Machine learning in trading: theory, models, practice and algo-trading - page 187

 
Vizard_:
And if you sound the ticker as well...

Interesting idea, I voiced it, but it came out shitty. I was expecting some cosmic howls or something, but it came out just noise :)

The volume needs to be turned up to maximum, or you can't hear anything.

I attached an Expert Advisor for mt5 to save bars and ticks to a file in case somebody wants to repeat it. And then I need to run such R script to create sound files:

library(audio)
range11 <- function(x){2*(x-min(x))/(max(x)-min(x))-1}
openPrices <- read.csv("dump_open.csv")[,"open"]
ticks <- read.csv("dump_ticks.csv")[,"ticks"]
save.wave(audioSample(range11(openPrices), rate=44100, bits=16, clip = TRUE), "dump_open.wav")
save.wave(audioSample(range11(openPrices), rate=5000, bits=16, clip = TRUE), "dump_open_slow.wav")
save.wave(audioSample(range11(ticks), rate=44100, bits=16, clip = TRUE), "dump_ticks.wav")
Files:
forex_music.zip  15867 kb
 
Vizard_:

Thank you, just what I needed. Got great space noise instead of white noise.

The frequency is slightly different, so that 1 second = 1 day. In glorious stereo!

p.s. 1_1.mp3 -> properties -> details -> genre -> Blues :D

Files:
forex_music_2.zip  6380 kb
 

Does anyone know where you can pull quotes for currencies and indices into the p-ku intraday?

maybe someone will be useful, a collection of sources of financial data for the r-ka, I think you can even download newshttp://www.thertrader.com/2013/11/08/financial-data-accessible-from-r-part-iii/

but what i need is not there, so my question is relevant

Financial Data Accessible from R – part III
Financial Data Accessible from R – part III
  • www.thertrader.com
I came across a new source of data which I think is really worth sharing: ThinkNum. It gathers around 2,000 sources of data but more importantly it allows the user to manipulate this data via funct…
 
mytarmailS:

Does anyone know where you can pull quotes for currencies and indices into the p-ku intraday?

maybe someone will be useful, a collection of sources of financial data for the r-ka, I think you can even download newshttp://www.thertrader.com/2013/11/08/financial-data-accessible-from-r-part-iii/

but what i need is not there, so my question is relevant

what prevents me from doing it directly from MT? or this is post number 1

https://www.mql5.com/ru/forum/73266

Исследования в мат. пакетах
Исследования в мат. пакетах
  • www.mql5.com
Форум трейдеров MQL5.community
 
ivanivan_11:

What prevents you from doing it from MT directly? Or is this post number 1

https://www.mql5.com/ru/forum/73266

Thanks, I didn't even know there was such a thing...

Is there any manual for this miracle?

 

In supporting Reshetov's talk.

Essentially a trinary classifier is a committee of two binary classifiers that capture the essence of the data, with respect to the output. Making the same move. Suppose one network said yes and the other said no. How to be in such a case???? Well of course the answer is "I don't know" If both said YES, then yes, if no, then no..... I think this approach is justified and do you know why?

The point is that when we build an output variable, we can always specify what class this or that signal belongs to (on the history of course) and there is no such a thing as "I don't know" on the history. But I noticed such a peculiarity, that when NS says Don't Know, it usually refers to some class. It is enough to look at the previous "don't know", determine what that signal was true or false, and in the future, when "Don't know" appears, we already know that this class is true, for example. The most important thing is that the division into classes is stable.

Well, that's what I'm saying.... thoughts aloud....

 

In other words, there is no "I don't know" in nature when it comes to the past. In the past, we always know what our signal was. That is why we defined today's "I don't know" as a lie, when we see not knowing, this means that one network points up and the other down, so we know when we don't know, but the first network says yes, and the second doesn't. This is a lie, when the second says yes, and the first no, then the truth. The main thing is not how it divides, but what makes it stable, for example, I have not trained my network for several days, let's see how it works together.

As we see the TS signals for buying are already focused, where "I do not know" it is a lie, well, in fact the signal itself shows us the correct lower one.

The last arrow is not yet oriented, because there works another NS, the signal is to sell, but it turned out to be false according to the NS, let's see......

 
Mihail Marchukajtes:

In keeping Reshetov in the conversation.

No one doubted it :)

 

I admit honestly that the signals sat I had to reorient, but overall not bad!!!

 
Mihail Marchukajtes:

The point is that when we build an output variable, we can always specify to which class this or that signal belongs (on the history of course) and there is no such thing as "I don't know" on the history. But I noticed such peculiarity, that when NS says Don't Know, it usually refers to some class. It is enough to look at the previous "don't know", determine what that signal was true or false, and in the future when "Don't know" appears, we already know that this class is true, for example. The most important thing is to make sure that division into classes is stable.

It reminds me of the missile control system.