Machine learning in trading: theory, models, practice and algo-trading - page 2116

 
Aleksey Vyazmikin:
No, according to the percentages there is a learning curve type - without MOI 40%-45% are profitable, and with MOI 60%-65%. But it's not an indicator for trading if the profit is not equal to the loss.

It would be an indicator if TP=SL.

 
elibrarius:

It would be an indication if TP=SL

That's what I wrote...

 
Aleksey Vyazmikin:

Please let me know if you find it, otherwise I will start building my own bicycle :)

elibrarius suggested the same idea - just build a branching tree and use it instead of clustering, taking information from leaves in order to reduce majority class.

I don't understand what you're writing and what this has to do with clustering

 
Maxim Dmitrievsky:

I do not understand what you are writing and what it has to do with clustering

This is essentially clustering with the target in mind.

Did you manage to find anything useful on this topic?

 
Aleksey Vyazmikin:

This is essentially clustering with the target in mind.

Were you able to find anything useful on this topic?

There's no such thing as target-based clustering.

 
Maxim Dmitrievsky:

there is no clustering with the target in mind

In textbooks - probably :)

It's clustering on a limited number of attributes simply.
 
Aleksey Vyazmikin:

In textbooks - probably :)

It's clustering on a limited number of attributes simply.

Read what clustering is, I don't want to bore you with it.

 
Aleksey Vyazmikin:

Did you manage to find anything useful on this topic?

it's a serious study on class balancing, haven't finished yet

 
Maxim Dmitrievsky:

there is no clustering with the target in mind

Each leaf can be called a cluster with maximum class partitioning.
You yourself agreed with my similar idea about half a year ago.
 
elibrarius:
Each leaf can be called a cluster with maximum class separation.
You yourself agreed with my similar idea about half a year ago.

I don't know what you're discussing here.

there are 2 feature spaces (I took 5 main components of each)

In the case of random sampling of transactions:

In the case of clustering into 2 clusters:

Problem: find the trade-off between correct labels and good class partitioning.

In the case of simple clustering, the labels are of course unsuitable for trading

In the case of transaction sampling, the feature space is not good