Machine learning in trading: theory, models, practice and algo-trading - page 840

[Maxim Dmitrievsky]

Finally, I found a decent RL video in Russian :)

You can tell by the faces on the splash screen, but there's not enough facepalm.

[Grigoriy Chaunin]

Alexander_K2:

I will not let this branch fall out of the Top level.

Gentlemen of the neural network - the common people are waiting for the Grail from you. Don't ruin it.

Here everyone makes his own grail. Here we share experience, debate and trolling. No one will give you a ready-made grail. So get involved and study the topic.

[Alexander_K2]

Grigoriy Chaunin:
Here everyone makes his own grail. And here we are sharing experiences, arguing and trolling. No one will give you a ready-made grail. So get involved and study the subject.

I do not need a ready-made grail - everyone has a right to secrecy in such cases. But I would not refuse a signal. It would be more trustworthy than the signals of self-taught novices without any mathematical apparatus. But, alas... I don't see any Signals in this thread. An inexplicable paradox!

I know for a fact that the problem of predicting BP is solvable, read Kolmogorov. The question with predictors remains to be solved. I stand by my opinion - they are returns in a sifted stream.

I am waiting for a real breakthrough here, a great signal, and getting my pockets ready. Having read on the forum of masters of epistolary genre, their money dreams and reveries, I too have become unrestrainedly indifferent to money.

PS To study new toolkit, except Vissim in which I am the professional, I am lazy, I am already old... And I do not have Vissim NeuralNet module and cannot find it anywhere.

[Dr. Trader]

Alexander_K2:

are returnees in a sifted stream.

Supported by

[Grigoriy Chaunin]

Alexander_K2:

I do not need a ready-made Grail - everyone has the right to secrecy in such matters. But I wouldn't refuse a signal. It would be more trustworthy than the signals of self-taught people without any mathematical apparatus. But, alas... I don't see any Signals in this thread. An inexplicable paradox!

I know for a fact that the problem of predicting BP is solvable, read Kolmogorov. The question with predictors remains to be solved. I stand by my opinion - they are returns in the sifted stream.

I am waiting for a real breakthrough here, a great signal, and getting my pockets ready. After reading on the forum of masters of epistolary genre, their monetary dreams and reveries I also became something unrestrainedly indifferent to money.

PS I am lazy to study new toolkit, except Vissim, in which I am the professional, I am already old... And I have no module of Vissim NeuralNet and can not find it anywhere.

I am looking for a grail for myself. Why should I get soaked in signals. It's a responsibility to people. And if I leak? What will I tell people. Yes, I know what risk management is.

[Maxim Dmitrievsky]

Dr. Trader:

Support

Which side do you support and how?

[Alexander_K2]

As a continuation of the post https://www.mql5.com/ru/forum/86386/page836#comment_7062634, I attach the already converted tick BP to Erlang flows of different orders.

The first two digits in the file name are the order of the Erlang flow.

For example: 01 AUDCAD ... - simple flow, 30 AUDCAD ... - 30th order Erlang flow

Data processed for April 02-06, 2018.

It would be interesting to see if the quality of the forecast improves as the order of the flow increases.

If necessary, I can continue transforming the flux up to order 100

[Maxim Dmitrievsky]

Alexander_K2:

As a continuation of the post https://www.mql5.com/ru/forum/86386/page836#comment_7062634, I attach the already converted tick BP to Erlang flows of different orders.

The first two digits in the file name are the order of the Erlang flow.

For example: 01 AUDCAD ... - simple flow, 30 AUDCAD ... - 30th order Erlang flow

Data processed for April 02-06, 2018.

It would be interesting to see if the quality of the forecast improves as the order of the flow increases.

If necessary, I can continue flow transformations up to order 100

Hang on a bit, I'll make the streams without any csv right in the terminal. I'm sure you can, just haven't delved into it - busy with other tricks so far.

May I clarify again: you take the source cotier, convert it to Erlang, and then take its returnee?

Or does it take a return and then convert it to Erlang?

My head is just seething with all sorts of NS, constantly busy ))

[Alexander_K2]

Maxim Dmitrievsky:

Wait a bit, I'll make streams without any csv directly in the terminal. I'm sure you can, I just didn't get into it - I'm busy with other tricks so far.

May I clarify again: you take the source cotier, convert it to Erlang, and then take its returnee?

Or does it take a return and then convert it to Erlang?

My head just boils over from all sorts of NS, constantly loaded )))

We take the tick flow, but read it not every tick, but in exponential time intervals (to be more exact, in time intervals obeying the discrete geometrical distribution with p=0.5). We have the simplest flow of events. Then we "sift" this initial VR. We obtain Erlang flows of different orders. It is necessary to input returns of these streams.

This experiment will unambiguously answer the question whether we need BP thinning or not.

[Slasher111]

Alexander_K2:

None of the participants in this thread has a trading signal. An inexplicable paradox!

Alexander, the absence of the signal can be easily explained. For example: a good machine learning model by itself is not a good trading system. In my personal experience it is not even a half of a good trading system. A more important part of TS is money management, by which I mean not only and not so much the position size, but also such things as where to open after the signal, how to open, what to do if the price moved in the right direction, what to do if the price did not move in the right direction, how to close at various changes of volatility, etc. Without a quality MM, it usually turns out not very good. Setting a stop and a take when opening a position and doing nothing more is a "can-do" approach. The only thing worse than that is not setting any stops and takeoffs at all.

A quality MM is a very valuable thing. It, in contrast to the machine learning model, does not become unusable over time. And if desired, it can be completely deduced from deals, i.e. also from the signal. At least for this reason I would not open a signal and would not show my trades.

Alexander, I sincerely advise you to pay more attention to MM in your trading. Take my word for it, the mere entering Buy on the "price will go up" signal and reversing into Sell on "price will go down" signal is not enough. And without any stops if memory serves me correctly.