NumPy: 1D numpy array to display list
How can I extract the elements of a list that match the indices contained in 1D numpy.ndarray
?
Here's an example:
list_data = list(range(1, 100))
arr_index = np.asarray([18, 55, 22])
arr_index.shape
list_data[arr_index] # FAILS
I want to be able to retrieve items list_data
that match arr_index
.
source to share
You can use numpy.take
-
import numpy as np
np.take(list_data,arr_index)
Example run -
In [12]: list_data = list(range(1, 20))
In [13]: list_data
Out[13]: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19]
In [14]: arr_index = np.asarray([3, 5, 12])
In [15]: np.take(list_data,arr_index)
Out[15]: array([ 4, 6, 13])
source to share
OR
import numpy as np
list_data = list(range(1, 100))
arr_index = np.asarray([18, 55, 22])
arr_index.shape
new_ = [list_data[i] for i in arr_index]
>> [19, 56, 23]
Note
list_data = list(range(1, 100)
can be replaced with list_data = range(1, 100)
arr_index = np.asarray([18, 55, 22])
can be replaced with arr_index = np.array([18, 55, 22])
source to share
I just ran some time tests:
In [226]: ll=list(range(20000))
In [227]: ind=np.random.randint(0,20000,200)
In [228]: timeit np.array(ll)[ind]
100 loops, best of 3: 3.29 ms per loop
In [229]: timeit np.take(ll,ind)
100 loops, best of 3: 3.34 ms per loop
In [230]: timeit [ll[x] for x in ind]
10000 loops, best of 3: 65.1 µs per loop
In [231]: arr=np.array(ll)
In [232]: timeit arr[ind]
100000 loops, best of 3: 6 µs per loop
It's clear that list comprehension is the winner. Indexing an array is clearly faster, but the overhead of creating that array is substantial.
Converting to an array of dtype objects is faster. I'm a little surprised, but it must be because it can convert indiscriminately:
In [236]: timeit np.array(ll,dtype=object)[ind].tolist()
1000 loops, best of 3: 1.04 ms per loop
source to share