rrpy is a scikit-learn compatible Python implementation of reduced rank ridge regression.
It is based on the rrs.fit method of the R package rrpack, which is in turn based on [1].
pip install git+https://github.com/krey/rrpy.git
This implementation does not support missing values, though such a feature could be added using https://github.com/aksarkar/wlra.
The ReducedRankRidge estimator has a memory parameter which allows rapid tuning of the rank parameter:
import sklearn.datasets
import joblib
from rrpy import ReducedRankRidge
X, Y = sklearn.datasets.make_regression(n_samples=1000, n_features=500, n_targets=50, random_state=1, n_informative=25)
memory = joblib.Memory(location='/tmp/rrpy-test/', verbose=2)
estimator = ReducedRankRidge(memory=memory, rank=10)
estimator.fit(X, Y)
estimator.rank = 20
estimator.fit(X, Y) # cached
memory.clear(warn=False)[1] Mukherjee, A. and Zhu, J. (2011) Reduced rank ridge regression and its kernel extensions.