How to turn a 3d numpy array into a pandas dataframe of numpy 1d arrays?

Question

I have a numpy 3d array. I’d like to create a pandas dataframe off of it which would have as many rows as the array’s 1st dimension’s size and the number of columns would be the size of the 2nd dimension. The values of the dataframe should be the actual vectors (numpy arrays) with the size of the third dimension.

Like if I have this array of size (2, 3, 5):

>>> arr
array([[[ 1.,  1.,  1.,  1.,  1.],
        [ 0.,  0.,  0.,  0.,  0.],
        [ 1.,  2.,  3.,  4.,  6.]],

       [[ 0.,  0.,  0.,  0.,  0.],
        [11., 22., 33., 44., 66.],
        [ 0.,  0.,  0.,  0.,  0.]]])

I want to turn it into this dataframe (and do this efficiently with native numpy and/or pandas methods):

>>> df_arr
                        col0                            col1                       col2
0  [1.0, 1.0, 1.0, 1.0, 1.0]       [0.0, 0.0, 0.0, 0.0, 0.0]  [1.0, 2.0, 3.0, 4.0, 6.0]
1  [0.0, 0.0, 0.0, 0.0, 0.0]  [11.0, 22.0, 33.0, 44.0, 66.0]  [0.0, 0.0, 0.0, 0.0, 0.0]

Asked By: Sergey Zakharov

||

Source

Answer 1

pd.DataFrame.from_records works:

pd.DataFrame.from_records(arr, columns=['col0', 'col1', 'col2'])
                        col0  ...                       col2
0  [1.0, 1.0, 1.0, 1.0, 1.0]  ...  [1.0, 2.0, 3.0, 4.0, 6.0]
1  [0.0, 0.0, 0.0, 0.0, 0.0]  ...  [0.0, 0.0, 0.0, 0.0, 0.0]

[2 rows x 3 columns]

Answered By: Psidom

How to turn a 3d numpy array into a pandas dataframe of numpy 1d arrays?

Question:

Answers: