Machine learning in trading: theory, models, practice and algo-trading - page 3375

New comment
 
In our case, the original data are time-series, not tables. If we don't use news analysis; if we do, we use tables.

So, in the first case, both NS and bousting are suitable, perhaps NS is better, it depends on the data representation. In the second case, bousting is better.

So we have learnt to distinguish between the initial data representation and its representation after processing.
 
Aleksey Nikolayev #:
If they can be written in a table but cannot be written in a matrix)
Well, NSs work better on homogeneous data. Tabular data can be written into a matrix if they are of the same type.
 
For tabular data, there is a neuronics TabNet architecture

It is positioned as a competitor to boosts.
I've tried it, it works well, it doesn't skam...
 
There are such networks, yes. But our topic requires networks for working with sequences rather than tables. Because they are sequences from the beginning.
 
Maxim Dmitrievsky #:
There are such networks, yes. But our topic requires networks for working with sequences rather than tables. Because they are sequences from the beginning.

I'm in a mood.

Can you prove that they are sequences? Apart from the fact that they are sequences.

 
Tabular data as I understand it from this tip-off

It's what's called tidy data. That is, "tidy data."

It is a table where each row is an observation and the column is a feature.

 
Maxim Dmitrievsky #:
topics need more networks to work with sequences rather than tables.
I don't get it, sequences can't be in table format?
 
Maxim Dmitrievsky #:
There are such networks, yes. But our topic requires networks for working with sequences rather than tables. Because they are sequences from the beginning.

The first option, tables - Excel spreadsheets, each row has a time marker. The most familiar form of financial data.

Second option, handwritten letters. Learning with a teacher, with a printed letter as the teacher, and a column below it of handwritten variants of that letter.

Comparing bousting and NS. Which is more suitable and for which case? Or is it equivalent?

PS.

From Rattle, which has rpart (simple tree), rf, ada, SVM, glm, nnet (probably the simplest NS). The worst result is with rpart, second from the end is nnet, the other four are about the same, depends on the input data.

 
Maxim Kuznetsov #:

I'm in such a mood.

can you prove that they're sequences? Apart from the fact that they're sequences.

Time-series is more accurate. It seems to me that you have to offer an alternative first. Otherwise it's either something or nothing.
 
mytarmailS #:
I don't get it, sequences can't be in table format?
Initially no. After processing it can look any way you want and lies entirely on the conscience of the date satanist.
1...336833693370337133723373337433753376337733783379338033813382...3699
New comment