Formatting 2D array with older version of NumPy

Question

I have a code snippet to format a 2D array to a specific string format. The code snippet works well with recent versions of NumPy (e.g. 1.9.2), but the same fails with NumPy 1.4.1, which is the current version for CentOS 6.6.

import numpy as np
cfmt = '%14.6E'
vfmt = np.vectorize(cfmt.__mod__)

Traceback (most recent call last):
  File "/usr/lib64/python2.6/site-packages/numpy/lib/function_base.py", line 1762, in __init__
    nin, ndefault = _get_nargs(pyfunc)
  File "/usr/lib64/python2.6/site-packages/numpy/lib/function_base.py", line 1694, in _get_nargs
    raise ValueError, 'failed to determine the number of arguments for %s' % (obj)
ValueError: failed to determine the number of arguments for <method-wrapper '__mod__' of str object at 0xb461e0>

From Numpy-discussion, this appears to be an issue with functools.partial, with a recommend workaround to use lambda. However, here are my failed attempts:

vfmt1 = np.vectorize(lambda x: cfmt.__mod__(x))
vfmt2 = np.vectorize(lambda x: cfmt % (x,), ['|S14'])  # attempt to specify otype

# A 2D array with shape (1, 3)
ar = np.array([[1.0e35, 9.999999, 10.0]])

print(vfmt1(ar))  # [['  1.0000' '  9.9999' '  1.0000']]
print(vfmt2(ar))  # [['  1.0000' '  9.9999' '  1.0000']]

These values appear to be strings clipped to 8 characters, and are essentially garbage.

The expected formatted result using a recent version of NumPy is:

array([['  1.000000E+35', '  9.999999E+00', '  1.000000E+01']], type='|S14')

Any suggestions on how to get the expected outputs using older versions of NumPy? Or generally how to vectorize formatting of numeric arrays to character arrays?

FYI: You are running into a bug that has been fixed in newer versions of numpy: https://github.com/numpy/numpy/issues/2485 — Warren Weckesser, Sep 28 '15 at 23:17
Thanks, however I require a solution for an older version of NumPy, since that is the supported version available to me. — Mike T, Sep 28 '15 at 23:20
Any `vectorize`d function that returns a string will result in a dtype of `S8`, giving you truncated strings. — Warren Weckesser, Sep 28 '15 at 23:23
@askewchan from Python 2.6.6 that style of formatting throws "ValueError: zero length field name in format" — Mike T, Sep 28 '15 at 23:26

Warren Weckesser · Accepted Answer · 2015-09-29T00:01:49.560

If your input array is one-dimensional, a simple list comprehension will work:

>>> ar
array([  1.00000000e+35,   9.99999900e+00,   1.00000000e+01])
>>> ar2 = np.array(["%14.6E" % v for v in ar])
>>> ar2
array(['  1.000000E+35', '  9.999999E+00', '  1.000000E+01'], 
      dtype='|S14')

For a 2-d array, a nested list comprehension will work. For example, here's a 2-d array x:

>>> x
array([[  1.00000000e+35,   9.99999900e+00,   1.00000000e+01],
       [  1.23450000e-04,   0.00000000e+00,   1.23000000e+02]])

Here's the nested list comprehension that creates a list of list of strings:

>>> [["%14.6E" % v for v in row] for row in x]
[['  1.000000E+35', '  9.999999E+00', '  1.000000E+01'], ['  1.234500E-04', '  0.000000E+00', '  1.230000E+02']]

If you need the result to be a numpy array, just put that in a call to np.array:

>>> np.array([["%14.6E" % v for v in row] for row in x])
array([['  1.000000E+35', '  9.999999E+00', '  1.000000E+01'],
       ['  1.234500E-04', '  0.000000E+00', '  1.230000E+02']], 
      dtype='|S14')

Formatting 2D array with older version of NumPy

1 Answers1