Machine learning in trading: theory, models, practice and algo-trading - page 292

 

I started to look at Vorontsov's lectures on machine learning on Yandex and was stunned at the level, unfortunately it is not for me and not for traders in general, machine learning is contrary to all canons of building profitable strategies in forex, there are more than enough parameters, it is reading the sweat.

I think this is a good way to build more profitable forex strategies than all these mathematical brainiacs with the sweeping.


 
Vasily Perepelkin:

I started to look at Vorontsov's lectures on machine learning on Yandex and was stunned at the level, unfortunately it is not for me and not for traders in general, machine learning contradicts all the canons of building profitable strategies in forex, there are more than enough parameters, it is reading the sweat.

I think this is a good way to build more profitable forex strategies than all these mathematical brainiacs with the sweeping.



Well well well.... of course. If you can not understand the question, it does not mean that it does not work. To be honest with you, in my models the number of inputs rarely reaches 8, mostly from 4-6 and it is enough to make the model work not long, but effectively. So ..... I like AI in trading very much :-)

 
Mihail Marchukajtes:


Well, well, well.... Sure. If you can not understand the question, it does not mean that it does not work. To be honest with you, in my models the number of inputs rarely reaches 8, mostly from 4-6 and it is enough to make the model work not long, but effectively. So ..... I like AI in trading very much :-)


I don't think that anyone "understands" the question here, you need to be a scientist, people thoughtlessly use library functions and that's all. You like AI in trading purely psychologically, it is as if you join the elite, you imagine yourself a billionaire and are able to hire a whole university of scientists to develop trading strategies for you like in the West. For example, I am a manager at the racetrack now, it's been almost a year, I used to be a groom taking care of horses and I have nothing to hide, I tell it like it is, you and I are not scientists but practitioners, we need to see and feel what we work with, no need to meddle in another's monastery, where abstraction and vanity rule.
 
Vasily Perepelkin:

No one "understands the question" here, it is necessary to be a scientist, people thoughtlessly use library functions and that's all. You like AI in trading purely psychologically, it is as if you join the chosen ones, you imagine yourself as a billionaire and are able to hire a whole university of scientists to develop trading strategies for you like in the West. For example I am a manager at a racetrack now, it's been almost a year, I used to be a groom taking care of horses and I have nothing to hide, I tell it like it is, you and I are not scientists but practitioners, we need to see and feel what we work with, no need to pry into another's monastery, where abstraction and vanity rule.

I like that the AI tells me the truth and falsity of my TS signals. It does what I purely physically can not do, which is to summarize the information and give an answer, not always correctly, but in general satisfactory. This is a very good clue in my trading, but to say that the AI doesn't work, or that you need to be a scientist, excuse me. I have two degrees in engineering and economics, I did my post-graduate studies, I was a scientist. I do not consider myself a scientist at all, but I have enough brains to understand how AI works, all the more I am not a developer of AI, I am its user. The ability to correctly train the network and choose the right model, this is an art that comes with the years. And you think that just sat down and trained and everything you got right away. That does not happen. To understand the essence of AI you need to train it a thousand times, and go through a thousand models just to get into it. And I don't think anyone will be able to do it so brashly, you need experience here....
 
Mihail Marchukajtes:

I like the fact that the AI prompts me about the truth and falsity of my TS signals. It does what I just physically cannot do, namely it summarizes the information and gives me an answer, not always correctly, but in general it is satisfactory. This is a very good clue in my trading, but to say that the AI doesn't work, or that you need to be a scientist, excuse me. I have two degrees in engineering and economics, I did my post-graduate studies, I was a scientist. I do not consider myself a scientist at all, but I have enough brains to understand how AI works, all the more I am not a developer of AI, I am its user. The ability to correctly train the network and choose the right model, this is an art that comes with the years. And you think that just sat down and trained and everything you got right away. That does not happen. To understand the essence of AI you need to train it a thousand times, and go through a thousand models just to get into it. And I don't think anyone can do it just like that, it takes experience....

So I am a scientist after all, and I have not graduated from any institute, however everything in the machine learning contradicts the common sense in forex trading. The model should be as simple as possible, not more complex.
 
Vasily Perepelkin:

No one "understands the question" here, it's necessary to be a scientist, people thoughtlessly use library functions and that's all. You like AI in trading purely psychologically, it is as if you join the elite, you imagine yourself a billionaire and are able to hire a whole university of scientists to develop trading strategies for you like in the West. For example, I'm the manager at the racetrack for almost a year now, I used to be a groom taking care of horses and I have nothing to hide, I tell it like it is, we are not scientists but practitioners, we need to see and feel what we work with, no need to meddle in someone else's monastery where abstraction and vanity rule.


Well, here are the employees of the racetrack already give an assessment of everything. Doesn't understand anything yet, but gives out evaluations.

