Machine learning in trading: theory, models, practice and algo-trading - page 3219

New comment
 
mytarmailS #:
We don't have a data scarcity problem

Renaissance had a problem for new instruments with little history. 6,000 instruments is a challenge.)

 
Valeriy Yastremskiy #:

Renaissance had a problem for new instruments with little history. 6,000 instruments is a challenge.)

It's not that difficult... with their capabilities, 6,000 instruments is like a holiday.

 
mytarmailS #:
There is a lot of things, Google to help.
For my TC simulation is not relevant, so I don't even want to get into this topic.

I don't need it either.

But instead of google I thought a bit and came to the conclusion that in R it is not relevant at all.

If parameters of some function, it is possible to find an optimum.

If the input data, it is possible if the data model is known. Then we change the parameters of the data model by apply. If the data model is not known, then all this is nonsense.


Again a lot of rubbish on the branch. It would be better if you had learnt R without being stupid.

 
СанСаныч Фоменко #:
apply

what does this have to do with apply???

What if we need to identify the covariance structure and connectivity of 5 pairs and then create a simulation of such series with the same regularity?

 
mytarmailS #:

what's that got to do with it ??

What if we need to identify the covariance structure and connectivity of 5 pairs and then create a simulation of such series with the same regularity?

You should start with the regularity, or rather first draw a histogram. And gradually simulate the random value at least by eye, bringing the histogram closer to the initial one. Without having the regularity of each series it is impossible to compare something with the result, it is impossible to answer the question: how much the result "resembles" the initial data.

 
СанСаныч Фоменко #:

You should start with a pattern, or rather first draw a histogram. And gradually model the random value at least by eye, bringing the histogram closer to the original. Without having the regularity of each series it is impossible to compare something with the result, it is impossible to answer the question: how much the result "resembles" the initial data.

You don't need any approximations and histograms.... You don't need any fancy stuff either...

Here is an example of simulation of cointegrated series
 
I didn't bother with the binary, you can get ticks in csv via export. There are also a lot of missing fields, you need to fill them in correctly
 
Maxim Dmitrievsky #:
I didn't bother with the binary, you can get ticks in csv via export. There are also a lot of missing fields, you need to fill in correctly 
#property script_show_inputs
#property link "https://www.mql5.com/ru/forum/86386/page3216#comment_49148211"

input string inFileName = "Ticks.bin";

void OnStart()
{
  MqlTick Ticks[];
  
  const int Size = (int)FileLoad(inFileName, Ticks);
  
  if (Size > 0)
  {
    const int Handle = FileOpen(inFileName + ".csv", FILE_WRITE | FILE_ANSI);
    
    if (Handle != INVALID_HANDLE)
    {
      for (int i = 0; i < Size; i++)
        FileWriteString(Handle, (string)Ticks[i].time + "." + IntegerToString(Ticks[i].time_msc % 1000, 3, '0') + " " +
                                DoubleToString(Ticks[i].bid, 5) + " " + DoubleToString(Ticks[i].ask, 5) + "\n");
      
      FileClose(Handle);
    }
  }
}

CSV: time bid ask.

 
fxsaber #:

CSV: time bid ask.

Thanks, MKL in general forgot already 😀 I'll try tonight.

Recently I started writing something, in the end all variables without type and without semicolons
 
mytarmailS #:
You don't need any approximations or histograms.... You don't need any of that stuff either.

Here's an example of a cointegrated series simulation.
h ttps://quant.st ackexchange.com/questions/3930/how-to-simulate-cointegrated-prices

You confirm my thoughts with a concrete example: if we know the regularities in the form of formulas and corresponding code, moreover, we know what we are going to trade, we can do simulation - this is a normal professional approach. And everything that goes beyond this pattern is usual alchemy.

The branch is talking about ticks, the statistical characteristics of which are not interesting. Then there is a conversation at the level of alchemists - something, somewhere..... Here for these people it is suggested to start with a histogram as a first step to a professional approach on the way to simulation.

1...321232133214321532163217321832193220322132223223322432253226...3574
New comment