Why not CDAT?

[DamienIrving]

In your main README file you ask the questions "Why not Pandas?" and "Why not Iris?" Another question you might want to ask is "Why not CDAT?" It was written quite a long time ago now but is still used extensively by UV-CDAT and thus by the ESGF. In particular, the cdms2, cdutil, genutil and MV2 libraries within CDAT do some of what xray does.

[shoyer]

Thanks for asking!

Looking at CDAT, it does indeed appear that it has objects which are very similar in spirit to the xray Dataset and DataArray. I think some of the major distinguishing features of xray would be:

1. Design:
   • xray is targeted at a broader audience: anyone who needs a labeled, multi-dimensional array. I would like to avoid tight coupling to any particular domain, and keep xray as a more generic analysis tool for working with labeled N-dimensional arrays.
   • xray.DataArray is not a numpy.ndarray, unlike cmds2.tvariable.TransientVariable. This makes the design cleaner, and more flexible -- we can really do whatever we want with the array behind the scenes. In contrast, subclassing numpy arrays is not very reliable or predictable (in my experience).
2. Performance:
   • xray represents missing values by NaN (like pandas) in a numpy.ndarray, instead using a numpy.ma.MaskedArray. MaskedArray is written in pure python, so it's a far slower than using the standard ndarray.
   • xray indexes coordinate labels, and can use them for fast lookups, array assignment and alignment, all based on the pandas, with very minimal overhead.
3. Legal:
   • CDAT is described as "public domain software with unrestricted use" on its website, but the actual license sure doesn't look like that to me. In contrast, xray has a permissive open source license.

However, I'm sure that CDAT has some useful features and designs. If there are any particular aspects that you particular appreciate and think might belong in xray, I would be very interested to hear about them.

Note: I found the source code for UV-CDAT on GitHub: https://github.com/UV-CDAT/uvcdat

[shoyer]

going to close this since we now mention CDAT in our FAQ.

[shoyer]

Note: looks like I may be wrong about (most of) the legal complications -- UV-CDAT now says its available under the GPL: http://uvcdat.llnl.gov/installing.html