Machine learning in trading: theory, models, practice and algo-trading - page 2742

New comment
 
СанСаныч Фоменко #:

It says classDist {caret}, i.e. it specifies a particular function that is part of the caret PACKAGE

As I understand, you do not know R. Then why are you wasting your time on this thread and on MO in general?

Without mastery of R, discussion of MO is meaningless.

I kept silent about entropy only because cross-entropy is a standard loss function for classification models.... MO is implemented not only in R! (knowing one library and not knowing the nature of the entities it operates on - you go without understanding the direction of your movement).

to you even more difficult question - why do you categorically disassociate from statistics, when you declare about"information theory"?... while it was created exactly as a

The field lies at the intersection of mathematics , statistics , computer science , physics , neurobiology , information engineering and electrical engineering .

indeed, the discussion is subjectless, if you operate in snippets and your ego (and even about someone, not only about yourself), and not the subject of the dialogue... the thread does not change, unfortunately (specificity and subject matter in answers is not added)
 
Maxim Dmitrievsky #:
Once again this obtuse mouthpiece is calling everyone to the truth, but has not yet decided which

You've already bored the moderator so much that he's tearing everything down.

Do not read the provocative post of user JeeyCi (his post is a provocation and demand to "continue the banquet").
Yesterday I deleted several posts with swearing and boorishness with personal attacks, guided by this - I have deleted the posts of JeeyCi .

Forum on trading, automated trading systems and testing trading strategies.

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

MetaQuotes, 2022.08.31 15:14

Continuation of personalities will lead to weekly bans.

I made two warnings in the thread, they were ignored, and then I deleted several posts with swearing.
The only literary post there (that was readable at all) was your post - this one (which started it all yesterday):

Forum on trading, automated trading systems and testing trading strategies

...

Maxim Dmitrievsky, 2022.09.10 12:15

There are model based, model agnostic and mixed feature selection. If you take agnostic, it is correlation and mutual information (entropy based). The latter differs from the former in its ability to capture non-linear dependencies, otherwise it is the same. It is hard to talk about any relation of feature to target in this case, even impossible. It's just a correlation. But it is useful to get rid of uninformative features.

You can do it in a sliding window, or in an elusive window, or in a sliding window, or in a rubbing window

If you want to determine causality specifically, that's causal inference, including using MO, which I don't know how to apply to a time series, haven't studied the topic.

And all the previous methods do not work for finding causality, but only for optimal training of algorithms.

So once again citizens can't concentrate and take the flies out of their cutlets.

About the great and omnipotent R we have already heard many times. Obviously, if you put a monkey behind it, it too can consider itself a statistician and analyst, so great is it.

Yes, I will occasionally delete swear words, especially if they last half a day and two pages of text for example (like yesterday).

----------------

This thread is very popular (it is even read on the English-speaking forum and is considered the key thread on this topic).
So please - less swearing.

 
mytarmailS #:

If you analyse the TS of more or less successful traders, you will see that all of them trade levels.

I have not seen a single successful trader who trades with the help of indicators.

A level is a clear and understandable entry point with a clear stop....

If you can trade with low risk, you don't need anything else, low risk per trade/precise entry is the most important thing!

With the help of MO you can look for levels of PD/SP those exact entries, it is not trivial, not simple, you can't read about it in blogs about MO here you need to use your own head....

You can also draw levels on the sb chart and it is also a time series. Everyone's had enough of you, we're not responding to you anymore. You're talking rubbish day in and day out.
 
mytarmailS #:

Here is an example on a randomly generated sample of 5 traits and 1 binary target

forrest and fiche selector

The task queue has been unloaded a bit - it became possible to run the script. I run it and get an error.

> install.packages("randomForest")
Warning in install.packages :
  unable to access index for repository https://cran.rstudio.com/src/contrib:
  cannot open URL 'https://cran.rstudio.com/src/contrib/PACKAGES'
Installing package into ‘C:/Users/S_V_A/Documents/R/win-library/4.0’
(as ‘lib’ is unspecified)
Warning in install.packages :
  unable to access index for repository https://cran.rstudio.com/src/contrib:
  cannot open URL 'https://cran.rstudio.com/src/contrib/PACKAGES'
Warning in install.packages :
  package ‘randomForest’ is not available (for R version 4.0.5)
Warning in install.packages :
  unable to access index for repository https://cran.rstudio.com/bin/windows/contrib/4.0:
  cannot open URL 'https://cran.rstudio.com/bin/windows/contrib/4.0/PACKAGES'

> library(randomForest)
Error in library(randomForest) : нет пакета под названием ‘randomForest’

I understand correctly that the programme wants an old version of R 4.0?

