Numpy index slice with None

Question

Working with a sliding window for a numpy example. I tried to understand ,None of start_idx = np.arange(B[0])[:,None]

 foo = np.arange(10) print foo print foo[:] print foo[:,] print foo[:,None] 

The effect of None seems to be to transpose the array.

 [0 1 2 3 4 5 6 7 8 9] [0 1 2 3 4 5 6 7 8 9] [0 1 2 3 4 5 6 7 8 9] [[0] [1] [2] [3] [4] [5] [6] [7] [8] [9]] 

But I'm not quite sure. I could not find the documentation that explains what the second parameter does ( None ). This is also a tough piece for Google. numpy array docs makes me think this has something to do with advanced indexing , but I'm not sure.

Answer 1

foo[:, None] extends 1 dimensional array foo in 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 to transpose?

 print(foo.T) [1 2] 

Same as array itself

 print(foo.T == foo) [ True True] 

This restriction has many implications, and it becomes useful to consider foo in a 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 because it is less than characters and means the same thing

 print(foo[None, :]) [[1 2]] 

---

Ok, let's see what else we could do. Notice that I used the None example in the first position, while the OP used the second position. This position determines which dimension is expanded. And we could do it further. May 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 matters when numpy prints an 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]]] ↓