Machine learning in trading: theory, models, practice and algo-trading - page 887

 
Maxim Dmitrievsky:

This is not even Low latency, hft is milliseconds and even microseconds, every extra meter of cable counts

You're just scalping.

Yes, I'm aware of that. My point is that HFT methodology can be used on longer distances as well. With some adaptation.

 
elibrarius:

Here are more articles with code examples
https://www.mql5.com/ru/users/vlad1949/publications

I read some of them three times and tried to figure out how those examples work. And then you can make something of your own.

Yes, thanks, great, read it, but I'll have more - it's very abstruse, at each level of training you have to go back, and otherwise half the words are not clear.

But I would like to understand how to load a new sample into Rattle :)

 
Yuriy Asaulenko:

Yes, I'm aware of that. My point is that the HFT methodology can be used at longer distances. With some adaptation.

Well, you can, if you have the stomach for it :D But it may not be in Forex, at least in the ECN market

 
Alyosha:

Shame on me! Shame!!! But now 5% of the error on random rambling, like the cool ones!

Have they realized that the market is a random walk and this is its natural state, while all sorts of jumps and dances are a deviation from the norm).

 

I added such a predictor as a bar hour - there was a significant jump.


It turned out that in Deductor Studio in the training version is not blocked unloading the model in a text file, and blocked unloading only for all sorts of beauties in excel and html. That is why I have a set of rules of this kind

A question arises, how to minimize (organize) these arrays of data for convenience and speed of work?

 
Aleksey Vyazmikin:

I added such a predictor as a bar hour - there was a significant jump.


It turned out that in Deductor Studio in the training version is not blocked unloading the model in a text file, and blocked unloading only for all sorts of beauties in excel and html. That is why I have a set of rules of this kind

Now I have a question, how to better summarize (organize) these data sets for convenience and speed?

I have never collapsed arrays. Explain how to do this, and most importantly, why?
Googled - CollapseArray(nArray) - Removes repetitive values in the array.
Is this it? To remove duplicate rows?
 
elibrarius:
I've never collapsed arrays. Explain how to do it, and most importantly, why?
Googled it - collapseArray(nArray) - It removes repeated values in the array.
Is this it? Remove duplicates?

Now I was walking around thinking about how to do it... In general, yes, I assume that there are repetitions, and they should be removed, but the main thing hit me, that all data is not needed, but only those that confirm the lesser of two 1 or 0. Then I thought to arrange arrays by stairs, like making truncation of repeating parts... or maybe just write in one line, and respectively also combine data in a line from predictors (input data in real time received), but this is until we go beyond string length limit by number of predicts.

 
Aleksey Vyazmikin:

Now I was walking around thinking about how to do it... In general, yes, I suppose that there are repetitions, and they should be removed, but the main thing that occurred to me, that all data is not needed, but only those that confirm the lesser of two 1 or 0. Then I thought to arrange arrays by stairs, like making truncation of repeating parts... or maybe just write in one line, and respectively also combine data in a line from predictors (input data in real time obtained), but this is until we go beyond string length limit by number of predicts.

I think repetition is necessary, it'll make a situation more important. If it was repeated 100 times in the past, it's more likely to happen in the future than it was once or twice
 
Alesha:

I agree, python is the most top-end kernel for HFTs.

For research maybe, for production - I don't believe it.

 
elibrarius:
I think repetition is necessary, it will reinforce the importance of the situation. If it was repeated 100 times in the past, it will probably happen in the future as well, than the one which was 1-2 times

So I will take only the ones that have great support and reliability. This is how I see it working - I have indicators that generate predictors in real time and on history, put them together into a string, and then this string is looked for in the array; if it is found, then we mark the bar as favorable for entering, and if not, we do nothing. Accordingly, the doubled lines will only increase the array. Of course, it is possible to make gradation by color, where will be information about reliability and support (by multiplying one by the other we will get the coefficient, which will change color depending on the value), but for this it is easier just to make a separate array of int type with an index. Or I don't understand something....