Writing similar pandas function on numpy

Question

When I am using pandas dataframe, I usually need to do this operation:

x.eq(0).sum()/((x.eq(0)&x.shift().ne(0)).sum())

However, I need to work with arrays, and perform this same function.

I noticed that:

x.eq(0).sum()

Can be written like this in numpy:

np.count_nonzero(x == 0)

This second part ((x.eq(0)&x.shift().ne(0)).sum()), I don't know which numpy function substitutes shift().

How to write this function ((x.eq(0)&x.shift().ne(0)).sum()) on numpy ?

https://stackoverflow.com/questions/30399534/shift-elements-in-a-numpy-array — BENY, Apr 03 '20 at 13:45

Bruno Mello · Accepted Answer · 2020-04-03T13:56:52.053

IIUC:

from scipy.ndimage.interpolation import shift

x.eq(0) -> x==0
x.shift() -> shift(x, 1, cval=1)
x.ne(0) -> x!=0

And thus:

((x.eq(0)&x.shift().ne(0)).sum()) -> np.sum(np.logical_and((x==0),shift(x, 1, cval=np.nan)!=0))

Finally:

x.eq(0).sum()/((x.eq(0)&x.shift().ne(0)).sum()) -> np.count_nonzero(x == 0)/np.sum(np.logical_and((x==0),shift(x, 1, cval=1)!=0))

But if I where you, I would just do:

x = pd.Series(x)

1 Answers1