Machine learning in trading: theory, models, practice and algo-trading - page 2848

 
Aleksey Vyazmikin #:

So you built it, and how do you propose to use it?

As for the survival function, I don't understand how to use it.

For example, in the first figure there is a noticeable decline around 8-9, which corresponds to your fence and trees. You can see more precisely the slope of the decline on the survival function graph - there it is defined by a horizontal section.

 
Aleksey Nikolayev #:

For example, in the first figure there is a noticeable decline around 8-9, which corresponds to your fence and trees. You can see the decline segment more precisely on the survival function graph - there it is defined by a horizontal section.

It is clear that the histogram can be divided into two parts, but it is a question of automation and unification of the process. If the spread is not large in absolute terms, the histogram will not show it.

 
Aleksey Vyazmikin #:

It is clear that the histogram can be divided into two parts, but here the question is about automation and unification of the process. If the spread is not large in absolute terms, the histogram will not show it.

That's why I wrote about the survival function (there is also a hazard function). Don't be lazy to learn new things and don't try to invent everything on your own - someone has already solved any of your problems.

 
Aleksey Nikolayev #:

Don't be lazy to learn new things and don't try to make things up on your own - any problems you have have have already been solved by someone else.

++++

 
Aleksey Nikolayev #:

That's why I wrote about the survival function (there is also a hazard function). Don't be lazy to learn new things and don't try to invent everything on your own - someone has already solved your problems.

Let's be specific - what do you propose to do - referring to etheric bodies without productive understanding of how and why to use them is a waste of time.

Ready to help with understanding of the algorithm of building the survival function - I'll write the code and we'll think further. And your assumptions about my laziness are simply insulting, considering the amount of information that I process for application in my tasks.

 
Aleksey Vyazmikin #:

Let's be specific - what do you propose to do - referring to etheric bodies without a productive understanding of how and why to use them is a waste of time.

If you are ready to help me with understanding the algorithm of building the survival function, I will write the code and we will think further. And your assumptions about my laziness are simply insulting, considering the amount of information that I process for application in my tasks.

Roughly speaking, you are not lazy to fell trees, but you are lazy to sharpen an axe.

Risk function, the simplest variant on R

#  x - выборка, y - функция риска
x <- sort(x)
nx <- length(x)
y <- log(nx/(nx:1))
plot(x, y, type = "l")

The sections of the curve close to the horizontal line correspond to dips in the histogram and here these sections can be determined more accurately, since there is no linkage to the partitioning (as in histograms). I use it, for example, when studying the distribution of heights of zigzag knees.

 
Aleksey Nikolayev #:

Risk function, simplest variant on R

What do you mean by nx:1? Don't you get more than one number in the vector y after looking up the logarithm? I don't know the syntax of R.

Aleksey Nikolayev #:

The sections of the curve close to the horizontal line correspond to dips in the histogram and here these sections can be defined more precisely, since there is no link to the partitioning (as in histograms). I use, for example, when studying the distribution of heights of zigzag knees.

"Close" - how to unify the degree of closeness? I want an algorithm, not manual estimation in the end.

So far we're making a simple predictor that will show the percentage of such outliers in the sample. Then we can think about the option of estimating the distribution of these outliers in the sample.

 
Aleksey Vyazmikin #:

What do you mean nx:1? Don't you get more than one number in the vector y after looking up the logarithm? I don't know the syntax of R.

It's a vector of length nx with values from nx to 1. Learn R, sharpen your axe.

Aleksey Vyazmikin #:

"Close" - how to unify the degree of closeness? I need an algorithm, not manual estimation in the end.

Depends on your task, it's up to you to decide how. My variant is unlikely to work for you - I use a deviation from the theoretical form of the risk function calculated for SB.

 
Aleksey Nikolayev #:

It is a vector of length nx with values from nx to 1. Learn R, sharpen your axe.

nx is the number of elements of the same. How can it be up to 1 if it is greater than 1?

The article generally emphasises that you need to know the distribution before applying the method itself.
 
Aleksey Vyazmikin #:

nx is the number of elements of the same. How can it be up to 1 if it is greater than 1?

It decreases from nx to 1. For example, 5:1 = (5,4,3,2,1), and 1:5 = (1,2,3,4,5).

Aleksey Vyazmikin #:

The article generally emphasises that you need to know the distribution before applying the method itself.

As usual in matstat - an empirical analogue is constructed from the sample. Like mean instead of expectation, frequency instead of probability, or ECDF instead of CDF.