search

tensorly / tensorly

TensorLy: Tensor Learning in Python.
http://tensorly.org
Other
1.51k stars 281 forks source link

Panel Dataset Time, Company, Feature #544

Closed firmai closed 2 months ago

firmai commented 3 months ago

I have a dataset in the following format, what I would love is a decomposition that would allow for a comparison of tickers over time, to decompose all the feature data into a singulr component.

Time, Company, Feature, to Time, Company, Single Component.

Is that somethig that this library can do?

Thanks in advance :)

aarmey commented 2 months ago

Hi @firmai! If you are wanting to decompose just one dimension of your dataset into component patterns, you can just use PCA/SVD. If you flatten time and company into one axis, such that your dataset is time-company x feature, you can then use PCA to derive the component patterns. Your time-company scores can then be reshaped back into time x company x component.

Hope this helps!