Machine learning in trading: theory, models, practice and algo-trading - page 3161

New comment
 

Remember _Vizard kept extending some points in his vids and then deleting them? )

https://github.com/abidlabs/contrastive

GitHub - abidlabs/contrastive: Contrastive PCA
GitHub - abidlabs/contrastive: Contrastive PCA
  • abidlabs
  • github.com
A python library for performing unsupervised machine learning on datasets with learning (e.g. PCA) in contrastive settings, where one is interested in patterns (e.g. clusters or clines) that exist one dataset, but not the other. Applications include dicovering subgroups in biological and medical data. Here are basic installation and usage...
 
Maxim Dmitrievsky #:

Remember _Vizard kept extending some points in his vids and then deleting them? )

https://github.com/abidlabs/contrastive

Then there's PLS. At first glance, the idea is similar.

 
Aleksey Nikolayev #:

Then there's PLS. At first glance, the idea is similar.

There's also t-sne, umap, lle.... And a bunch of other stuff


One thing I don't understand, what is the point of enthusiasm? The director of IT department has never done PCA ? )))

 
mytarmailS #:

There's also t-sne, umap, lle.... And a bunch of other stuff.


One thing I don't understand, what's the point of enthusiasm? The director of IT department has never done PCA ? )))

These seem to be non-linear, and those are both linear like PCA, if I'm not confused.

 
Aleksey Nikolayev #:

Then there's PLS. At first glance, the idea seems similar.

Does it use an additional dataset too?
I didn't see it in the description.
 
Maxim Dmitrievsky #:
Is an additional dataset used there too?

No, the problem formulation is formally different and labels are used there. The similarity seems to me to be in the search of spaces on which the projection is made. I think both approaches are also used in genetics, where the dimensionality of features is huge, larger than the number of examples.

 
Aleksey Nikolayev #:

No, the problem formulation is formally different and labels are used there. The similarity seems to me to be in the search for spaces on which the projection is made. I think both approaches are also used in genetics, where the dimensionality of traits is huge, more than the number of examples.

I will try it later, I wonder what it will show on new data.
 
Maxim Dmitrievsky #:
Does it use an additional dataset too?
I didn't see it in the description.

Had a closer look. Yes, the difference is more significant, although I was not mistaken about the linearity of the method.

cPSA, supposedly, can help to visually find subtle differences between market phases. Let's become wizards too)

 
Aleksey Nikolayev #:

I took a closer look. Yes, the difference is more significant, although I was not mistaken about the linearity of the method.

cPSA, supposedly, can help to visually find subtle differences between market phases. Let's become wizards too)

Yes, when there is a lot of noise, the components with maximum variance can't show anything. I didn't understand how the second dataset is involved there, but it is also something from kozul and there was something about tritment in his videos.
 
Maxim Dmitrievsky #:
Yes, when there is a lot of noise, components with maximum variance can't show anything. I didn't understand how the second dataset is involved, but it's also something from kozul and there was something about tritment in his videos.

Well, it seems that the second dataset is needed just to find the contrast with the first one. It seems that somehow the covariance matrices for both datasets are cleverly compared.

1...315431553156315731583159316031613162316331643165316631673168...3516
New comment
Reason: