How do you practically assess the contribution of a "specific" input to the NS? - page 6

New comment
 
TheXpert:
There is a logistical "regression"
What difference does it make? If it doesn't replicate the work of the network how can you draw any conclusions? And if it does, how can it be solved?)
 
faa1947: Nothing to do with the NS
Then what will your tests have to do with it?
 
Figar0:

Anyway, I've got it all figured out. I don't need any more answers.


That's too bad. None of the above applies to me.
 
Mathemat:
Then what will your tests have to do with it?
None. Take any TS. The variables are the numbers of entries. The left side of the regression is the profit and loss of TS. we need the statistics, at least 30 profits. We estimate. At the level of idea.
 

This is as arbitrary a construction as all econometric tests.

Делаем регрессию:

Profit = s(1) * A0 + . c(n) * A(n)

We estimate the coefficients of this regression.

We obtain at once

probability that a particular coefficient is equal to zero. - we delete such an input

the probability that all the coefficients taken together are equal to zero

by ellipses we obtain correlation coefficients

perform the test for redundant inputs

Test for skipped inputs

carry out the test of stability of the coefficient values (evaluate the randomness of their change)

Why do you think these particular tests are enough?

 
Figar0:

I have the NS doing the classifications.

What are the classes, if that's not a secret?
 
Mathemat:

Why do you think that these particular tests are enough?


This is as arbitrary a construction as all econometric tests.

Did you manage to pass the test, or are you judging again without reading?

Why do you think these particular tests are enough?

Necessity, sufficiency - what lofty matters.

To begin with I would like to get at least some characteristics of the inputs. Maybe all the coefficients will be equal to each other? I had an Expert Advisor with 6 inputs but they were intricately correlated in some way, and I refused from Expert Advisors with many inputs.

By the way, if the inputs have a set of parameters, it is even more interesting.

 
TheXpert:
Which classes, if you don't mind me asking?
Why the kick in the pants? I look forward to hearing from you.
 
TheXpert:
Which classes, if it's not a secret?


Growing - Falling, and Don't know.

-1; 0; 1 :)

 

Maybe someone can explain to me.

We take the NS and draw samples of the letter "a" in different handwriting. We teach them to recognise the letter "a". And in kotir - what do we teach? After all, we need to highlight what we teach "head and shoulders" or something.

I've suggested to one of the coryphaei of NS to feed the NS input ZZ and teach it to recognize reversals. ZZ patterns can be bred a lot. But the coryphaeus has gone into the mire, not even an eye.

1234567891011121314
New comment