Machine learning in trading: theory, models, practice and algo-trading - page 3726

[[Deleted]]

Rorschach #:
So masterforex wrote that everything is managed by a single quotation machine. It's a joke, but some MM algorithms resemble something like that, like there are some constants on time and price range. They've been trying to write decoders for a few years now, but so far without success. Past winners of the LCHI and participants of the Robinson Cup, who realised who it is well done

Anyway, engaged in very useful activity of supporting weak-minded theories and selling courses :)

[fxsaber]

Forum on trading, automated trading systems and testing trading strategies

Machine Learning in Trading: Theory, Models, Practice and Algorithm Trading

Maxim Dmitrievsky, 2025.11.08 06:01

As a consequence, it is necessary to put into the algorithms an overshoot of instruments.

I wonder if anyone does this.

[[Deleted]]

fxsaber #:

I wonder if anyone does this.

Probably did it for interest for a simple TC. Picked 2-3 pairs.

[[Deleted]]

In DC Saber the logic works (low spread) and on low TFs (here M15). It needs to be optimised a bit for combat conditions, but it's fun because there is almost 0 retraining.

Basic and meta model errors:

>>>  options[-1][1].get_best_score()['validation']
{'Logloss': 0.004198935695863803, 'F1': 0.9987859166329423}

>>>  options[-1][2].get_best_score()['validation']
{'Logloss': 0.015636819123292726, 'F1': 0.9590488771466316}

[[Deleted]]

5-minutes at fix spread 10 (bottom) and at real spread (top). If there is a possibility to fix it with limiters - it would work well, as on the bottom one.

[Aleksei Kuznetsov]

Maxim Dmitrievsky #:

5-minutes at fix spread 10 (bottom) and at real spread (top). If there is a possibility to fix it with limiters - it would work well, as on the bottom one.

What instrument are you testing?

[[Deleted]]

Forester #:
What instrument are you testing?

AUDCAD

c 22 to 25 something with spread, because at fix spread there is no chatter there.

[E38]

Maxim Dmitrievsky #:
One graal flat currency pair, on which almost no difference for what period to do training.

What pair is that? I can't see it on the screenshots.

[[Deleted]]

E38 #:
What's the pair? I can't make it out in the screenshots.

The screenshots are Australocanadian.

[[Deleted]]

Maxim Dmitrievsky #:

Errors of the core and meta models:

The trick of markup is that it is not looking ahead, i.e. without looking ahead. For example, signals using any oscillator (overbought/oversold). Then generalisation.

Therefore, the errors of the models are small, because they approximate the indicator in essence. But it is not possible to find such a magic oscillator for all pairs.