Home / Questions / 39070
Accepted
Jose Hermosilla Rodrigo
Asked: 2020-12-14 15:56:23 +0800 CST 2020-12-14 15:56:23 +0800 CST

How to insert rows to a numpy array?

  • 772

I have an array created with numpy:

matriz = numpy.matrix([]);

I try to insert rows in a loop (something like):

for i in range(3) : matriz = numpy.insert(matriz, i, [i + 1, 0, 0])

I expected the result to be:

matrix([[1, 0, 0], [2, 0, 0], [3, 0, 0]])

However I get:

matrix([[ 1.,  2.,  3.,  0.,  0.,  0.,  0.,  0.,  0.]])

What am I doing wrong?

python

1 Answers

  1. Best Answer

    This happens because you do not indicate the axis to which you want to add the data ( 0for filasand 1for columnas). If this parameter is not indicated the array/matrix is ​​flattened first as you can see in the documentation of numpy.insert():

    axis : int, optional

    Axis along which to insert values. If axis is None then arr is flattened first.

    The correct syntax would be:

    matriz=numpy.insert(matriz, i, [i + 1, 0, 0], axis=0)

    The problem is that you are trying to add a row with three columns to a 1x0 matrix. That throws an error. The solution may be to initialize a 3 row but empty array. For this we can use numpy.empty:

    import numpy

matriz =  numpy.matrix(numpy.empty(shape=(0,3), dtype=int))
for i in range(3):
    matriz=numpy.insert(matriz, i, [i + 1, 0, 0], axis = 0)

    Anyway there are more efficient ways to do this like:

    import numpy

matriz = numpy.matrix([[i+1, 0, 0] for i in range(3)], dtype=int)

    I specify the data type with dtypebecause if I don't, it will be the floatdefault.

    In both cases we get:

    matrix([[1, 0, 0],    
        [2, 0, 0],    
        [3, 0, 0]])

    Although the first method (using np.insert()) is much more inefficient than the second (list comprehensions), I get the following results for adding 100,000 rows:

    np.insert(): 9.337375402450562 seconds
    list comprehensions: 0.07504963874816895 seconds

    • 2