Why don't you study, get up to the level of the participants, and then modestly enter the discussion?

 
Vasily Perepelkin:

I see, so a scientist after all, and I did not go to college, nevertheless everything in machine learning contradicts the common sense in forex trading. The model should be as simple as possible, not more complex.


Well, for example, this model is working for me now. Do you think it is complicated????

double getBinaryClassificator2(double v0, double v1, double v2, double v3, double v4, double v5) {
   double x0 = 2.0 * (v0 + 2576.0) / 3282.0 - 1.0;
   double x1 = 2.0 * (v1 + 339.0) / 1599.0 - 1.0;
   double x2 = 2.0 * (v2 + 324.0) / 1266.0 - 1.0;
   double x3 = 2.0 * (v3 + 320.0) / 5863.0 - 1.0;
   double x4 = 2.0 * (v4 + 3928.0) / 25542.0 - 1.0;
   double x5 = 2.0 * (v5 + 30394.0) / 61071.0 - 1.0;
   double decision = 0.9739394040899727 * sigmoid(x0 + x4)
  -0.784271525951731 * sigmoid(x1 + x2 + x3 + x4)
  -1.2984574672986418 * sigmoid(x5)
  + 1.6301941145929288 * sigmoid(x0 + x4 + x5)
  -0.784271525951731 * sigmoid(x1 + x3 + x4 + x5)
  + 0.6917195638702329 * sigmoid(1.0 + x1 + x2)
  + 0.1320196961830271 * sigmoid(1.0 + x3)
  + 2.585886077009396 * sigmoid(1.0 + x0 + x1 + x3)
  + 2.3569514462290853 * sigmoid(1.0 + x0 + x1 + x2 + x5)
  -0.3401611201366674 * sigmoid(1.0 + x3 + x5)
  -2.8359572579429546 * sigmoid(1.0 + x0 + x1 + x4 + x5)
  -0.2533120069673573 * sigmoid(1.0 + x2 + x4 + x5)
  + 0.12667693905237243 * sigmoid(1.0 + x0 + x1 + x2 + x4 + x5)
  -1.2009563367779323 * sigmoid(1.0 + x0 + x3 + x4 + x5);
   return decision;
}


And you're absolutely right, no matter how many times I've tried to build such complicated models, the one that is simpler always works well. Amazing but true....

 
Vladimir Perervenko:


Well, here are the employees of the racetrack already give marks to everyone. He does not understand anything yet, but gives out estimates.

Why don't you study, get up to the level of the participants, and then modestly enter the discussion?

We are not creating the Hadron Collider here, you do not have to be a mathematician to trade.
 
Mihail Marchukajtes:


Well, for example, this model works for me now. Do you think it's complicated????

double getBinaryClassificator2(double v0, double v1, double v2, double v3, double v4, double v5) {
   double x0 = 2.0 * (v0 + 2576.0) / 3282.0 - 1.0;
   double x1 = 2.0 * (v1 + 339.0) / 1599.0 - 1.0;
   double x2 = 2.0 * (v2 + 324.0) / 1266.0 - 1.0;
   double x3 = 2.0 * (v3 + 320.0) / 5863.0 - 1.0;
   double x4 = 2.0 * (v4 + 3928.0) / 25542.0 - 1.0;
   double x5 = 2.0 * (v5 + 30394.0) / 61071.0 - 1.0;
   double decision = 0.9739394040899727 * sigmoid(x0 + x4)
  -0.784271525951731 * sigmoid(x1 + x2 + x3 + x4)
  -1.2984574672986418 * sigmoid(x5)
  + 1.6301941145929288 * sigmoid(x0 + x4 + x5)
  -0.784271525951731 * sigmoid(x1 + x3 + x4 + x5)
  + 0.6917195638702329 * sigmoid(1.0 + x1 + x2)
  + 0.1320196961830271 * sigmoid(1.0 + x3)
  + 2.585886077009396 * sigmoid(1.0 + x0 + x1 + x3)
  + 2.3569514462290853 * sigmoid(1.0 + x0 + x1 + x2 + x5)
  -0.3401611201366674 * sigmoid(1.0 + x3 + x5)
  -2.8359572579429546 * sigmoid(1.0 + x0 + x1 + x4 + x5)
  -0.2533120069673573 * sigmoid(1.0 + x2 + x4 + x5)
  + 0.12667693905237243 * sigmoid(1.0 + x0 + x1 + x2 + x4 + x5)
  -1.2009563367779323 * sigmoid(1.0 + x0 + x3 + x4 + x5);
   return decision;
}


And you're absolutely right, no matter how many times I try to build such complex models, the one that is simpler always works well. Amazing but true....


I don't need it in forex, moreover there are scientists, for example Djuric, who wrote the best mash-ups, no one can beat him.