问题
I found a very similar question to mine, but not exactly the same. This one: here However in ntimes's case the size of the array matches the number of the dimensions the tuple is point at. In my case I have a 4-dimensional array and a 2-dimensional tuple, just like this:
from numpy.random import rand
big_array=rand(3,3,4,5)
tup=(2,2)
I want to use the tuple as an index to the first two dimensions, and manually index the last two. Something like:
big_array[tup,3,2]
However, I obtain a repetition of the first dimension with index=2, along the fourth dimension( since it technically hasn't been indexed). That is because this indexing is interpreting a double indexing to the first dimension instead of one value for each dimension,
eg.
| dim 0:(index 2 AND index 2) , dim 1:(index 3), dim 2:(index 2), dim 3:(no index)|
instead of
|dim 0(index 2), dim 1(index 2), dim 2:(index 3), dim 3:(index 2)|.
How can I 'unpack' this tuple then? Any ideas? thanks!
回答1:
You can also pass in your first tuple alone to get the slice of interest, then index it seprately:
from numpy.random import rand
big_array=rand(3,3,4,5)
chosen_slice = (2,2)
>>> big_array[ chosen_slice ]
array([[ 0.96281602, 0.38296561, 0.59362615, 0.74032818, 0.88169483],
[ 0.54893771, 0.33640089, 0.53352849, 0.75534718, 0.38815883],
[ 0.85247424, 0.9441886 , 0.74682007, 0.87371017, 0.68644639],
[ 0.52858188, 0.74717948, 0.76120181, 0.08314177, 0.99557654]])
>>> chosen_part = (1,1)
>>> big_array[ chosen_slice ][ chosen_part ]
0.33640088565877657
That may be slightly more readable for some users, but otherwise I'd lean towards mgilson's solution.
回答2:
Since you're using numpy:
big_array[tup+(3,2)]
should work. When you call __getitem__ (via the square brackets), the stuff is passed to __getitem__ as a tuple. You just need to construct the tuple explicitly here (adding tuples together concatenates into a new tuple) and numpy will do what you want.
来源：https://stackoverflow.com/questions/12265751/tuple-as-index-of-multidimensional-array