Machine learning in trading: theory, models, practice and algo-trading - page 2840

 
Andrey Dik #:
It's the criterion that's at fault.
You mean the problem statement
 
Learning with a teacher implies that you want to get a specific result and test your conclusions on new data. If you shove custom criteria in there, you will check them. Then you can not prepare labels at all, randomly do.

That's it, I'm off to play Metro exodus. It is more relevant than ever.
 
Maxim Dmitrievsky #:
I can't imagine it in my head... we have a labelled dataset, we want to train as close to these labels as possible. If we take another criterion not related to them, then these labels stop mattering?

Then the learning process completely changes the strategy. We get a customised fit to a custom criterion.

That's the point, we are moving a bit away from the kind of optimisation (learning) that is common in classical MO. We are moving towards optimisation in a broader sense (as in MT5, for example). But at the same time we want to preserve the power and flexibility of the models used in MO.

I have always been confused by the conceptual gap between MT5 optimisation and MO application. It would be nice to have options for intermediate approaches.

 
Andrey Dik #:


It is as if Fomenko does not hear what is being said. I have already said several times that the tester does not affect profitability or the ability of the TS to work profitably in the future. The tester is a tool, nothing more. An optimisation algorithm is a tool and nothing more. It is like discussing the "success" of a shovel for making money.

That's right, a conversation between the deaf and the blind.

I write that optimisation together with criteria is not necessary because financial markets are NOT stationary, and you write that I understand something about optimisation.


Success, alchemists have been converting everything into gold for several hundred years.

 
Aleksey Nikolayev #:

The point is that we are moving a bit away from the type of optimisation (training) that is accepted in classical MO. We are moving towards optimisation in a broader sense (as in MT5, for example). But at the same time we want to preserve the power and flexibility of the models used in MO.

I have always been confused by the conceptual gap between MT5 optimisation and MO application. It would be good to have possibilities for intermediate approaches.

Well, in this form it is possible
 
mytarmailS #:
So the mission statement
yes
 
Andrey Dik #:
The more adequate the evaluation criterion, the more adequate the model behaves on new data. choosing the best AO means choosing the best tool to optimise the CRITERION. It cannot be the AO's fault or the tester's fault. The criterion is at fault.

The robustness of the TS has NOTHING to do with the evaluation criteria, because the criterion is exactly one - guessed the direction of the trade or not. But the latter depends on the set and properties of predictors

 
СанСаныч Фоменко #:

That's right, a deaf man talking to a blind man.

I write that optimisation together with criteria is not necessary, because financial markets are NOT stationary, and you write that I understand something about optimisation.

You are also, in fact, doing optimisation. You have invented some criterion of "stationarity of signs" and take the signs that are optimal according to it. It's the same optimisation in history, but in profile.

SanSanych Fomenko #:

The robustness of the TS has NOTHING to do with the evaluation criteria, because the criterion is exactly one - guessed the direction of the trade or not. But the latter depends on the set and properties of predictors

Here, it is absolutely necessary to invent a criterion of TS robustness and optimise according to it) Again we will get the sameoptimisation on history, but in a different profile).
 
Aleksey Nikolayev #:
Here, it is absolutely necessary to invent a criterion of TS robustness and optimise according to it) Again we will get the sameoptimisation on history, but in a different profile).

There you go. I don't understand the allergy of some comrades to the word "optimisation".

Optimisation should be considered as a process of finding the best solution. the best solution of a robust model. If the model is not robust (weak evaluation criterion), then, as they say, "don't blame the mirror" (blame the optimisation).

 
Andrey Dik #:

There you go. I don't understand the allergy of some comrades to the word "optimisation".

Optimisation should be considered as a process of finding the best solution. the best solution of a robust model.

Not an exact definition, and if the search process is not in the model, then it is not optimisation? )

I, for example, create code using optimisation tools...

Optimisation is "mathematical search of unobservable parameters according to a chosen criterion of usefulness".

something like this