Machine learning in trading: theory, models, practice and algo-trading - page 378

 
Maxim Dmitrievsky:

https://www.youtube.com/channel/UCLk-Oih8VlqF-StidijTUnw

found something to do for the weekend :) ar for nubas

And here's a guy engaged in algotrading even

The Internet is full of good books on data processing in R. The case is worse with the application of packages - it is necessary to look for information on each of them separately.
 
Yuriy Asaulenko:
. With the application of packages is worse - it is necessary to look for information on each of them separately.

In 99% of the cases, you don't have to search for anything.

Each package has very good documentation on its use.

Often there is a link to wignets, introdakshin... besides documentation.

Each function of the package necessarily has a link to the description of the algorithm that exactly this function implements, which is very important, since the same function can have a different author of algorithms with the same name. This is especially important if the need arises to compare the same functions in different packages, e.g. with matlab.

There are websites, for example, where you can search for ready-made answers, or ask your own question.

Usually all of this is enough to raise eyebrows.


If none of that is satisfactory, google it. Just type in the name of the function and you get ....


PS.

For general development you get here for very limited money (10 rubles/book). A bunch of books with a lag of a couple of months.

CRAN Packages By Name
  • cran.r-project.org
The package will formally test two curves represented by discrete data sets to be statistically equal or not when the errors of the two curves were assumed either equal or not using the tube formula to calculate the tail probabilities
 
SanSanych Fomenko:

In 99% of the cases, you don't have to search for anything.

Each package has very good documentation on its use.

Often there is a link to wignets, introdakshin... besides documentation.

Each function of the package necessarily has a link to the description of the algorithm that exactly this function implements, which is very important because one and the same function can have a different author of algorithms with the same name.

There are websites, for example, where you can search for ready-made answers, or ask your own question.

Usually all of this is enough to the bureaus.

If none of that is satisfactory, google it. Just type in the name of the function and you get ....

PS.

For general development you get here for very limited money (10 rubles/book)

That's what we do. On their own, PDF docs to packages are actually hard to come by.

Of the books, Kabakov is very good, imho. You can start from scratch.

There's also O'Reilly, but in Aglitsky. It's a hard book to read. It's hard to assess).

 
Yuriy Asaulenko:
The Internet is full of good books on data processing in R. The case is worse with the application of packages - it is necessary to look for information on each of them separately.

Well, here you can solve various examples and look around, in general, it's useful.
 

The most effective way to learn a language is to write and analyze scripts from working and reproducible code examples. At the beginning of this thread the topicstarter several times tried to post code, but got no support in the discussion. Unfortunately there is only verbiage on the thread now, without any examples, code or data.

It is useless blowing of air.

For absolute beginners in learning the R language I will add a link to a resource in Russian.

Good luck

Введение в R
  • Alexander Novopoltsev
  • rstudio-pubs-static.s3.amazonaws.com
Установить среду R Установить графическую оболочкуRStudio. Установка R Markdown (для создания автоматически генерируемых отчетов): в RStudio автоматически при первом создании файла с расширением “.Rmd”. Установка библиотек расширений: набрать в консоли install.packages(“pname”), где “pname” - название библиотеки. Полный список библиотек по...
 
Vladimir Perervenko:

The most effective way to learn a language is to write and analyze scripts from working and reproducible code examples. At the beginning of this thread the topicstarter several times tried to post code, but got no support in the discussion. Unfortunately there is only verbiage on the thread now, without any examples, code or data.

It is useless blowing of air.

For absolute beginners in learning the R language I will add a link to a resource in Russian.

Good luck


The topic is titled Machine Learning - Theory and Practice, so it's okay... In fact, there's nothing to learn, the most difficult thing is to develop a strategy using NS (and you have to have a good imagination for that), and you can teach something in two lines.

And of course, if you do everything by the book and use ready-made examples, you will never make money :)

P.S. I'm in a branch of ready-made bot, which earns better than anything else here :))

 
Maxim Dmitrievsky:

P.S. I'm in a branch of ready-made bot, which earns better than anything else that is here :))

Put more in a branchWhat is your trading strategy! There the man badly wants to share his strategy with him, and make a million bucks out of 100 bucks.) Let him suffer).
 
Has anyone tried training two nets instead of one, one network to buy, the other to sell? Maybe you can improve accuracy that way?
 
elibrarius:
Has anyone tried to train 2 nets instead of one, one for buying and one for selling? Maybe this way it is possible to improve accuracy?


https://habrahabr.ru/post/243211/

Here's a good post about the retraining of the NS, I reproduced the results, the predictions of ARIMA I got the same, but the NS gave something wrong :)


Нефтяные ряды в R
Нефтяные ряды в R
  • habrahabr.ru
Пометьте топик понятными вам метками, если хотите или закрыть
 
Maxim Dmitrievsky:


https://habrahabr.ru/post/243211/

Here is a good post about the overtraining of the NS, I reproduced the results, the predictions of ARIMA I got the same, but the NS gave something wrong :)


I have already expressed my negative opinion about NS. the main complaint: there is not even a hint of NS overtraining in this thread. Moreover, supporters of NS do not even try to do so. And without evidence of the lack of retraining, it's all a numbers game, a very interesting intellectual game, about layers, perseptrons, and so on and so forth...