Machine learning in trading: theory, models, practice and algo-trading - page 920

 
Mihail Marchukajtes:

And don't let such big investors hang you by your cokeys..... :-)

I've been wearing armor for a long time, just in case

 
Is there any way to use clustering for our purposes? I tried to divide the sample into 10 clusters and it turned out to be bullshit - 0 correct solutions, or should I make 100-1000 clusters?
 
Aleksey Vyazmikin:
Is it possible to use clustering for our purposes? I tried to divide a sample into 10 clusters and it turned out to be bullshit - 0 correct solutions, or should I make 100-1000 of these clusters?
Yes, it's bullshit.
 

There is an option to open my RL class as a shared project for those who wish, if anyone has an unbridled desire to do something more with it

possibilities at the moment:

  • creating an ensemble of any number of agents
  • each actor has its own characteristics (number of trees, r parameter)
  • different predictors can be assigned to each agent (any arbitrary number).
  • dropping agents after training
  • training of selected agents for dropping, not for earning
  • easy addition and removal of agents in a few lines
  • Encouragement f-thing is only one for now, but you can easily add different variants to choose from, for each agent in the ensemble (the most creative thing to do)

Or an article with a detailed description?

Or a stomping ground

 
Maxim Dmitrievsky:

There is an option to open my RL class as a shared project for those who wish, if anyone has an unbridled desire to do something more with it

possibilities at the moment:

  • creating an ensemble of any number of agents
  • each actor has its own characteristics (number of trees, r parameter)
  • different predictors can be assigned to each agent (any arbitrary number).
  • dropping agents after training
  • training of selected agents for dropping, not for earning
  • easy addition and removal of agents in a few lines
  • encouragement f-thing is only one for now, but you can easily add different options to choose from, for each agent in the ensemble (the most creative thing to do)
Or an article with a detailed description?

In any format, the information will be useful, the main thing that the material was set out in accessible language!

 
Aleksey Vyazmikin:

In any format the information will be useful, the main thing that the material was presented in accessible language!

It's a simple class and a few methods, I don't know if it's any simpler

//+------------------------------------------------------------------+
//|Multiple RL agents class                                          |
//+------------------------------------------------------------------+
class CRLAgents
  {
private:
   struct Agents
     {
      double            inpVector[];
      CRLAgent         *ag;
     };
   double            dropoutAgent[];
   bool              dropout;
public:
                     CRLAgents(int,int,int,double);
                    ~CRLAgents(void);
   Agents            agent[];
   void              updatePolicies(double);
   void              updateRewards();
   double            getTradeSignal();
   void              setAgentSettings(int,int,int,double,bool);
   void              dropoutAgents(int&[]);

  };

I can not give lessons on the alglib, I'm not a teacher

 
Maxim Dmitrievsky:

There is an option to open my RL class as a shared project for those who wish, if anyone has an unbridled desire to do something more with it

possibilities at the moment:

  • creating an ensemble of any number of agents
  • each actor has its own characteristics (number of trees, r parameter)
  • different predictors can be assigned to each agent (any arbitrary number).
  • dropping agents after training
  • training of selected agents for dropping, not for earning
  • easy addition and removal of agents in a few lines
  • Encouragement f-thing is only one for now, but you can easily add different variants to choose from, for each agent in the ensemble (the most creative thing to do)

Or an article with a detailed description?

Or a stomping ground

Article with sources. If the public repository will be stored forever, just a link to it.
If you choose to stomp on it, then send the project first in your personal)
 
Maxim Dmitrievsky:

There's a simple class and a few methods, I don't know if it's simpler.

I can not give lessons on the alglib, I'm not a teacher

It would be nice to see a comment for each line - what to write there - you can only guess from the name. Why guess?

 
Aleksey Vyazmikin:

It would be nice to see a commentary on each line - what to write there - you can only guess from the title. Why guess?

I will do it for 50 rubles.)

 
Maxim Dmitrievsky:

I will do it for 50k :)

That's how it turns out, a lot of articles are written, but the code there is often unreadable for novice programmers... Why do you think people wrote them? Who would understand them or live without them?