issues: 1165654699

This data as json

id node_id number title user state locked assignee milestone comments created_at updated_at closed_at author_association active_lock_reason draft pull_request body reactions performed_via_github_app state_reason repo type
1165654699 I_kwDOAMm_X85Fenqr 6349 Rolling exp correlation 5635139 closed 0     1 2022-03-10T19:51:57Z 2023-12-04T19:13:35Z 2023-12-04T19:13:34Z MEMBER      

Is your feature request related to a problem?

I'd like an exponentially moving correlation coefficient

Describe the solution you'd like

I think we could add a rolling_exp.corr method fairly easily — i.e. just in python, no need to add anything to numbagg: ewma here means rolling_exp(...).mean - ewma(A * B) - ewma(A) * ewma(B) for the rolling covar - divided by sqrt of (ewma(A**2) - ewma(A)**2 * ewma(B**2) - ewma(B)**2 for the sqrt of variance

We could also add a flag for cosine similarity, which wouldn't remove the mean. We could also add .var & .std & .covar as their own methods.

I think we'd need to mask the variables on their intersection, so we don't have values that are missing from B affecting A's variance without affecting its covariance.

Pandas does this in cython, possibly because it's faster to only do a single pass of the data. If anyone has correctness concerns about this simple approach of wrapping ewmas, please let me know. Or if the performance would be unacceptable such that it shouldn't go into xarray until it's a single pass.

Describe alternatives you've considered

Numagg

Additional context

No response

 
{
    "url": "https://api.github.com/repos/pydata/xarray/issues/6349/reactions",
    "total_count": 0,
    "+1": 0,
    "-1": 0,
    "laugh": 0,
    "hooray": 0,
    "confused": 0,
    "heart": 0,
    "rocket": 0,
    "eyes": 0
}
   completed 13221727 issue

Links from other tables