Machine learning in trading: theory, models, practice and algo-trading - page 1262

 
Yuriy Asaulenko:

This is a dead-end version, except for Market and not more. For my beloved, there are better and more interesting options. It seems that you have come to this point too. Didn't you say it before?

Yes, I'm doing something in python, I'm working on it mostly with Bayes... I don't really care where, as long as I have the libs

by speed - mql is faster
 
Maxim Dmitrievsky:

Yes, I'm doing something in Python, I'm mostly doing Bayesian now, it's fun... in principle, it doesn't matter where, as long as the libs are there

I have no idea how to use mql, but it's faster by speed.

Do you need speed? What for? What are you going to do with it?

Imho, you need speed only for pipsing. I can't use it for forex, for Forts even the hand reaction is enough, which takes 0.2-0.5 seconds.

 
Yuriy Asaulenko:

Do you need speed? What for? What will you do with it?

Imho, the speed is needed only for pipsing. In forex it's impossible, in Forts even the hand reaction is enough for free, and that's min 0.2-0.5s.

Monte Carlo simulation on python ~10 minutes, on mql <2

well, this is the cost of dynamic typing

 
Maxim Dmitrievsky:

Monte Carlo simulation on python ~10 minutes, on mql <2

Well, that's the cost of dynamic typing.

Strange. Monte Carlo in Python for 55 t points only takes 13 seconds, together with the calculation of quite complex indicators.

Next is training, but it takes me almost 24 hours, which doesn't bother me at all)). Compared to the system design time, it's nothing.

Python 3.7 Anaconda.

 
Yuriy Asaulenko:

Strange. Monte Carlo on 55 t. points only 13 s.

Next is training, but I have almost 24 hours, which does not embarrass me at all.) Compared to system design time, it's nothing.

150k I have about 10 minutes, through the pyMC3 lib, well +-... not critical, just a comparison. But the code is minimum.

on your ultrabook ) it's weak
 
Maxim Dmitrievsky:

150k I have about 10 minutes, through the libu pyMC3, well +-... yeah not critical, just a comparison. But the code is minimum.

On an ultrabook ) it's weak.

Why do you need a lib? Monte Carlo = HSCI + a few lines of code.

My laptop isn't very powerful either, it's from 2008. I think it's time to change it, but it suits me fine so far.

 
Yuriy Asaulenko:

Why do you need a lib? Monte Carlo = GSCH + a few lines of code.

There's all sorts of stuff in there that I'm exploring.

Here, if you're interested https://docs.pymc.io/

PyMC3 Documentation — PyMC3 3.6 documentation
  • docs.pymc.io
Sometimes an unknown parameter or variable in a model is not a scalar value or a fixed-length vector, but a function. A Gaussian process (GP) can be used as a prior probability distribution whose support is over the space of continuous functions. PyMC3 provides rich support for defining and using GPs.
 
Maxim Dmitrievsky:

there's all kinds of stuff in there that I'm exploring

Here, if you're interested https://docs.pymc.io/

What's interesting there is variation problems and Theano.

I'm still going to use variational methods to tune the system, but I haven't found the approach yet.

 
Yuriy Asaulenko:

My laptop isn't very strong either, it's from 2008. I think it's time to change, but so far this one is fine.

Mine is 2 years old, i5u processor, zen book from asus. I like it because it's easy to carry around. I also want to have a more powerful phone.

 
Yuriy Asaulenko:

What is interesting there is variation problems and Theano.

Well, in general, the topic in general, probabilistic approach. I learned a lot of new things, I've never encountered this before. Only Rolling regression alone is already interesting