Pandas : transform a 2D dataframe to a 3D one

user_4180859 0
Viewed 611

I have a Pandas dataframe like :

     p1 p2 p3
t1   a  b  c
t2   d  e  f
t3   g  h  i

and I want either several df (or one xarray), one for each t(y axis) like:

for t1: the value of pn minus pn for t1 :

     p1    p2   p3
p1   a-a  b-a  c-a 
p2   a-b  b-b  c-b
p3   a-c  b-c  c-c

idem for t2, t3 and so on...

I have made a loop like

for t in tlist :
    for p in plist :
        for q in plist :
            res = p - q

but it's hugly and slow.

So if anyone know a pythonic way to do that, [s]he's very welcome !

python-xarray pandas python
2 Answers

This can be done with broadcasting:

df = pd.DataFrame([[0,1,2],[3,6,8]])

# update as commented by piRSquared
a = df.to_numpy()    # use a=df.values if on Pandas < 0.24

a  = (a[:, None, :] - a[:,:,None]).reshape(-1, df.shape[1])

idx = pd.MultiIndex.from_product((df.index,df.columns), names=('t','p'))
pd.DataFrame(a, index=idx, columns=df.columns)

Output:

     0  1  2
t p         
0 0  0  1  2
  1 -1  0  1
  2 -2 -1  0
1 0  0  3  5
  1 -3  0  2
  2 -5 -2  0
user_42384080

This can be done in a quite simple manner leveraging numpy's broadcasting. Let's try with the following ndarray:

a = np.random.randint(1,10,(3,3))

print(a)
array([[9, 6, 4],
       [2, 3, 6],
       [8, 9, 2]])

a[:,None] - a[...,None]

array([[[ 0, -3, -5],
        [ 3,  0, -2],
        [ 5,  2,  0]],

       [[ 0,  1,  4],
        [-1,  0,  3],
        [-4, -3,  0]],

       [[ 0,  1, -6],
        [-1,  0, -7],
        [ 6,  7,  0]]])

This works by adding new axes to the arrays in such a way, that subtracting them yields the desired cartesian operation, since:

print(a[:,None])

array([[[9, 6, 4]],

       [[2, 3, 6]],

       [[8, 9, 2]]])

print(a[...,None])

array([[[9],
        [6],
        [4]],

       [[2],
        [3],
        [6]],

       [[8],
        [9],
        [2]]])
user_96986840
Related