Re: [Matplotlib-users] plt.hist() doesn't recognize the masked array?

[Benjamin R. <ben...@ou...>]

On Mon, Jan 2, 2012 at 11:10 AM, Chao YUE <cha...@gm...> wrote:

> Dear all matplotlib users,
>
> Happy New Year.
> I try to check the distribution of a 2D array and I find that the
> histogram plot function doesn't respect the numpy masked array?
>
>
> In [188]: a=range(1,6); b=np.array(a+a[::-1])
>
> In [189]: b=np.ma.masked_equal(b,2); b=np.ma.masked_equal(b,5)
>
> In [190]: b
> Out[190]:
> masked_array(data = [1 -- 3 4 -- -- 4 3 -- 1],
>              mask = [False  True False False  True  True False False  True
> False],
>        fill_value = 5)
>
>
> In [191]: n,bins,patches=plt.hist(b)
>
> In [192]: n
> Out[192]: array([2, 0, 2, 0, 0, 2, 0, 2, 0, 2])
>
> In [193]: n.sum()
> Out[193]: 10
>
> it seems that all the elements (masked or not) are counted in the history
> plotting?
> and the original value is used but not the fill_value?
>
> I attach a figure below.
>
> In [194]: plt.show()
>
>
Yes, this is a known issue (at least, from the comments within the
function).  Looks like hist() uses np.asarray() instead of np.asanyarray(),
which would result in the array being stripped of its mask.  However, I
don't think the fix is as straight-forward as changing that to
np.asanyarray().  I will take a peek and see what can be done.

Ben Root