Machine learning in trading: theory, models, practice and algo-trading - page 3377

New comment
 
fxsaber #:

The nature of the profit curve does not change by OOS: Size(OOS_Left) = Size(OOS_Right) = Size(Sample). All in all, a result you can't go past.

Well through re-optimisation with a check on OOS can be found :) which is the simplest case of wolf forward OR cross-validation with one fold.
 
Maxim Dmitrievsky #:
Well through re-optimisation with OOS checking can be found :)

Please disclose in a few sentences.

 
fxsaber #:

Please disclose in a few sentences.

Everyone seems to know about wolf-forward. When optimised for sample, the results are taken from oos. The best overall result with averaged parameters is taken so that the curves don't differ.
 
It won't prove that it's not a fit. It will give the most robust parameters. It is impossible to prove that it is not a global fit at all. And especially not through any global minima.
 
Maxim Dmitrievsky #:
I think everyone knows about wolf-forward. When optimised for sample, the results are taken from oos. The best overall result with averaged parameters is taken so that curves do not differ.

Let's say 100 steps are taken - we get 100 sets of input. If we form the average set according to the principle "each input set is equal to the average of the corresponding input 100 sets", it is unlikely that this set will pass well the whole initial interval.

 
fxsaber #:

Let's assume that 100 steps have been taken - 100 input sets have been obtained. If we form an average set according to the principle "each input set is equal to the average of the corresponding input 100 sets", it is unlikely that this set will pass well through the whole initial interval.

If it doesn't, then there are no good sets at all, logically. In terms of confidence in the future.

That's the great and merciless matstat.
 
Maxim Dmitrievsky #:
If it doesn't, there are no good sets at all, logically.

Not logical! Sets depend on FF, for example.

 
fxsaber #:

Illogical! Sets depend on the FF, for example.

You think a few times, then formulate the question taking into account that what you are doing is known only to you. And if you want to be helped you need to be as clear and understandable in your expressions as possible, as well as give as much information as possible....
 
mytarmailS #:
I wonder, does anyone read this endless stream of articles called "neural networks are easy"?
It seems to me that if you calculate the average reading time of this nonsense, it won't exceed 10-15 seconds.


Psst. The guy's just trying to make money to buy a car.

I scroll down to the balance and realise that the man doesn't know the following saying:

"it's not enough to see here, here you have to look, here you have to think ... "

he studies the material superficially and his conclusions are not quite competent.
 
fxsaber #:

Illogical! Sets depend on FF, for example.

FF is nothing at all, nothing depends on it. It just matches inputs with desired outputs. I don't see what this has to do with robustness.

I forgot that in the optimiser you can't set transactions as you like, probably because of that there is a lot of fuss about FF and giving it some special look or sacred meaning :).
1...337033713372337333743375337633773378337933803381338233833384...3574
New comment