search

cornellius-gp / gpytorch

A highly efficient implementation of Gaussian Processes in PyTorch
MIT License
3.57k stars 562 forks source link

Non stationary time series, kernel for induced points #1945

Closed lion241 closed 2 years ago

lion241 commented 2 years ago

Hi,

I just wanted to ask a quick question the case of non stationary time series.

Can I use induced points to pick most relevant data points based on a temporal kernel?

for example: lets say i have a series from 0 to 1000

it happens that the sequence 200 to 300 is similar to the sequence 900 to 1000. I should be able to use an RBF kernel to compare over the sequences.

i do not see how i can do that in the induced points are selected now. In general for the exact case i would introduce the RBF separately and multiply but do i need to do it here to get the same functionality ?

wjmaddox commented 2 years ago

I'm not entirely sure I follow what you're trying to do here. If you look at something like a PCA of the RBF kernel evaluated on either sequence chunk, you're going to be only considering the data in an un-supervised fashion.

What exactly are you trying to do?

lion241 commented 2 years ago

I have actually resolved it , you can close the issue