SAS Transport FIle

[Tim Churches]

Jeff wrote:
> Does anyone know of a python module to read a SAS Transport (XPORT)
> file?

I am not aware of any native Python module to read SAS portable
transport (XPORT) files (but that doesn't mean you shouldn't search for
one using Google etc). However, the format of the XPORT files is
publically available (see
http://support.sas.com/techsup/technote/ts140.html )and I would be happy
to help test such a module should you or anyone else care to write one.
Be aware that the XPORT file format harks back to the days when SAS was
only available on IBM mainframes, and teh specs contain some
peculiarities for a post-mainframe world, such as the use of IBM double
precision representation of floats (7 bits for the exponent and 56 bits
for the mantissa, whereas IEEE
format uses 11 bits for the exponent and 52 bits for the mantissa) -
thus you have to take additional care if you want to retain the full
precision expressible in the XPORT format.

There are alternative routes, though.

1) The foreign package (see
http://cran.r-project.org/doc/packages/foreign.pdf ) for the R
statistical system (see http://www.r-project.org) contains a function
read.xport which does as its name suggests. You can then use RPy (see
http://rpy.sf.net) to convert the resulting R data object(s) into Python
objects. Obviously you need to install R and RPy (and ideally Numeric
Python too) for this to work, but they are worth installing in their own
right. If your XPORT daatset is very large, then R/RPy may not be an
option, as R stores everything in memory.

2) Recent (within the last 6 years or so) versions of SAS come with an
XML engine, and it easy to export a SAS dataset to an XML file, which
would be easier to parse and convert into Python objects than teh SAS
XPORT file format, I dare say. The DTD used by SAS is also publicly
available (Google for it)

3) Recent versions of SAS also come with PROC EXPORT, which you can use
to dump you SAS dataset to a CSV or tab-delimited (or fixed format) text
file - but be careful with representation and precision issues. You can
write a SAS data step programme to accomplish the same thing in earlier
versions of SAS (in fact, that's all that PROC EXPORT does, behind the
scenes).

Tim C