Machine learning in trading: theory, models, practice and algo-trading - page 3319

New comment
 
Andrey Dik #:
If you know, don't say anything. And if you say you know, show me. Otherwise it's not nice: "I know but I won't tell you, and you're all fools!".

Are you able to absorb new information? I showed you, read the link.

 
Andrey Dik #:
What is the stopping criterion?

any

 
This is an external (custom) criterion by which the NS is not optimised (trained)
 
Maxim Dmitrievsky #:

Are you able to absorb new information? I showed you, read the link.

Maxim Dmitrievsky #:

any

You haven't given an answer. When do you need to stop the training?
 
Andrey Dik #:
You haven't given an answer. When is it necessary to stop learning?

by any custom criteria, I answered the question.

 
Maxim Dmitrievsky #:

by any custom criteria, I answered the question.

What, any? Name two very specific criteria.
 
Andrey Dik #:
By what, any? Name two very specific criteria.

Accuracy, ROC/AUC, R^2.

 
Maxim Dmitrievsky #:
This is an external (custom) criterion by which the NS is not optimised (trained)

So it is possible to train several different neural networks, and each according to its own criterion?

If I understood correctly and it is so, this is suitable for me.

I have to do it

 
Renat Akhtyamov #:

So you can train several different neural networks, and each on a different criterion?

If I understand it correctly and that's the case, that works for me.

I'm going to have to look into it.

You can stop training by any criterion, but they are usually trained by logloss.

depending on the criterion you will get different models with different properties
 
Maxim Dmitrievsky #:

Accuracy, ROC/AUC, R^2

Well, can you give graphs of these criteria during training and show where exactly on this graph you need to stop training?

You don't need to throw links, you can draw by hand and show where exactly.

1...331233133314331533163317331833193320332133223323332433253326...3700
New comment