Perhaps you would be interested in MaskedArrays? Take a peek at
np.ma.masked_invalid().

Cheers!
Ben Root
On Jun 27, 2013 8:29 AM, "mmokrejs" [email protected] wrote:

I was annoyed by numpy giving me sometimes this weird message. It turned
out to be related to average() while not to median(). I see a pull #961https://github.com/matplotlib/matplotlib/issues/961but containing same warning messages in comment from 'jdh2358'.

Anyway, here is my testcase:

$ python
Python 2.7.3 (default, Apr 20 2013, 18:28:22)
[GCC 4.6.3] on linux2
Type "help", "copyright", "credits" or "license" for more information.

import numpy as np
np.average([])
/usr/lib64/python2.7/site-packages/numpy/core/_methods.py:57:
RuntimeWarning: invalid value encountered in double_scalars
ret = ret / float(rcount)
nan
np.median([])
nan

Sure, sometimes upstream code passes an empty list but I believe that is
perfectly valid requirement. Please make your code more defensive. ;-)

Personally am not happy at all with those NaN values but if you insist,
return NaN in case of average() like you already do for median(). I zap
them myself with None.

A more general note on why I don't like NaN and how I got into the issue:

def fix_clean_lists(indexes, func, _in, _out):
"""Overwrite NaN values (numpy.float64) with None or keep the float for
average or round the float to integer for medians.
"""

if _in:
if func == 'median':
for _indexpos in indexes:
if _in[_indexpos]:
_v = np.median(_in[_indexpos])
try:
_rounded = int(_v)
except ValueError:
# get rid rid of NaN which is a numpy.float64
_out[_indexpos] = None
else:
_out[_indexpos] = _rounded
elif func == 'average':
for _indexpos in indexes:
if _in[_indexpos]:
# needed to get rid of:
# fix_clean_lists() entered here for func average for [[], [], [], [], []]
# /usr/lib64/python2.7/site-packages/numpy/core/_methods.py:57: RuntimeWarning: invalid value encountered in double_scalars
# ret = ret / float(rcount)

            _v = np.average(_in[_indexpos])
            try:
                _avg = int(_v)
            except ValueError:
                # get rid rid of NaN which is a numpy.float64
                _out[_indexpos] = None
            else:
                _out[_indexpos] = round(_v, 2)

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/2157
.

Closing as this should have been reported to numpy, not to mpl.