Well, I searched for an old version and didn't find it. Terrible incompatibility is repulsive of course.

 
Aleksey Vyazmikin #:

The task queue has been unloaded a bit - it became possible to run the script. I run it and get an error.

Do I understand correctly that the programme wants the old version R 4.0?

I have R-3.6.3.

I'm writing this on the old R-3.6.3 for my own reasons, so it's my problem...

I couldn't imagine that the package would be removed from the tap....

Aleksey Vyazmikin #:

I understand correctly that the programme wants the old version of R 4.0?

correctly

Aleksey Vyazmikin #:

Well, in general I searched for the old version and did not find it. Terrible incompatibility is repulsive of course.

Listen, maybe you can't go into trading, with such a smikalka ??? ))

With compatibility there everything is fine, python, for example, only envy such compatibility....


Also see

https://stackoverflow.com/questions/62541885/package-randomforest-is-not-available-for-r-version-4-0-2

Try it on the current version

urlPackage <- "https://cran.r-project.org/src/contrib/Archive/randomForest/randomForest_4.6-12.tar.gz"
install.packages(urlPackage, repos=NULL, type="source")

 
To summarise Sanych's theory (since he himself failed to formalise it properly and give examples):

*his way of feature selection is based on correlation, since "relation" and "connection" are definitions of correlation.

*this way we do implicit fitting to history, similar in meaning to LDA (linear discriminant analysis) or PCA, simplify the learning process, reduce error.

*There is not even a theory that the trained model should perform better on new data (not involved in estimating the links between features and targets), because the features have been fitted to the trait or (worse) to the available history.

*The situation is somewhat improved by averaging QC in a sliding window, like you can estimate the spread and select more stable ones. At least we have some statistics to rely on.

*I was thinking of causality or a statistically significant relationship, but that's not the case in his approach.
 
Maxim Dmitrievsky #:
To summarise Sanych's theory (since he himself failed to formalise it properly and give examples):

*his way of feature selection is based on correlation, since "relation" and "relationship" are definitions of correlation.

*This way we make an implicit fit to history, similar in meaning to LDA (linear discriminant analysis) or PCA, simplify the learning process, reduce error.

*There is not even a theory that the trained model should perform better on new data (not involved in the estimation of feature-target relationships) because the features were previously fitted to the trait or (worse) to the entire available history.

*By relation I meant causality or a statistically significant relationship, but that is not the case in his approach.

With all due respect, but this is not a summarisation (not a digest or summary). It's filled with personal attitudes and unfounded attacks.

you'd think someone would have a valid theory where "a trained model should work on new data" :-) and validated..yep.

 
Maxim Kuznetsov #:

with all due respect, but this is not a summarisation (not a digest or summary). This is a personal attitude and unfounded attacks.

You'd think someone would have a valid theory where "a trained model should work on new data" :-) and validated..yep.

And if you read carefully, you can see the ambush in point 2, i.e. the initial fit to the story. That's why it has a learning error drop.

Point 4 is a bit more optimistic if it is not done on all available history. It should only be done for traine sampling, for goodness of fit. To get an adequate estimation of the model on new data.

Not known to be into psychology, so it never squirts anywhere. And I don't know anyone personally.
 
СанСаныч Фоменко #:

There is not enough power to get to the EA level. But the result of the model fitting error: from 8% to 22% is a fitting error that differs little in the fitting section and out of sample.

This kind of hints that the fit to the whole history was done before training. If this is not the case, please correct me. At what interval were the features estimated/selected and at what interval was training done?

I just have a similar method, I can share the results this weekend. Only if there's substantive communication rather than a word game.
 
Maxim Dmitrievsky #:
is based on correlation because "relation" and "relationship" are definitions of correlation.
Relation and relationship are definitions of correlation???? Seriously???

The links in a chain around your neck are connected, they have a relationship. That's correlation?

I'm in a relationship with a girl, the relationship between us is correlation???

Correlation is first and foremost a measure! Stupid.

1...273527362737273827392740274127422743274427452746274727482749...3688
New comment