python numpy 切片表示法（COMMA VS STANDARD INDEX）

Question

Is there a performance difference between using a comma and explicitly exploding out the index references for perhaps more conventional readers? Since both seem to yield the same results but the latter may be more intuitive to some

x = numpy.array([[1,2,3,4],
                 [5,6,7,8]])

comma_method = x[0,1:3] 
>>> numpy.array([2,3])

conventional method = x[0][1:3]
>>> numpy.array([2,3])

Answer 1

Pretty much always go for the comma, not for performance reasons, but because indexing twice isn't quite equivalent:

In [2]: x = numpy.array([[0, 1], [2, 3]])

In [3]: x[:1, :1]
Out[3]: array([[0]])

In [4]: x[:1][:1]
Out[4]: array([[0, 1]])

That said, the comma also appears to have a speed advantage:

In [7]: %timeit x[0][0]
The slowest run took 25.41 times longer than the fastest. This could mean that a
n intermediate result is being cached 
1000000 loops, best of 3: 357 ns per loop

In [8]: %timeit x[0, 0]
The slowest run took 41.92 times longer than the fastest. This could mean that a
n intermediate result is being cached 
1000000 loops, best of 3: 148 ns per loop

I'm not sure what's up with the slowest run and the fastest run having such a time difference.

Answer 2

第二种情况效率较低，因为在随后索引的第一个索引之后创建了一个新的临时数组。