Machine learning in trading: theory, models, practice and algo-trading - page 3112

[Maxim Dmitrievsky]

Evgeni Gavrilovi #:

What features did you use? A complete list would be appreciated)

Various mock returns, nothing special. I've written before that sometimes just volatility works well, it's again converted retournes.

Too much is bad, too little is bad. Somehow I get an optimum around 10 fics with different periods increasing by a fixed step. Let's say, range(10, 100, 10)

but this is how it is for my implementation, I don't pretend to have derived some magic formula.

Is there anything else in the time series? )

[mytarmailS]

I got banned, the thread died))

[mytarmailS]

I sparred all MT5 signals with the filter "reliable" to look at the distribution by years, well, to understand how much and how long "reliable TSs" "live".

Here is the distribution

years
2015 2018 2019 2020 2021 2022 2023 
   1    4    5   34  130  549  957 

Something to think about....

Pars code, if anyone needs it

library(rvest)
library(readr)
n_pages <- 35 
url <- "https://www.mql5.com/ru/signals/mt5/page"
urls <- paste0(url, 1:n_pages)
years <- lapply(urls, \(Url) Url |> read_html() |> html_nodes(".signal-card__growth-label") |> html_text() |> parse_number()) |> unlist()
hist(years)

df <- data.frame(table(years))
library(ggplot2)
ggplot(df,aes(years,Freq)) +
  geom_bar(stat='identity',fill=colors()[128]) 

[mytarmailS]

Sparse all of Sanych's posts from the first 100 pages of the forum

converted the text into a "bag of words"

removed "stop words"

visualised it as a "word cloud" (the bigger the word, the more often it occurs).

Cool ))))

library(xml2)
library(rvest)
urls <- paste0("https://www.mql5.com/ru/forum/86386/page", 2:100)
author <- "СанСаныч Фоменко"

li <- list()
for(url in urls){
txt <- url |> read_html() |> html_nodes(".text") 
txt |> html_node(".fquote") |> xml_remove() #  remove fquote

Authors <- txt |> html_node(".author") |> html_text()
Content <- txt |> html_node(".content") |> html_text()
Comment_date <- txt |> html_node(".comment__date") |> html_text()

if(any(Authors==author,na.rm = T)){
la <- lapply(seq_along(Authors), function(i) {
       list(author=Authors[i], text=Content[i], date=Comment_date[i]) })
la <- la[Authors==author]
la <- la[!sapply(la, is.null)]
li <- append(li, la)
}
print(url)
}
text2word <- function(text) strsplit(text, "\\P{L}+", perl = TRUE) |> unlist()
words <- lapply(li, function(x) text2word(x$text))  |> unlist() 
bag_of_words <- table(words)

library(stopwords)
stopw <- stopwords("ru", source = "stopwords-iso")
bag_of_words <- bag_of_words[!names(bag_of_words) %in% stopw]

library(wordcloud2)
bag_of_words |> as.data.frame() |> wordcloud2()

And this is what I was talking about back in 2016.

......

Yes, you can do almost anything in R in one screen, it's a marvellous language....

[Renat Fatkhullin]

For a new inoculation/fuelling take a look at: Quantitative trading

[СанСаныч Фоменко]

mytarmailS #:

Sparred all of Sanych's posts from the first 100 pages of the forum

minified the text into a "bag of words"

minus the "stop words"

visualised as a "word cloud" (the bigger the word, the more frequent the occurrence)

That's cool.)

And this is what I was talking about back in 2016.

......

Yes, you can do almost anything in R in one screen, it's a marvellous language.....

That's great, thanks!

[СанСаныч Фоменко]

Renat Fatkhullin #:
For a new inoculation/refuelling take a look at: Quantitative trading

Just an amazing selection!

[mytarmailS]

СанСаныч Фоменко #:

That's just great, thanks!

The original purpose was to count how many times you said the word GARCH, because it seems to me that fifty, not less)))))

but while learning to parse posts on the forum, I already forgot and forgot what I was doing it for....

Let it remain a secret)))

[mytarmailS]

Renat Fatkhullin #:

Is there any news on the mt package?

[Renat Fatkhullin]

mytarmailS #:
Is there any news on the mt package?

Not yet, I can't get my hands on it.