Convert numpy array with indices to a pandas dataframe

Question

I have a numpy array which I want to print with python ggplot's tile. For that I need to have a DataFrame with the columns x, y, value. How can I transform the numpy array efficiently into such a DataFrame. Please consider, that the form of the data I want is in a sparse style, but I want a regular DataFrame. I tried using scipy sparse data structures like in Convert sparse matrix (csc_matrix) to pandas dataframe, but conversions were too slow and memory hungry: My memory was used up.

To clarify what I want:

I start out with a numpy array like

array([[ 1,  3,  7],
       [ 4,  9,  8]])

and I would like to end up with the DataFrame

     x    y    value
0    0    0    1
1    0    1    3
2    0    2    7
3    1    0    4
4    1    1    9
5    1    2    8

cs95 · Accepted Answer · 2017-08-25T10:07:56.230

2

arr = np.array([[1, 3, 7],
                [4, 9, 8]])

df = pd.DataFrame(np.hstack((np.indices(arr.shape).reshape(2, arr.size).T,\
                    arr.reshape(-1, 1))), columns=['x', 'y', 'value'])
print(df)

   x  y  value
0  0  0      1
1  0  1      3
2  0  2      7
3  1  0      4
4  1  1      9
5  1  2      8

You might also consider using the function employed in this answer, as a speedup to np.indices in the solution above:

def indices_merged_arr(arr):
    m,n = arr.shape
    I,J = np.ogrid[:m,:n]
    out = np.empty((m,n,3), dtype=arr.dtype)
    out[...,0] = I
    out[...,1] = J
    out[...,2] = arr
    out.shape = (-1,3)
    return out

array = np.array([[ 1,  3,  7],
                  [ 4,  9,  8]])

df = pd.DataFrame(indices_merged_arr(array), columns=['x', 'y', 'value'])
print(df)

   x  y  value
0  0  0      1
1  0  1      3
2  0  2      7
3  1  0      4
4  1  1      9
5  1  2      8

Performance

arr = np.random.randn(1000, 1000)

%timeit df = pd.DataFrame(np.hstack((np.indices(arr.shape).reshape(2, arr.size).T,\
                         arr.reshape(-1, 1))), columns=['x', 'y', 'value'])
100 loops, best of 3: 15.3 ms per loop

%timeit pd.DataFrame(indices_merged_arr(array), columns=['x', 'y', 'value'])
1000 loops, best of 3: 229 µs per loop

edited Aug 25 '17 at 10:07

answered Aug 24 '17 at 08:04

cs95

379,657
97
704
746

I tried to clarify what I want in the question. I am not sure how your answer is helping. – Make42 Aug 24 '17 at 08:53
Your first answer works, your second answer throws `ValueError: Shape of passed values is (3, 6), indices imply (3, 3)`. – Make42 Aug 24 '17 at 11:33
@Make42 Copy paste error. Had to include `columns=`. It works. – cs95 Aug 24 '17 at 11:34
@Make42 Interestingly, speed wise they're almost the same. It's a matter of preference as to what you'd want to use. – cs95 Aug 25 '17 at 10:05
But aren't they doing exactly the same? So what has this to do with preferences? – Make42 Aug 25 '17 at 10:06
@Make42 Nevermind. The second is much faster on large data (check performance stats). When I said preference I meant wrt how much code you like writing. – cs95 Aug 25 '17 at 10:08
@Make42 Yes, sorry. – cs95 Aug 25 '17 at 10:12

score 1 · Answer 2 · answered Mar 11 '22 at 17:46

You can try this solution by using np.ndenumerate:

arr = np.array([[1, 3, 7],
                [4, 9, 8]])

df = pd.DataFrame(np.ndenumerate(arr), columns=["coord","val"])

df[["x","y"]]  = df["coord"].tolist()

df.drop('coord', 1, inplace=True)

df = df[["x","y","val"]]

output

Convert numpy array with indices to a pandas dataframe

2 Answers2