Machine learning in trading: theory, models, practice and algo-trading - page 2583

 
Replikant_mih #:
What exactly do you do in the models?
 
mytarmailS #:
What are you doing in the models?

Predicting a class, which is defined by an increment in n candles. It used to be greater than/less than zero. Then in training I started balancing classes to have roughly equal objects, so "zero" is shifted from real zero for these purposes. Or wasn't that the question?)

 
Replikant_mih #:

Predict the class, which is defined by the increment in n candles. Earlier it was more/less than zero. Then I started balancing classes to have about equal objects, so "zero" is shifted from real zero for these purposes. Or wasn't that the question?)

Well, yes about that...
And that you can build something working/earning on that?
 
mytarmailS #:
Well, yes about that...
And that it is possible to build something working/earning on it?

It is possible, the main thing is to give good signs). And some other nuances.

 
Replikant_mih #:

It is possible, the main thing is to give good signs). And some other nuances.

(Can you reveal the veil of secrecy or is it a secret? )

 
Very voluminous topic), and there are those who in combat trade uses ML?
 
mytarmailS #:

(Will you reveal the veil of mystery or is it a mystery? )

Well, it's a mystery as well, and somewhere without context it will be incomprehensible, and with context for a long time).

About the signs - I call them trader's signs. On increments it's difficult to build something sensible, unless you have some kind of advanced neural network. And with normal ML without rocketsaensis you need trader traits - such that you could use in normal trade, such as: Oh, here we have XXX, so I enter, well this XXX you can and feed. It's not the fact that the sign is really important, but it's better than increments anyway.

 
Replikant_mih #:

Well, the mystery including, well, somewhere without context it will be incomprehensible, and with the context long).

About signs - I call them trader's signs. It's hard to build something good on increments, unless you have some advanced neural network. And with normal ML without rocketsaensis you need trader traits - such that you could use in normal trade, such as: Oh, here we have XXX, so I enter, well this XXX you can and feed. It's not the fact that the sign is really important, but it's better than increments, anyway.

very vague

 
mytarmailS #:

very fuzzy.

Sort of)).

 

I think it would be worthwhile to better study the issue of customizing the loss function for our trader's needs.

As an example, here isan article on the subject.

Improving the Prediction of Asset Returns With Machine Learning by Using a Custom Loss Function
  • papers.ssrn.com
The loss function in supervised deep learning is a key element for training AI algorithms. For models aiming at predicting asset returns, not all prediction err
Reason: