Machine learning in trading: theory, models, practice and algo-trading - page 3094

[mytarmailS]

Forester #:

If you compare - it would be interesting to know the result

I am deeply convinced that if the same thing could be calculated by a primitive formula such as

Bal*EP = balance * fabs( balance / number of deals)

no one would be doing training, crosvalidation and a lot of other things. they are not inventing all this from a sweet life.

[СанСаныч Фоменко]

Forester #:

MT formulas are simple, everything from here can be pulled out and replicated

https://www.mql5.com/ru/articles/1486

https://www.metatrader5.com/ru/terminal/help/algotrading/testing_report#recovery_factor

https://www.mql5.com/ru/forum/4485

What I described in #30913 is not realisable in metaquote testers in principle. Methaquot tester is not about retraining at all, you can easily get a retrained TC.

[Forester]

СанСаныч Фоменко #:

What I described in #30913 is not realisable in metaquote testers in principle. Methaquot tester is not about retraining at all, you can easily get a retrained TC.

I am not talking about the tester, but about the formulas for evaluating the obtained result. You can easily write them in R and choose the best model using them. It is much better than by classification error, because it takes into account profit, and Bal*RF will also show the model with minimal drawdown with a good balance.

[Forester]

mytarmailS #:

I'm deeply convinced that if the same thing could be calculated with a primitive formula such as

Bal*EP = balance * fabs( balance / number of trades)

then nobody would be doing training, crosvalidation and a lot of other things. they are not inventing all this from a sweet life.

This package calculates by Sharpe, but they write that you can use any other metric.

And the suggestion was to compare them to see what's best.

[СанСаныч Фоменко]

Forester #:
I'm not talking about the tester, but about formulas for evaluating the result obtained. You can easily write them in R and choose the best model using them. It is much better than by classification error, because it takes into account profit, and Bal*RF will also show the model with minimal drawdown with good balance.

You wrote above that R was removed.

There are no problems at all on R as an instrument. Problems with brains.

[Valeriy Yastremskiy]

СанСаныч Фоменко #:

Brain problems.

More loyal and kinder definitions would be nice.))))) Everyone's moschis are different)))) And weight doesn't matter)

[СанСаныч Фоменко]

Valeriy Yastremskiy #:

Would be nice with more loyal and kind definitions.))))) Moshi is different for everyone)))) And weight doesn't matter)

I'm actually talking about my own brains, not other people's brains

[mytarmailS]

Well, I finally wrote the code...

The first tests show that it does work...

[Maxim Dmitrievsky]

Valeriy Yastremskiy #:

Would be nice with more loyal and kind definitions.))))) Moshi is different for everyone)))) And weight doesn't matter)

it's called aggressive marketing, what do you even understand?

[Maxim Dmitrievsky]

I've already noted several times that GPT is sometimes more convenient than searching the internet, using python as an example. In the same Pandas there are a million phrases, you can't remember them all, and it answers quickly, while Google takes longer. And the answer is immediately given in the context of the question, not just an abstract example. Not bad.