On Mon, Oct 8, 2012 at 1:38 PM, Mathieu Blondel <mathieu@mblondel.org>wrote:

On Tue, Oct 9, 2012 at 2:11 AM, Johannes Schönberger < hannesschoenberger@gmail.com> wrote:

...

I'm not sure if everyone has already noticed, that there is support for "fused types" in the recently released Cython 0.17 ( http://docs.cython.org/src/userguide/fusedtypes.html).

Fused types were actually introduced in v0.16 ( http://wiki.cython.org/ReleaseNotes-0.16) so v0.17 is the second release supporting them.

Mathieu

--

I actually tried to bump up the Cython requirement to 0.16 (see https://groups.google.com/forum/?fromgroups=#!topic/scikit-image/Fzdtoo8AUZ0) about a month ago, but that turned out to be **way** too soon. For example, the current Ubuntu, 12.04, ships with Cython 0.15. I guess Ubuntu 12.10 will be released soon, and that release includes Cython 0.16. That said, I'm not sure if targeting the most recent Ubuntu is what we want to do (since it will be too fast for some people). Andreas or Mathieu: Does scikit-learn have a policy for upgrading dependencies (in particular Cython)? I'd lean toward more conservative versions of numpy, but more recent versions of Cython. I'm not sure how that would actually translate into hard numbers. Maybe numpy matches latest Ubuntu LTS, cython matches latest Ubuntu? (I don't run Ubuntu, btw, I just find it a convenient reference.) Or maybe some predetermined time delay after those releases? Best, -Tony