Machine learning in trading: theory, models, practice and algo-trading - page 2672

New comment
 
Valeriy Yastremskiy #:

No, it is possible, of course, and you can even compare strong signals, but not weak ones yet. In order to compare weak or all signals, we need a more frequent time decomposition, perhaps, or something else. We realise that the impulse gives a damped wave. But we can't calculate it yet.

Well, then we don't understand each other.

 
Valeriy Yastremskiy #:

No, it is possible, of course, and you can even compare strong signals, but not weak ones yet. In order to compare weak or all signals, we need a more frequent time decomposition, perhaps, or something else. We realise that the impulse gives a damped wave. But we can't calculate it yet.

Nature has already calculated everything for us. It's only necessary to observe carefully.

 
Valeriy Yastremskiy #:

No, it is possible, of course, and you can even compare strong signals, but not weak ones yet. In order to compare weak or all signals, we need a more frequent time decomposition, perhaps, or something else. We realise that the impulse gives a damped wave. But we can't calculate it yet.

Do you have a sample on this problem? It's interesting to poke around.

 
mytarmailS #:


is that what it's supposed to look like?


P[i] - log( mean(P[ii] ) ) * sd( P[ii] )*150

where " P[ii ] " is the last 20 prices

and " P[i] " is the current price.

That's what it should look like with a factor of 150.

250

And if you increase the period of Mashka it would look like this


 
Maxim Dmitrievsky #:

the point is to cram as much information as possible into a single feature, while making it stationary

then compare it with your tags by some information criterion, and go through it until you get the maximum.

This one will be the best in every sense.

and it's faster than forest or NS.

How do you measure informativeness? What do you use to change the oldness? How do you measure that this feature is better than others?
 
mytarmailS #:
What do you use to measure information? What do you use to change the old-fashionedness? How do you measure that this attribute is better than others?

I'm just concerned with stationarity. So that the signs don't go out of range on new ones and there's not a lot of sticking in a certain position.

The rest is above.

you and your sinusoids have distanced yourself from the MO in general, it seems :) there are no specialists left in the topic, only Alexey sometimes writes something about statistics and distributions.
 
Maxim Dmitrievsky #:

the rest above

you and your sinusoids have gone away from the MoD altogether, it seems :) there are no specialists left in the subject.
What good is it to have everyone doing the same thing?

Sinusoids are a type of basis functions, i.e. one of the methods of approximation/training, as well as polynomials, neural networks, forrests....

Only sinusoids are noise-free, with clear parameters and it is clear how to do invariance with them, so I think I am at least keeping up with the experts))).


Ahah, you are raping your own machine, and you are laughing with Sinusoids))))
 
mytarmailS #:
What good is it if everyone does the same thing?

Sinusoids are one kind of basis functions, i.e. one of the methods of approximation/training, as well as polynomials, neural networks, forrests....

Only sinusoids are noise-free, with clear parameters and it is clear how to do invariance with them, so I think I am at least keeping up with the experts))).


Ahah, you yourself are raping the machine, and you are laughing with Sinusoids)))).
What do they have to do with quotes?)
They are used to encode time components with strict periodicity, nowhere else, I think. To translate categorical variables into continuous variables. And even with that they do a poor job, rbf kernels are brighter.
Define the scope. It feels like you hit the MOE ceiling, realised that nothing works and decided to start degrading so that you can push yourself off the bottom with renewed vigour 🙄
 
Maxim Dmitrievsky #:
What do they have to do with quotes?)
They are used to encode time components with strict periodicity, nowhere else I think. And even with that they do a poor job, rbf kernels are brighter.
Outline the scope of the application. It feels like you hit the MOE ceiling, realised that nothing works and decided to start degrading so that you can push yourself back from the bottom with new strength 🙄
The field of application is wider than neural networks... For example, the automatic creation of data, including target data. And a thousand other areas.

You should read at least something written by normal engineers who solve real problems, because on blogs about python written by 11th graders you won't get far.

You should read what is filtering, adaptive filtering, spectrum, what it is for, why it is needed, etc.
1...266526662667266826692670267126722673267426752676267726782679...3699
New comment