I have a numpy 2d array A, and a list of row numbers row_set. How can I get new array B such as if row_set = [0, 2, 5], then B = [A_row[0], A_row[2], A_row[5]]?
I thought of something like this:
def slice_matrix(A, row_set):
slice = array([row for row in A if row_num in row_set])
but I don't have any idea, how can I get a row_num.
Use take():
In [87]: m = np.random.random((6, 2))
In [88]: m
Out[88]:
array([[ 0.6641412 , 0.31556053],
[ 0.11480163, 0.00143887],
[ 0.4677745 , 0.43055324],
[ 0.49749099, 0.15678506],
[ 0.48024596, 0.65701218],
[ 0.48952677, 0.97089177]])
In [89]: m.take([0, 2, 5], axis=0)
Out[89]:
array([[ 0.6641412 , 0.31556053],
[ 0.4677745 , 0.43055324],
[ 0.48952677, 0.97089177]])
You can pass a list or an array as indexes to any np array.
>>> r = np.random.randint(0,10,(5,5))
>>> r
array([[3, 8, 9, 8, 4],
[4, 1, 5, 9, 1],
[3, 6, 8, 8, 0],
[5, 1, 7, 6, 1],
[6, 1, 7, 7, 7]])
>>> idx = [0,3,1]
>>> r[idx]
array([[3, 8, 9, 8, 4],
[5, 1, 7, 6, 1],
[4, 1, 5, 9, 1]])
Speed comparison: take() is faster.
In [1]: m = np.random.random((1000, 2))
i = np.random.randint(1000, size=500)
%timeit m[i]
Out[1]:
10000 loops, best of 3: 27.2 µs per loop
In [2]: %timeit m.take(i, axis=0)
Out[2]:
100000 loops, best of 3: 7.24 µs per loop
This remains true for very large m and i
