Machine learning in trading: theory, models, practice and algo-trading - page 2407

New comment
 
mytarmailS:

Discussing which model is better is a kindergarten, I suffered from such bullshit 7 years ago


how to generate new informative features - YES

how to create fitness functions objective to the market - YES

how to transform incoming information - YES

how to build functions adaptive to the market - YES

but discussing what is better NS or Forrest on the same empty data is just FEIS PALM...

the difference in error between all dozens of different MO algorithms is 0.5 - 2%

2% Carl!!! we need to discuss something that gives a gap of 20-30%

If the signs are informative, then any AMO will work well, the opposite is also true!!!!

These are the words of a man who understands nothing of what he has been discussing in this thread for YEARS.

"The difference in error" on a forward test between conventional dumblinear regression and NS can be MORE than 20-30% and not in favor of the network.

And this is easy to check and demonstrate.

That's why networks are the usual fit

 
Dmytryi Nazarchuk:

These are the words of a man who understands nothing of what he has been discussing in this thread for YEARS.

"The difference in error" on a forward test between ordinary dumblinear regression and NS can be MORE than 20-30%.

And this is easy to check and demonstrate.

Demonstrate
 
mytarmailS:
Demo

Blue curve (GBP D1 - GBP LineReg), orange curve(GBP D1 - GBP NN).

Training sample 2009-2020, forward 2021.

The independent variables are the same.


 
Dmytryi Nazarchuk:

Blue curve (GBP D1 - GBP LineReg), orange curve(GBP D1 - GBP NN).

Training sample 2009-2020, forward 2021.

The independent variables are the same.


Please send me the data in the format signs, target, neural outputs, regression outputs, as well as labels, where trace where test
 
mytarmailS:
Please send me the data in the format of signs, target, neuron outputs, regression outputs, as well as labels, where the trace is where the test

)No, I can't reset the signs.

 
Dmytryi Nazarchuk:

)No, I can't reset the signs.

Why?

Forget it, I just wanted to show you that if you have an understanding, you can make a lin. Regression model is no worse or even better than NS...

But it all started out not about that, and that no AMO does not give considerable advantage over another on standard data, the difference of 2% is nothing when you need 20-30%
 
mytarmailS:
Why?

Well, that's my know-how.

 
Dmytryi Nazarchuk:

Well, that's my know-how.

So if you send me the matrix of ready-made signs and ready-made target, it will somehow reveal your kind of know-how of technology?

Dmitry, you're a genius! )) I'd better go read a book.

 
mytarmailS:

So if you throw in a matrix of ready-made traits and a ready-made target, it will somehow reveal your kind of know-how of technology?

Dmitry, you are a genius! )) I'd better go read some books.

Thank you!

 
Sometimes people don't understand the difference between a regression model and a regression model with activation f-ions, hence the curved conclusions. For example, if you don't normalize the data for the second one properly, there will be a gradient explosion and it won't learn anything.
Also, they don't understand that the first model is the simplest neuron without activation f-ions, from which the second model is sostritic.
1...240024012402240324042405240624072408240924102411241224132414...3547
New comment
Reason: