Машинное обучение в трейдинге: теория, модели, практика и алготорговля - страница 1073

[Maxim Dmitrievsky]

Alexander_K:

Индусы как программеры - умные.

Я с ними частенько сталкивался в американских и немецких компаниях.

Пусть делает все подчистую, под твоим руководством, и выкладывает нейронный Грааль прям сюды.

Да какую-нибудь х.. очередную родим скоро :)

[FxTrader562]

Hi Maxim,

if you are still online...just check this code quickly since I will run it on test:

 this line of code is showing compilation error:

ERROR: inpVector array required

 ArrayCopy(ag1.agent[i].inpVector[l],arr,l,0,1);

[FxTrader562]

Maxim Dmitrievsky:

sorry

Fixed...but ArrayCopy error..hehe:)))

I will figure it out..

[FxTrader562]

Maxim Dmitrievsky:

double arr[1];

:)

No, ERROR: ArrayCopy: some operator expected

[FxTrader562]

Maxim Dmitrievsky :

this is how I make my codes

Got it :)))))))))))))))))))))))

You can imagine my situation if I'd have done everything myself ... hehe :))))))))

Next, if I'll program candle simulation by myself , then what will happen:))))))))))))))))))))))))))))

[FxTrader562]

Hi Maxim, 

I have completed the GMDH library using the first version of the code given by you...I have ignored the version which you gave yesterday since I didn't find any link.... I mean I have just replaced the kernel function by my GMDH function(CalculateNeuron()) That's all......But there are lot of compilation error coming. So I can't figure it out..If you have already completed the GMDH Library, then it is great!!!!...

But if you are still trying to write the GMDH code yet...then, I can provide you both the modified version of .mqh file and MQL5 file and I guess you can easily figure out the compilation errors.... or if required then you can do slight modification of the code as per your need..

[FxTrader562]

Maxim Dmitrievsky:

Hi, I did not do anything today. You can just show and explain how do you see it in combination with RDF

[FxTrader562]

Maxim Dmitrievsky:

This will not work, better to call this function CalculateNeuron(ker,degree)in all places where it needed. After it, need to save all coefficients and predictors numbers into file, and use it after learning

this is hard logic to implement ) I'll try my method, but selection algorithm with RDF is hard.. I'll show you later

Yes, that's exactly. I have not done much changes to the EA code except 2 lines inside "calsignal()" function. You can just call the function " CalculateNeuron(ker,degree)" at anytime and based on random value of base components it will transform the feature.

So if your previous RDF code is correct:))..which I guess so it is correct..then, it should work...

Anyway, if you have better ways to do it..no problem:))...

I was finding it very difficult to implement initially and now I got this idea of shortcut:)).. and hence, I thought to update you....

Because there is no point re-writing the whole code again. You can just add this function and it will do this job..

[FxTrader562]

Maxim Dmitrievsky:

First line of selection is totally correct now. 2-st line we must combine all best variables with each other in loop, and learn RDF with each combination of features. Also here we can appy different polynomial transformations

this is how I see it

It took me lot of time to understand your first code to implement this:)))..so now please don't give me another code:)))

Whenever you want to transform a feature like close price..please make sure to collect it in a variable either from new price data or old trained data from matrix..that you should do.:)))

Then, just call the function "CalculateNeuron(ker,degree)" with the variable and then, when you get the new feature update it in Matrix or whatever you want to do it:))

I mean whole task of feature transformation using GMDH will be done inside function which I understand it properly...Now, how you feed the values and retrieve the values that is your choice...)))))

[FxTrader562]

Maxim Dmitrievsky :

well, just let me more time .. I dont understand nothing yet, but all be good :)

too many combinatorics here

but on output all must be clear - just n-dimentional array with selected features and formulas

Well,if you already understood your previous code(which I understood a little so far:)).. then it should be just 2 minutes work for you:))... because I am not doing anything here..just copying your previous code and replacing it by GMDH:))

But if you want to do it another way...you can take your own time:)))

Of course, I can't guarantee about my implementation of GMDH:))... and we can't know anything until we run the final EA in LIVE trading mode:)))..even backtesting results don't seem to be reliable...

So you can try your own way or you can just let me know in case if you need the code for "(CalculateNeuron(ker,degree)" to implement

If it will work, then I can even expand the base components to 20 or 30... and it can be slow during training and may be slow in trading also due to multiple for loops..but since it will just check one value of degree at a time and hence, we can expect average speed..