Last year I worked on integrating some statistics code with a GIS. After looking at as many open-source cross-platform GIS as I could find, I got stuck with OpenEV. It did the job - just. Its big attraction for me was the built-in python scripting and API - a python script could create new menu items, Gtk dialogs, access map data, create new layers and so on.

My code had to talk to R, a statistics package (www.r-project.org). I did it using Rserve - in this way, R opens a socket on localhost and listens for requests. There were no python bindings for Rserve so I dissected the protocol and wrote my own. Job pretty much done.

Then late last year I upgraded my desktop box to Ubuntu (from RH8). Upgrade time for R, OpenEV, python and so on. Plus the realisation that this code would all have to run on Windows, by non-geeks, in remote villages in Africa.

OpenEV can come as a binary distribution for Linux and Windows, complete with python. But that python didn't have the required files to add extra packages, so I couldn't use that. Plus I wanted to stop using Rserve (which is very clunky), and use Rpy, which links straight into R as a library. But the python in the OpenEV distribution had a thread issue that didn't let Rpy work. Compiling from source looked... messy.

So I decided that what with all those problems, plus the fact that the OpenEV interface is rather non-standard in several aspects of its behaviour, to switch camps and go looking for a new GIS again.

Main candidates were: QGis, Thuban and SAGA. I went for QGis since it was the only one I could get to work reliably. Even the bleeding edge 0.9.0 svn version compiled and I was running python plugins with Rpy straightaway. Woo hoo props to y'all.

If I was a java man I'd take a good look at OpenJUMP, which does some sweet things. But I'm not, so it had to be python, plus I had large chunks of the code done in python already.

I'm now well into the rewrite. The plugin doesn't do an awful lot: basically it registers as a plugin, then when activated it pops up a Qt dialog (made with designer-qt4), when OK'd, this gets some layers from the map and some other parameters from the dialog and runs some statistical methodology for predicting disease prevalence in a defined study area. The output is a grid which is saved as an ESRI raster file and is loaded in as a new map layer.

The plan is to get this all done and to make a demo (with Wink) to show to some WHO people in Geneva by the end of the week. I'm not going, only my .swf file is. Once that's done I'll try and post some more blog entries on technical aspects of writing this thing.

Thanks to Martin D and all on #qgis for help so far and help to come...

Barry

PS: apologies for posting www.doingitwrong.com on #qgis and the ensuing loss of productivity.....