Discussing the article: "Creating a mean-reversion strategy based on machine learning" - page 2

[[Deleted]]

fxsaber #:

Finished reading to this point and didn't understand where the origin.

Predicts the selected cluster, one out of 10. If it predicts it, it allows trading. Analogous to using some other bevel filter.

[fxsaber]

Сигналы на закрытие работают по обратной логике.

What's this?

input int stoploss = 2000;             //Stop loss
input int takeprofit = 200;            //Take profit

[[Deleted]]

fxsaber #:

What's this?

The bot's inputs, you can use them.

[fxsaber]

Maxim Dmitrievsky #:
Predicts the selected cluster, one out of 10. If it is predicted, it allows trading. Analogous to using some other bevel filter.

That's what I don't get. Why predict a cluster if it is immediately clear which one we are in?

[fxsaber]

Maxim Dmitrievsky #:
Bot's Inputs

So you didn't participate in the training, but just added it for experimentation in MT5?

Maxim Dmitrievsky #:
can be accessed.

It's painful actually, as the code was not written at all for acceptable speed of calculations.

[[Deleted]]

fxsaber #:

I don't get it. Why predict a cluster, if we know at a glance which one we are in?

Model outputs are called predictions

[fxsaber]

Maxim Dmitrievsky #:
Outputs of the model are called predictions

Thanks, glad for the enlightenment.

I read the article. No water at all. Thank you.

[[Deleted]]

fxsaber #:

I.e. you didn't participate in the training, but just added it for experimentation in MT5?

They are not involved in training in any way, but they participate in model selection in the python script. You can change them arbitrarily. This strategy works easiest with longer stops. A separate topic is why and how to fix it. It is related to clustering, when there can be large time gaps between observations.

[[Deleted]]

fxsaber #:

Thank you, glad for the enlightenment.

I read the article. No water at all. Thank you.

Yu ar Welcom :)

[[Deleted]]

fxsaber #:
It's a pain really, since the code wasn't written at all for acceptably fast calculations.

Models take a long time to calculate. I can think of a couple of tricks to get around it, but I'm not up to it yet. For example, write all predictions on history into an array, purely for optimisation.