Machine learning in trading: theory, models, practice and algo-trading - page 2961

[Forester]

mytarmailS #:

Well, then there's nothing to ask, nothing to do but throw it away.

Just in case, an example of how to replace NAVs, since I've already written an example.

and the solution

Thanks, maybe the code will be useful to someone.

[Rashid Umarov]

The onnx help on the website has been updated - https://www.mql5.com/ru/docs/onnx

[СанСаныч Фоменко]

mytarmailS #:

I've been thinking about your example.

Big doubts.

Firstly, am I understanding it correctly.

On some section of the quote, entry points have been found that will give some kind of perfect balance line.

If so, this is an over fitting on history. The entry/exit points found do not at all satisfy the basic MO idea of "history repeats itself". With MO, one looks for some abstract patterns, with the hope/justification that they will repeat in the future. And here is a markup of some price area....

Is there any other way? Or am I missing something?

[mytarmailS]

СанСаныч Фоменко #:

Thought about your example.

Big doubts.

Firstly, if I've got it right.

On some section of the quotir, entry points have been found that will give a certain ideal balance line.

If so, this is an over fitting on history. The entry/exit points found do not at all satisfy the basic MO idea of "history repeats itself". With MO, one looks for some abstract patterns, with the hope/justification that they will repeat in the future. And here is a markup of a certain price area....

Is there any other way? Or am I missing something?

The point/purpose of the example is to show that it is possible to train the model not only with ready-made targets, but also with loss functions of any complexity minimising or maximising the FF.

In this example (by request of those who are eager) it is shown how to train AMO for maximum stable profit, but it can be absolutely any FF to your liking....

Also nobody prevents to add test and validation sampling for training so that there would be no overtraining, but it would complicate the code and is beyond the scope of the example.

[СанСаныч Фоменко]

mytarmailS #:
The point/purpose of the example is to show that it is possible to train the model not only on ready-made targets, but also on loss functions of any complexity by minimising or maximising the FF.

In this example (at the request of those who are interested) it is shown how to train AMO for maximum stable profit, but it can be absolutely any FF to your liking....

Also nobody prevents to add test and validation sampling for training so that there would be no overtraining, but it would complicate the code and is beyond the scope of the example.

I see, very curious

[Andrey Dik]

СанСаныч Фоменко #:

I see, very curious

what's curious? it was said a couple of months ago in this thread in dialogues with my participation))) here many argued that max/min ff should not be in any way)))))

as you set the ff so the ship will sail....

[Maxim Dmitrievsky]

Andrey Dik #:

what's curious? so it was said a couple of months ago in this thread in dialogues with my participation)) here many argued that max/min ff should not be in any way))))

as you set the ff, so the ship will sail....

The algorithm has its own ff, which can not be changed (will not work), it is just an add-on for curve fitting to be beautiful. It doesn't affect anything globally. There was already here a variant with custom loss on the profit factor, on the trayne is beautiful as usual.

We go round and round and are surprised every time. Amnesia is a pleasant affliction, every day news 😀

[Andrey Dik]

Maxim Dmitrievsky #:
The algorithm has its own ff, which cannot be changed (will not work), it is just an add-on for curve fitting, to make it beautiful). It does not affect anything globally.

Max, you can set any FF, and it is good to set it according to the training goal.

If the learning goal is curve fitting, then it will be curve fitting)).

But it does not cancel the fact that any training is the essence of optimisation (max/minimisation) of some FF.

[Maxim Dmitrievsky]

Andrey Dik #:

Max, FF can be set any way you want, and it's a good idea to set an appropriate learning objective.

If the learning objective is kurwafing, then it will be kurwafing)).

but this does not cancel the fact that any training is the essence of optimisation (max/minimisation) of some FF.

But I can't imagine how TC can be pulled out through this :) maybe someone has a super-duper FF, but he keeps silent

[mytarmailS]

Maxim Dmitrievsky #:
It doesn't affect anything globally.

The main thing is to believe in it and repeat it more often, or vice versa.