Hussain Sultan

For ADMM, my understanding is that you don't have to be precise for local updates for the solution to converge, which leads me to believe that some numeric testing will...

@moody-marlin Thanks for raising this. > However, we don't want to adhere to this too strongly (especially in output), because we started with the goal of intelligently toggling between inferential...

Ok... these are my thoughts on how to make this work with sklearn interface: ```python from sklearn.base import BaseEstimator, ClassifierMixin def grad(w, X, y, lambduh): ... return out, grad class...

thanks @jcrist I wasnt aware that duck typing was sufficient for sklearn interface. This is super helpful! @moody-marlin i am happy with what you proposed with slight modification that we...

this is asynchronous variant of ADMM: http://jmlr.org/proceedings/papers/v32/zhange14.pdf However, i am not sure if we need it since the global updates are the fastest (unlike line search step above).

@moody-marlin for admm, does it matter if data is sorted or randomized?

ADMM will still converge if the data is not randomized but take a long time, is that what you meant by not a good idea?