Machine learning in trading: theory, models, practice and algo-trading - page 541

 
Aleksey Terentev:
https://github.com/Artelnics/OpenNN is an easy to learn library. But it lacks many modern techniques. Regression is there, but scaffolding is not.
https://github.com/Microsoft/CNTK - Multitool. Not studied. As a dll option.
https://github.com/BVLC/caffe - Also quite powerful, for a dll option.

looked it up, did not find any descriptions on how to get a linear regression trait estimate

So far, I liked http://dlib.net/, but have not yet had time to get into it. It looks friendly and portable

Instead of scaffolding, I think DNN would be fine, the main thing is that it should work fast

maybe it is possible to unlock some R package, I will have to look for it :)

dlib C++ Library
dlib C++ Library
  • dlib.net
Dlib is a modern C++ toolkit containing machine learning algorithms and tools for creating complex software in C++ to solve real world problems. It is used in both industry and academia in a wide range of domains including robotics, embedded devices, mobile phones, and large high performance...
 
Maxim Dmitrievsky:

looked it up, did not find any descriptions on how to get a linear regression trait estimate

So far, I liked http://dlib.net/ but did not have time to get into it, it looks friendly and portable

Instead of scaffolding, I think DNN would be fine, the main thing is that it should work fast

maybe you can work out some R package, I'll have to look for it :)


Why do you have such an aversion to R?

It's got everything, it's well documented, it's a world-class system, unlike some village trinkets, because even Matlab can't compete with R!

The question of performance is very controversial. 12% is implemented in Srr, all of them computationally complex algorithms. All cores load fine, you can load the neighboring computers too... What else do you need? No, this is another exotic idea.

 
SanSanych Fomenko:

Why do you have such an aversion to R?

It has everything, and it is well documented, a world-class system in contrast to various village trinkets, because even Matlab can no longer compete with R!

The question of performance is very controversial. 12% is implemented in Srr, all of them computationally complex algorithms. All cores load fine, you can load neighboring computers too... What else do you need? No, this is just another condemnation of something exotic.


I don't particularly dislike it, it's just that when you only need a couple of classes it's kind of weird to connect the whole environment and make the system inconvenient, it would fall from build to build terminal... Plus, I'm a fan of MT5, if in the future they port another couple of libraries like alglib, but more modern with MO, then all the research can be done in it almost as quickly as in R. Caesar's to Caesar, trader's to trader :)

Or Alexey wrote a little bit of code for MO and couldn't install it on the market, because dll calls aren't supported there... Otherwise it would be a very cool product, and now I have to make doctored parts. And some things have to be dumped on the market for the development of the community

+ anyways slow as you mentioned, from 5 to 1000 times (the language itself, not cpp libs)

 
Maxim Dmitrievsky:


+ in any case slow as you mentioned, from 5 to 1000 times (the language itself, not cpp libs)


I don't remember writing such a thing.

All matrix operations are executed at maximum speed, for example. On the subject of speed, you must be specific: here's a program, and here's an analog, and here's the result. And you have to be careful because you have to write code in R instead of repeating similar constructs in R. For example, by repeating loops instead of matrix operations.

In fact, this is all nonsense.

Once you switch to R, you'll forget that anything is missing. That seems to me to be the main thing.

 
SanSanych Fomenko:

I don't remember writing such a thing.

All matrix operations are executed at maximum speed, for example. As for speed, you have to be specific: here is the program, and here is the analog, and here is the result. And you have to be careful because you have to write code in R instead of repeating similar constructs in R. For example, by repeating loops instead of matrix operations.

In fact, this is all nonsense.

Once you switch to R, you'll forget that anything is missing. That seems to me to be the main thing.


But the disadvantages, as I mentioned above, outweigh me for now)

i already have one, it's better to adapt to it.

 
Maxim Dmitrievsky:

I've already listed the minuses above, they still outweigh me )

Real trading and statistical analysis are a bit different things, a lot depends on infrastructure, and if you already have it it's better to adjust to it.

In fact the infrastructure is not the most important thing, sometimes it's better to give it all up. It's easy - I quit three times.) Maybe even four.)

I'm not agitating for R, I started, mastered it and quit. Maybe there are some diamonds in R, but I'm not ready to dive into this unsystematic rubbish dump. Unless, of course, life will make me do it, I will not make any guarantees.

 
Yuriy Asaulenko:
.....

Yuri, how's it going?

I remember the beginning was not bad, now what?

 
Renat Akhtyamov:

Yuri, how's it going?

I remember the beginning is not so bad, now what?

I'm doing okay. But I haven't left any of my systems unattended. When I have time, I do it.

I already told you, I do not publish reports from the real. There will be no details.

 
Yuriy Asaulenko:

Everything is fine. But I don't leave any of my systems unattended. When I have time, I turn it on.

Already said - I do not publish reports from the real. I won't give you any details.

That's right.

I have been waiting for a sensible answer to this question for a long time.

I repeat, the subject is very serious and extensive.

In order to start doing this, I needed confirmation of positive results.

 
Renat Akhtyamov:

That's right.

I have been waiting for a sensible answer to this question for a long time.

Again, this is a very serious and broad topic.

I am not going to do it, I need confirmation of the positive results.

I don't know if this topic will help you.

It helped me that here very smart people, and I learned how not to do.

Well, how to do - no one will not tell you.)