Machine learning in trading: theory, models, practice and algo-trading - page 718

[СанСаныч Фоменко]

And here is the lime package (available in R).

This is how its selection principle is explained.

The general approach lime takes to achieving this goal is as follows:

1. For each prediction to explain, permute the observation n times.
2. Let the complex model predict the outcome of all permuted observations.
3. Calculate the distance from all permutations to the original observation.
4. Convert the distance to a similarity score.
5. Select m features best describing the complex model outcome from the permuted data.
6. Fit a simple model to the permuted data, explaining the complex model outcome with the m features from the permuted data weighted by its similarity to the original observation.
7. Extract the feature weights from the simple model and use these as explanations for the complex models local behavior.

PS.

Note that both packages determine the influence of the predictor on the Predictor, not its importance in the construction of the model, which is considered a black box.

[Mihail Marchukajtes]

I use the package vreat to select predictors and I understand it as follows. So... thoughts aloud....

This package determines the importance of predictors and those predictors that have the maximum score (conventionally) are considered important. When pre-processing the data and getting the result R only says that there are both latent and explicit dependencies between the given set of predictors and the output variable. It means he only says there are dependencies and it's up to the Optimizer to find those dependencies.

R- says there are dependencies, Optimizer says. Yes, here are these dependencies...... Something like this....

[Alexander Ivanov]

Hi!

Guys, when will the super bot with AI finish?

You'll get old waiting.)

[Mihail Marchukajtes]

Alexander Ivanov:

Hi!

Guys, when will the super bot with AI finish?

You'll get old waiting.)

What do you mean you're done? You think you created the AI and put your hands in your pants... go walk......??? Is that how it is???

The AI requires weekly grooming as well as daily monitoring of its performance. Personally, I already have all the codes finished and polished. Now all I do is train and test the resulting models. So there you go...

[Maxim Dmitrievsky]

Mihail Marchukajtes:

What do you mean you're done? You think you've created an AI and put your hands in your pants...? go walk......??? Is that it???

The AI requires weekly grooming as well as daily monitoring of its performance. Personally, I already have all the codes finished and polished. Now all I do is teach and test the resulting models. That's about it...

No, the AI is a person and doesn't want to be controlled

[Konstantin Nikitin]

Maxim Dmitrievsky:

No, the AI is a person and doesn't want to be controlled.

Well any personality has been controlled by someone during its growth and formation. Parents, teachers, etc., etc., so until that person reaches a responsible relationship to what they are meant to do. It is necessary to control her, or they will deprive her of parental rights.

[Mihail Marchukajtes]

Konstantin Nikitin:

Well any personality during its growth and formation is controlled by someone. Parents, teachers, etc., etc., so until that person has reached a responsible relationship to what they are meant to do. Control her by all means, or they'll take away her parental rights.

Especially since he can be such a prankster that you can't leave him unattended. You live and learn. It turns out that as long as there will be AI, he will always sit at the desk, no matter how clever he would not be.... IMHO

[Maxim Dmitrievsky]

Konstantin Nikitin:

Well any personality during its growth and formation is controlled by someone. Parents, teachers, etc., etc., so until that person reaches a responsible attitude towards what he is meant to do. Control her by all means, or they'll take away her parental rights.

We are too dumb for them, and they are perfect... we will create them and die out.

[Alexander Sevastyanov]

Mihail Marchukajtes:

... Personally , I have all the codes already finished and polished. Now I'm just learning and testing the resulting models. Like so ...

Is there anywhere you can see the result of "finished and polished" codes? How they work as a result of training and testing? Something like myfxbook monitoring?

[Mihail Marchukajtes]

Alexander Sevastyanov:

Is it possible to see the results of "finished and polished" codes anywhere? How do they work as a result of training and testing? Something like myfxbook monitoring?

Unfortunately no. I train models every week. My signal statistics are depressing because there is a huge difference between tests and real trading. Not in the divergence of results, but in the nuances of real time.

I don't know why I got errors when I tried to rewrite it.

If I wanted to rewrite it for one bar then it would be wrong ... If I wanted to rewrite it for one bar then it would be wrong ...

   if(prev_calculated>rates_total || prev_calculated<=0)// проверка на первый старт расчёта индикатора
     {
      limit=100; // стартовый номер для расчёта всех баров
        }else{
      limit=rates_total-prev_calculated; // стартовый номер для расчёта новых баров
     }