[Numpy-discussion] loadtxt improvement

[Neil]

I'd like to add a way for loadtxt to infer a dtype from the data it
reads in.

Record arrays/structured arrays are the best thing ever, and ideally
I'd like to read in a csv-style file into a structured array in one
easy step.  loadtxt almost does this - if I know the number and type
of fields before reading the file, I can specify the dtype keyword,
and loadtxt will give me a structured array.  But having to know the
dtype before you read a file (including the the required string
lengths!) is a real pain. It would be great if you could tell loadtxt
to read a file into a structured array and guess the dtype for each
field.

I've made some changes to  lib/io.py that does this by adding a
'names' keyword to loadtxt.  If a list of field names is given, then
loadtxt reads the file data into a structured array, trying int, float
and str types for each column and keeping whichever is suitable for
all the data in that column.

Does this sound like a good approach?