Lars

`NULL` is not a valid value for a `double` because it's a pointer value.

I tried fixing some unchecked `malloc` calls, but I got stuck in the code as most functions are just declared `CNAME` and the name is generated by the build system....

Yes, `SGDClassifier` does ``` for i in xrange(n_iter): shuffle(dataset) for x in X: update() ``` It also wastes a little bit of time in each update, checking whether it should...

Ack. Patches are appreciated.

`partial_fit` is a method that some scikit-learn estimators have (traditionally `SGDClassifier` and `MiniBatchKMeans`, since 0.14 also Naive Bayes). They allow incremental learning: after each `partial_fit`, you have a functioning estimator/model,...

Sweet! As for the time, have you tried profiling with [`kernprof.py`](http://pythonhosted.org/line_profiler/)? I suspect calling `logsumexp` from inside Cython might have to with this, too.