understand how to cycle and (*/+-) repeat this.

_, vecs = np.linalg.eig(covmat)
v = -vecs[:,1])
Xnew = dot(v,Xcentered)
print Xnew

OUT: [ -9.56404107 -9.02021624 -5.52974822 -2.96481262 0.68933859 0.74406645 2.33433492 7.39307974 5.3212742 10.59672425]

dot(X,Y) - Product of Honor (this is how we multiply vectors and matrices in Python)

[Maxim Dmitrievsky]

elibrarius:

Maybe someone else knows?

Here step 4, there is this code to create component columns, but so far I don't understand how to cycle and (*/+-) repeat this.

_, vecs = np.linalg.eig(covmat)
v = -vecs[:,1])
Xnew = dot(v,Xcentered)
print Xnew

OUT: [ -9.56404107 -9.02021624 -5.52974822 -2.96481262 0.68933859 0.74406645 2.33433492 7.39307974 5.3212742 10.59672425]

dot(X,Y) - Product of Honor (this is how we multiply vectors and matrices in Python)

https://gist.github.com/freemancw/2981258

[Forester]

Maxim Dmitrievsky:

https://gist.github.com/freemancw/2981258

Seen it. This is not it. It just rewrote the 3x3 matrix into variables. But the new component vectors are not calculated.
The result is to get 6 rows for each component (based on this example).