Machine learning in trading: theory, models, practice and algo-trading - page 3242

[Aleksey Nikolayev]

Maxim Dmitrievsky #:
I am struck by the desire of offtoppers to assert themselves at the expense of others' failures, apparently they don't have enough of their own because of their uninteresting existence :))

It seems that this allows them to feel themselves great practitioners of trading, when more sensible ways of doing it are not available for obvious reasons).

[Aleksey Nikolayev]

Aleksey Vyazmikin #:

Another MM question, will there be any restrictions here?

I also immediately thought about the necessity of martingale).

[blef]

My understanding:

Robot template should be inserted as a fragment of the

Code in the *.mq5 file, while making a check of MQLInfoInteger() for the attribute

MQL_TESTER?

[Vladimir Perervenko]

Maxim Dmitrievsky #:

... Neural networks are probably more difficult, maybe only in python.

Any preprocessing, at the execution level of an already trained model, is usually pretty simple and can be rewritten in another NPS.

1. There is no problem with neural networks in R. There's a ported Torch(1.13.1), and there's H2O. Both can be converted to ONNH with some gymnastics.

2. The complexity of preprocessing depends on the knowledge and skills of the programmer. And not "any" can be rewritten or converted to ONH.

The question about the contest is different:

• will it be possible to use your own predictors or will everyone use some common ones proposed by the developers.
• If the testing will be done on the 2023 history, what period should the models be trained on? It is possible to present a model trained on the same period and it will show good results on the test. How do you test this?

" ONNX can be compared to a specialised programming language in mathematical functions. It defines all the necessary operations and the Machine Learning Model should implement the output function with this language". If we take into account that this "language" is constantly expanding and changing rather quickly, then the problems with debugging the ONNX code itself and its implementation in the MCL will be a sea of lost time.

Otherwise, the idea is interesting for those who have a lot of free time.

Good luck

[mytarmailS]

I'm telling you, docker is the real solution.

Any code, any language, any libraries.

But that's not the case.

[Aleksey Vyazmikin]

Vladimir Perervenko #:

• will it be possible to use your own predictors or will everyone use some common ones proposed by the developers.
• If the test will be conducted on the history of 2023, on which period should the models be trained? It is possible to present a model trained on the same period and it will show good results on the test. How will you check this?

I think they said that you can use your own mqh file with your own code, so you can use your own predictors, as I understand it.

But again, if you need special indicators, you can't attach them through the resource....

[fxsaber]

Forum on trading, automated trading systems and testing of trading strategies

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

fxsaber, 2023.09.13 19:28

On the discussion of the robot template for Tester.

Checking the template on a standard MA, only on ticks.

#include "EMA.mqh" // https://www.mql5.com/ru/code/download/22770/ema.mqh

input int inPeriod1 = 50;
input int inPeriod2 = 150;
input int inSensitivity = 100;

// https://www.mql5.com/ru/forum/86386/page3239#comment_49322318
double SignalONNX( const MqlTick &Tick )
{  
  static EMA EMA1(inPeriod1);
  static EMA EMA2(inPeriod2);
  static int Sensitivity = 0;
    
  const double Price = Tick.bid * Tick.ask;
  
  Sensitivity += Price ? 1 - ((EMA1.Get(Price) > EMA1.Get(Price)) << 1) : 0;  
  Sensitivity = MathMax(-inSensitivity, MathMin(Sensitivity, inSensitivity));

  return((Sensitivity == inSensitivity) ? 1 : -(Sensitivity == -inSensitivity));
} 

Working template.

[Renat Akhtyamov]

fxsaber #:

Checking the pattern on a standard MA, only on ticks.

Working pattern.

Now on real cents, 10 quid, preferably without reinvesting.

and you're gonna lose, 100%.

because inreal trading there will be a lot more negative trades than positive ones.

and the epic with the glorious MO will finally be over.

[fxsaber]

Renat Akhtyamov #:

and the epic with the illustrious MoD will finally be over.

The advisor has nothing to do with the MoD. This is a technical test of the proposed template.

Why go to the MO branch and report that MO does not work - I do not know.

[blef]

fxsaber #:

Checking the pattern on a standard MA, only on ticks.

...

A working pattern.

and still unclear: how onnx-signal will get into this working template?