WIP: Headbang re-work

[thequackdaddy]

Hello,

If I understand correctly, the headbang triples spatial smoothing is not beign supported because the current one doesn't work too well in python 3. After reviewing, I think I may have stumbled across a few bugs that make getting the old unittests to pass very difficult.

I decided to re-write it. Please let me know if this is something you'd consider replacing. Right now, I've saved this in another module (it is easier that way because i work with python 3 and the 2to3 process makes it hard to save a diff only for the changing part of the file. I'd incorporate this in pysal.esda.smoothing.

I've followed the algorithm described in this paper.

Right now, I've only re-written the Headbanging_Triples algorithm. I haven't experimented with the Headbanging_Median_Rate class yet. (It may be we can fully recycle that.)

Please let me know if this is something of interest to this project. Happy to finish it, or close this PR if you don't want to maintain.

Right now, I'm missing docstrings (I probably could just copy from the old method... but there's a few small simplifications to the API I woudl argue for... mainly dropping the requirement for data and k as they are already provided by the spatial weight w.

Also, the current unittests use the old nose template. There's a lot happening in the setUp function that doesn't really do anything.

[sjsrey]

@thequackdaddy thanks for this. certainly it is of interest if you want to finish it we will merge. modernizing/pruning the tests would be very helpful.

[thequackdaddy]

@sjsrey I would be happy to continue. I'd likely need a rather thorough review of this... there's lots of potential assumptions and not extremely thorough examples... So its highly likely what I do will be somewhat incorrect. Not having a good worked example makes writing unittests challenging.

When I get some time, I'll experiment some more with it shortly. If you have other resources/documentation, please let me know.

[sjsrey]

Here is some more documentation: https://surveillance.cancer.gov/headbang/

[sjsrey]

@thequackdaddy this is looking good - i will get a closer review in the coming days. One thing that comes to mind is whether we can assume k from the w argument. For triples (the logic of headbanging) we need k>=3 and for some w objects this might not hold for all i. On the other hand, for i with k_i<3 we could go the extrapolation route, so maybe your idea of dropping k holds? Food for thought.