Hi all,

Although dask-ml already has PCA and wrappers.Incremented, but there is no IncrementalPCA.
Is this in the scope of development?

With dask-ml's PCA, we can handle large data, but if the data is too large, some approximated algorithms may be necessary.
With wrappers.Incremental, we could use scikit-learn's IncrementalPCA, but it internally uses scipy.linalg.svd, which won't run with large feature size of data.

For me it would be nicer if dask-ml has dask-compatible IncrementalPCA.

I implemented it in my folk, most part of which is stolen from scikit-learn but uses dask.array.linalg.svd_compressed internally.
If it is in the scope, I will clean it a bit and send a PR.

Thank you!

cc:
@demaheim
@yasahi-hpc