Machine learning in trading: theory, models, practice and algo-trading - page 2760

[JeeyCi]

СанСаныч Фоменко #:

Less aplomb and more learning from other people. and you'll be digitally happy.

clearly not from you to learn, as your aplomb here - also contradicting the common sense logic of studying and choosing predictors (from Aleksey Nikolayev )...

start following your own advice

p.s. I left explanations in the p.s. of the previous post.... after your ma-keys on this issue any substantive dialogue is impossible in principle ... if you see insults in all your arguments, you should have dialogues with other specialists....

[Aleksei Kuznetsov]

СанСаныч Фоменко #:

That's the problem!

That is why I write that years have been spent on this thread mostly on nonsense. If you manage to find predictors, you will have a trading system. If you fail, it is not your destiny.

PS.

As a teacher, take the price increment, or rather a sign. VLADIMIR PERERVENKO has given very good predictors in his articles.

I tried it according to his articles. In the article a successful period/instrument - the balance line is still more or less decent. On other periods the balance line approaches the horizontal line (on new data).
So I try other target periods. But on them the balance is also horizontal at best.

[СанСаныч Фоменко]

elibrarius #:

Tried it on his articles. In a successful period/instrument - the balance line is still more or less decent. On other periods the balance line approaches the horizontal line (on new data).
So I try other target periods. But on them the balance is also horizontal at best.

What does balance have to do with it?

First we need to achieve a prediction error of less than 30%, and then balance.

[[Deleted]]

This is all very interesting, but you have a non-stationary source series on which trades are opened and a stationary sign.

The sign sometimes repeats market movements, sometimes not. If you train on those moments when it repeats, there will be high correlation and mutual information between the sign values and the target ones. Because the target is in the future relative to the trait, i.e. it predicts it. This is what I put in the concept of informativeness and modell agnostic ficha selekshn . And you can't think of anything else for forex.

[Aleksei Kuznetsov]

СанСаныч Фоменко #:

What's balance got to do with it?

First you need to achieve a prediction error of less than 30%, and then balance.

Balance is the main goal.
According to the articles, it turns out to be less than 30%. But it won't earn money, because the balance is not drained at best.

[[Deleted]]

And all these diagrams are just a consequence of a random selection of signs, without any understanding

Predicting the sign of increments is the worst case scenario, no sense at all is made of it

Random labelling works even better.

[СанСаныч Фоменко]

Maxim Dmitrievsky #:
And all these diagrams are just a consequence of randomly picking traits, without understanding

Predicting the sign of the increments is the worst case scenario, no sense is made at all

Random markup works even brighter.

1. Where did you see a randomisation of features? There's not even a whiff of that in there.

2. There is no Buy and Sell and nothing else in the terminal.

3. Interesting to see. What is random markup? Random what?

[СанСаныч Фоменко]

elibrarius #:
Balance is the main goal.
According to the articles, it turns out to be less than 30%. But it will not give you a chance to earn, as the balance is not drained at best.

Don't confuse God's gift with eggs.

Do you have a model with prediction error less than 30%? Can you show it?

PS. Developing an Expert Advisor using signals from MO is a separate problem. Let's leave it aside for now.

[[Deleted]]

СанСаныч Фоменко #:

1. Where did you see a random selection of traits? It doesn't even smell that way.

2. There is no Buy and Sell and nothing else in the terminal.

3. Interesting to see. What is random markup? Random what?

I didn't see any meaningful selection of signs in those articles. Meaningful in the sense of not choosing from the pile, but making informative markup chip-target at once. Any feature can be matched with a target feature. This is impossible with increments. You will have to match the target features to the target ones.

Random is when deals are seeded from scratch in all directions with different durations and dreamed depending on the result

[СанСаныч Фоменко]

Maxim Dmitrievsky #:
I didn't see any meaningful feature selection in those articles. Meaningful in the sense of not choosing from a pile, but making an informative markup chip-target at once. It is possible to select target features for any feature. This is impossible with increments. We will have to select target attributes under target attributes.

Random is when deals are seeded from scratch in all directions with different durations and dreamed depending on the result

"Meaningful" - this is according to the pictures I have given, which make"informative markup chip-target at once".

And what do you mean by "meaningful"?