Machine learning in trading: theory, models, practice and algo-trading - page 1095

[mytarmailS]

Maxim Dmitrievsky:

he's a genius, if you ask me, the book of the year, even though it's old, along with Wapnick.

The one I threw was a different book, more like a brochure...

I tried to read the one you threw two years ago, but got stuck in the formulas and gave up (( I'm a nun))

[mytarmailS]

Vizard_:

***I tried it a long time ago. I remember exactly this file.
You have to, otherwise it will show up "tomorrow like yesterday". Take at least the first difference for sure and convert it to + if negative doesn't work.
You've got to do better than that. I don't know what kind of implementation you've got there. Maybe it's already pre-processing. Try different ones.
and compare it to a metric you're happy with. Why wait for realtime, simulate it this way, it makes no difference. X <- (X - min(X)) / (max(X) - min(X)) etc., for experiments...

So far I'm doing exactly as you said, everything in its simplest form, but I'm taking the series as it is, I'm not deleting the trend

#  фун. нормализации
norm <- function(X) {X <- (X - min(X)) / (max(X) - min(X)) ; return(X)}
#   два фейк вектора
x <- rnorm(10)
y <- rnorm(10)
#   нормализирую их
x <- norm(x)
y <- norm(y)
#   фун. евклидовой близости
euclid.dist <- function(x1, x2) sqrt(sum((x1 - x2) ^ 2))
#   рез. насколько близки два вектора
euclid.dist(x,y)

I'll try it with live data tomorrow and then I'll see.

[mytarmailS]

Vizard_:

At the expense of the rattle... Oooooh very long time I spin it, nothing outstanding achieved, I do not know what you do, probably some kind of cunning overkill hz, but here's an interesting point. maybe you can help.

If you take a row and divide it into a buyer and a seller.

Everything above zero buy, everything below zero sell, it's like an attempt to bypass fractality.

x <- cumsum(rnorm(1000))+1000
x <- diff(x)

buyers <- x[x>0]
sellers <- x[x<0]

The difference of these predictions will be the final forecast, try it, I was able to make much better and more frequent trend predictions than the usual stated method

[mytarmailS]

Igor Makanu:

Search will not work.

I wrote myself that it does not work, because not properly formed forecast, there is a nuance that everyone missed, including me, but now I understand, after several years

[mytarmailS]

Vizard_:

At the beginning of the branch wrote that it is rubbish. I do not use. In this thread, there was
a moe competition. I was lucky and got first place)))

I see) just do not understand why you asked not to call a rattle

[mytarmailS]

Vizard_:

The half-wits from the neighboring branches are climbing... Let them guess)))

))

[Igor Makanu]

Maxim Dmitrievsky:

Well, in the Hilbert space the questions are probably easier to solve

Yeah, then we'll remember about the strange and scary attractor.... I'm still inclined to the opinion, that if anything works in the part of matrix model of market prices, it will be based on work with matrixes, at least the SSA algorithm I took apart the screws, I read more literature from this topic, everywhere the matrix is based on similar formulas: matrix of states (trajectories) and then make different manipulations with matrixes, and it all very similar to work with neural networks

I can't even imagine how and where to put it... I don`t even know, but I`m sure there is one big matrix

[mytarmailS]

Igor Makanu:

I read here that you used to do parsing... Tell me, is it realistic to collect posts by keywords all over the web, all over the web, articles, threads, blogs, Facebooks, tweets, and other swamps and sort them by date and time?

[mytarmailS]

Vizard_:
https://radikal.ru/video/Fm0qUroH5A1

strongly...

you're not predicting a price, you're predicting movement in the sign space

[Igor Makanu]

mytarmailS:

I read that you used to do parsing... Tell me, is it realistic to collect posts by keywords all over the web, all over the web, articles, threads, journals, facebook, tweets and other swamps and sort them by date and time?

Unfortunately WEB programming is not engaged, your question from this area, you need a search bot on the server

I can and did parsing specific network resources - the site, ie, I took the necessary information to my program from a third-party site and processed, if I am not mistaken, then on this forum, I could mention a site with some online data, like the gold was my interest, somewhere I took online prices for the metals

ZS: google GET and POST https://ru.wikipedia.org/wiki/POST_(HTTP)