Machine learning in trading: theory, models, practice and algo-trading - page 2156

[Uladzimir Izerski]

Aleksey Vyazmikin:

You did not answer the question on the merits - okay.

What are you offering me, cooperation, exchange of experience?

I always judge the behavior of a forum member by his adequate behavior on the forum, his posts and the level of professional training.

I do not need someone's secrets. Experience and digging up secrets is for newbies level.

[Aleksey Vyazmikin]

Maxim Dmitrievsky:

I only have hard resampling with class separation in mind so far, but I think there are easier ways

How did you do it, which letter to read?

When saving the sample, I just reduced the financial result by a given value, and then if it was greater than zero, I put 1 for the target. Now I made a separate script for this, so it would be quicker to change the target. The shift can be made in different directions.

The article does not provide any details.

      double F_Rez_Buy=0.0;//Финансовый результат в случае покупки
      double F_Rez_Sell=0.0;//Финансовый результат в случае продажи
      double P_Open=0.0;//Цена открытия позиции
      double P_Close=0.0;//Цена закрытия позиции
      int Metka=0;//Метка для целевой
      P_Open=iOpen(Symbol(),Signal_MA_TF,iBarShift(Symbol(),Signal_MA_TF,Load_Data_Start,false));
      P_Close=iOpen(Symbol(),Signal_MA_TF,iBarShift(Symbol(),Signal_MA_TF,Load_Data_Stop,false));
      F_Rez_Buy=P_Close-P_Open;//Прошлый вход был на покупку
      F_Rez_Sell=P_Open-P_Close;//Прошлый вход был на продажу
      if((F_Rez_Buy-comission*Point()>0 && Load_Vektor_P>0) || (F_Rez_Sell-comission*Point()>0 && Load_Vektor_P<0))Metka=1;
      else Metka=0;

[Aleksey Vyazmikin]

Uladzimir Izerski:

I always judge the behavior of a forum member by his adequate behavior on the forum, his posts and the level of training.

I do not need someone's secrets. Experience and digging up secrets is for beginner's level.

Why assess "forumers", what is the purpose of the task?

[Valeriy Yastremskiy]

Maxim Dmitrievsky:

Ask Valery, he started to catch up...

I'm hard to think of any other wording for it

probabilistic corridor. time of change of orbit of electron, if it went out of limit then it stayed in place. how to enter a parameter into the mo ..... the problem is clear.

[Maxim Dmitrievsky]

Aleksey Vyazmikin:

When saving the sample, I simply reduced the financial result by a given value, and then, if it was greater than zero, I put 1 for the target. Now I made a separate script for this, so it would be faster to change the target. The shift can be made in different directions.

There are no details in the article.

Hmm... I'll have to think about it. My chips and labels are converted after the tester is done, so you can't run it in the tester after that - it will be nonsense. You can only run a model that has already been trained.

But then I transform dataset and it loses seriality, the model stops accepting spread, and the chips live their own life.

On the one hand, conversion gives cool results. On the other hand, trained model can not defeat spread as an artifact

[Valeriy Yastremskiy]

Aleksey Vyazmikin:

Why give an assessment to "forumers", what is the purpose of this task?

I went to first grade in 73 and he was in the army.... consider ... fathers and children eternal book)))))

[Maxim Dmitrievsky]

Valeriy Yastremskiy:

probabilistic corridor. time of electron's orbital change, if it is out of the limit then it remains in place. how to introduce a parameter into the mo ..... the problem is clear.

it is necessary to work with densities for each group of marks. I see no other options, only hardcore variations.

[Aleksey Vyazmikin]

Maxim Dmitrievsky:

Hmm... I'll have to think about it. I have chips and tags are converted after the tester, after that in the tester is no longer run - it will be nonsense. You can only run the model already trained after that

And what prevents you from storing financial result and type of transaction in a separate array, and then make a markup for it with the necessary corrections and replace the targets? Of course, I don't know the capabilities of python, but in an extreme case, the selection may be saved in a file and converted to MT5 or, at least, to Excel :)))

[Maxim Dmitrievsky]

Aleksey Vyazmikin:

And what prevents you from saving the financial result and type of transaction in a separate array, and then make a markup for it with the necessary corrections and replace the target ones? Of course, I do not know the capabilities of python, but as a last resort, the selection may be saved in a file and converted into MT5 or, at least, into excel :)))

After conversion it's another dataset with other attributes and labels

[Aleksey Vyazmikin]

Maxim Dmitrievsky:

I delete trades that do not cover the spread... but then I convert the dataset and it loses its seriality, the model stops accepting the spread, and the currents take on a life of their own.

On the one hand, conversion gives cool results. On the other hand, a trained model cannot handle spreads as an artifact.

So we need to consider virtual inputs - my basic strategies activate the signal to be checked by the model - that's why there are no such problems.

I have increased the difference significantly - 50 points in the article, while the spread is 1-3 points :)