foo[:, None]
extends the 1 dimensional array foo
into the second dimension. In fact, numpy
uses the alias np.newaxis
to do this.
consider foo
foo = np.array([1, 2])
print(foo)
[1 2]
A one dimensional array has limitations. For example, what's the transpose?
print(foo.T)
[1 2]
The same as the array itself
print(foo.T == foo)
[ True True]
This limitation has many implications and it becomes useful to consider foo
in higher dimensional context. numpy uses np.newaxis
print(foo[np.newaxis, :])
[[1 2]]
But this np.newaxis
is just syntactic sugar for None
np.newaxis is None
True
So, often we use None
instead because it's less characters and means the same thing
print(foo[None, :])
[[1 2]]
Ok, let's see what else we could've done. Notice I used the example with None
in the first position while OP use the second position. This position specifies which dimension is extended. And we could've taken that further. Let these examples help explain
print(foo[None, :]) # same as foo.reshape(1, 2)
[[1 2]]
print(foo[:, None]) # same as foo.reshape(2, 1)
[[1]
[2]]
print(foo[None, None, :]) # same as foo.reshape(1, 1, 2)
[[[1 2]]]
print(foo[None, :, None]) # same as foo.reshape(1, 2, 1)
[[[1]
[2]]]
print(foo[:, None, None]) # same as foo.reshape(2, 1, 1)
[[[1]]
[[2]]]
Keep in mind which dimension is which when numpy prints the array
print(np.arange(27).reshape(3, 3, 3))
dim2
────────?
dim0 → [[[ 0 1 2] │ dim1
[ 3 4 5] │
[ 6 7 8]] ↓
────────?
→ [[ 9 10 11] │
[12 13 14] │
[15 16 17]] ↓
────────?
→ [[18 19 20] │
[21 22 23] │
[24 25 26]]] ↓