Machine learning in trading: theory, models, practice and algo-trading - page 3230

[Andrey Dik]

Renat Fatkhullin #:

Scalpers will be banned by rule.

The goal is clearly stated - to stimulate the development of ML models for trading, and not to give an opportunity to make money in the old way, pour banal scalpers under the guise of models and so on.

I didn't mean scalpers (it's probably when the target is 5 pips and without SL, although the term "scalpers" can mean anything and can even be used as a swear word).

It is also possible to set in the rules a restriction on the average profit / average loss ratio, a restriction on the maximum number of trades per day and the minimum number of trades per day, etc., in order to weed out obviously toxic methods that work on the principle of "just in case".

By the way, past championships have shown that simple strategies can win in the short term.

Maybe a simple strategy on a single neuron will work well, you don't have to add ChatGPT to compete....

I'm just interested in the limits of strategies that are acceptable for the championship.

[fxsaber]

Renat Fatkhullin #:

The goal is not to give another scalper easy money.

That's the point, your ideas about scalpers are still there, N years ago. Back then, indeed, the inefficiencies were just wild. Things have changed somewhat in the meantime. Whatever.

[Aleksey Vyazmikin]

Renat Fatkhullin #:

Please don't ignore the substantive question. What will be the predictors in the template? Or does the model on OHLCV only need to be done?

[Maxim Dmitrievsky]

fxsaber #:
That's the thing, your ideas about scalpers are still there, N years ago. Back then, indeed, the inefficiencies were just wild. Things have changed somewhat in the meantime. Whatever.

You have to do it through ONNX anyway, you don't have it like that

At the same time it will be possible to see the results of real MOShnikovs (those who use neural networks), not regulars of this topic. Which is very important from an educational point of view. I, for example, lack role models and communication with practitioners.

[Andrey Dik]

fxsaber #:
That's the thing, your ideas about scalpers are still there, N years ago. Back then, indeed, the inefficiencies were just wild. Things have changed somewhat in the meantime. Whatever.

I think it could be an anti-MO advert if simple methods (or non-simple, but without MO) win. although I believe very much in the prospect of MO.

ZY*. if, for example, adaptive systems with auto-optimisation but without neural networks in pure form will participate, isn't it interesting? - in this context, isn't MO a selection of parameters (weights, coefficients) of a model? it would be all the more interesting to see "live" systems that are trained on the fly, not just trained on history (some of them will be lucky enough to win prizes, of course, being formally "serious machine learning models").

ZZY**. "the evening stops being dark" - very interesting what will come out of this, after all you do care, fxsaber, don't you? - and so do I)))))

* and ** - thoughts aloud.

[Renat Fatkhullin]

Aleksey Vyazmikin #:
Please don't ignore the substantive question. What will be the predictors in the template? Or is the model on OHLCV only to be made?

Essentially everything will be as soon as we work out the conditions of the competition - it is a huge labour.

[mytarmailS]

Renat Fatkhullin #:

Can you please tell me if the R port is done?

I mean, we don't have to wait.

[Renat Fatkhullin]

mytarmailS #:

Can you please tell me if the R port is done?

I mean, we don't have to wait.

Not yet.

[Forester]

mytarmailS #:

Can you please tell me if the R port is done?

I mean, we don't have to wait.

Why do you need it? When I was experimenting with R, I used the dll-ku ordered by SanSanych. It works fine and has more possibilities, as it transmits any data both ways, not only quotes (as it was in the first R port). Plus I even managed to run multithreaded optimisation through it.

[mytarmailS]

Forester #:
Why do you need it? When I was experimenting with R, I used the dll-ku ordered by SanSanych. It works fine and has more possibilities, because it transfers any data both ways, not only quotes (as it was in the first R port). Plus I even managed to run multithreaded optimisation through it.

Official support as it seems to me can give a good push /hype for both R and MKL ....

New articles on all countries, new ideas, new people to talk to, new algorithms....