creating a masked array from text fields

The numpy documentation shows an example of masking existing values with ma.masked a posteriori (after array creation), or creating a masked array from an list of what seem to be valid data types (integer if dtype=int). I am trying to read in data from a file (and requires some text manipulation) but at some point 开发者_JAVA百科I will have a list of lists (or tuples) containing strings from which I want to make a numeric (float) array.

An example of the data might be textdata='1\t2\t3\n4\t\t6' (typical flat text format after cleaning).

One problem I have is that missing values may be encoded as '', which when trying to convert to float using the dtype argument, will tell me

ValueError: setting an array element with a sequence. 

So I've created this function

def makemaskedarray(X,missing='',fillvalue='-999.',dtype=float):
    arr = lambda x: x==missing and fillvalue or x    
    mask = lambda x: x==missing and 1 or 0
    triple = dict(zip(('data','mask','dtype'),
                      zip(*[(map(arr,x),map(mask,x)) for x in X])+
                      [dtype]))
    return ma.array(**triple)

which seems to do the trick:

>>> makemaskedarray([('1','2','3'),('4','','6')])
masked_array(data =
 [[1.0 2.0 3.0]
 [4.0 -- 6.0]],
             mask =
 [[False False False]
 [False  True False]],
       fill_value = 1e+20)

Is this the way to do it? Or there is a built-in function?

The way you're doing it is fine. (though you could definitely make it a bit more readable by avoiding building the temporary "triple" dict, just to expand it a step later, i.m.o.)

The built-in way is to use numpy.genfromtxt. Depending on the amount of pre-processing you need to do to your text file, it may or may not do what you need. However, as a basic example: (Using StringIO to simulate a file...)

from StringIO import StringIO
import numpy as np

txt_data = """
1\t2\t3
4\t\t6
7t\8t\9"""

infile = StringIO(txt_data)
data = np.genfromtxt(infile, usemask=True, delimiter='\t')

Which yields:

masked_array(data =
 [[1.0 2.0 3.0]
 [4.0 -- 6.0]
 [7.0 8.0 9.0]],
             mask =
 [[False False False]
 [False  True False]
 [False False False]],
       fill_value = 1e+20)

One word of caution: If you do use tabs as your delimiter and an empty string as your missing value marker, you'll have issues with missing values at the start of a line. (genfromtxt essentially calls line.strip().split(delimiter)). You'd be better off using something like "xxx" as a marker for missing values, if you can.