Pavan Ramkumar

Hey @inferentialist thanks for your thoughtful suggestions regarding alternate solvers. We've made some improvements to our convergence criteria recently and have made a new release on PyPI along with various...

is this simply a case of adding this line to travis.yml: ``` flake8 --count tests ``` seems to pass locally except for some docstring complaints

sure, i can work on this. the only thing we'd have to decide is whether to clone the bleeding edge github version or provide the pypi version.

@jasmainak i looked at it, but i don't know whether we can use this approach for all the distrs. also, how much slower is it with line search in each...

sounds great! thanks @rvraghav93 I'm also dropping here is alex gramfort's PR for gap safe screening rules in case we add it to our coordinate descent implementation: https://github.com/scikit-learn/scikit-learn/pull/5075/files

Just fyi, I have implemented and tested newton coordinate descent in pure python for the `poisson` case. It works well, by 5-10x slower than current batch gradient descent code. I'll...

Sounds great! My WIP PR will give you a pure python code so that you can focus on translating to cython or numba. I'll aim to get it in to...

Ok, PR is now updated to solve the `multinomial` case with the cdfast method.

Hey guys! @jasmainak @raghavrv I haven't been able to carve out time for cython experiments myself, but I ran into SWIG today: http://www.swig.org/tutorial.html A useful alternative to cython could be:...

@genezin we aren't maintaining this code too actively at this point. we do have certain custom theano routines so i'd suggest using a theano backend instead of tensorflow. let us...