@cdeil, since you looked into the rotation issue, do you agree with this?

Here's one way you can show that the current rotation is wrong:
http://nbviewer.ipython.org/gist/larrybradley/9808977

Ok, that looks wrong :) Could you add a regression test? Maybe you could do it with a simple low-res 5-5 array?

@astrofrog I've added the test, updated the change log, and rebased.

I agree this fix is correct to be consistent with the definition in Gaussian2D

The rotation angle increases clockwise.

It took me a while to convince myself this fix is correct. It could be useful to add in the notes section explaining the relation of the covariance matrix to the formulas given, namely that cov_matrix = [[a, b], [b, c]] and maybe even that positive b corresponds to a theta in the range 0 to 90 deg. (if that is correct).

But ... :-)

Should we re-consider the choice of rotation angle direction?
The other tools I use (sherpa gauss2d and SExtractor both use a rotation angle that increases counter-clockwise.
I think having the same convention as other common tools would be enough motivation to change.
@larrybradley @astrofrog What do you think? Are there other important tools that use rotation angles that increase clockwise?

I am in favor of consistently using counter-clockwise rotation (including RotateByAngle which does a clockwise rotation now) as this seems more intuitive to me. 👍 to changing all of them to counter-clockwise.

I am also in favor of consistently using counter-clockwise rotation. I never liked that it was clockwise. If @astrofrog concurs, then I'll change the rotation.

While we are discussing this, I'd like to bring up one other point. Currently there is a degeneracy in the model such that one can swap x_stddev and y_stddev and add 90 degrees to theta and get the same exact model. This could lead to some confusion when fitting the model. Does it make sense to perhaps change the inputs to something like major/minor axis stddev and then have theta represent the angle of the major axis (w.r.t to +x axis for example). There is still the 180 degree degeneracy, but nothing can be done about that. Thoughts?

Counter-clockwise makes sense!

OK, I've changed it to counterclockwise rotation. This should be ready to merge.

I'll update the MatrixRotation2D model next.

👍 to merge.

Any clarifications on the addition of this change to the 0.3.2 milestone? The changelog entry is under 0.4. Also as this constitutes a major change in behavior I would probably leave in as 0.4 anyway.