mean() and median() in PyTorch

By /

mean() can get the mean(average) values as shown below:

*Memos:

  • mean() can be called both from torch and a tensor.
  • The 2nd argument is one or more dimensions with torch.
  • The 1st argument is one or more dimensions with a tensor.
  • mean() can only accept floating-point or complex numbers so you need conversion to them if they are not as I explain it in my answer(17.1) otherwise there is the error.
import torch my_tensor = torch.tensor([[5, 4, 7, 7], [6, 5, 3, 5], [3, 8, 9, 3]])
torch.mean(my_tensor.float())tensor(5.4167)
my_tensor.float().mean()
# tensor(5.4167)

torch.mean(my_tensor.float(), 0)
my_tensor.float().mean(0)
torch.mean(my_tensor.float(), (0,))
my_tensor.float().mean((0,))
torch.mean(my_tensor.float(), -2)
my_tensor.float().mean(-2)
torch.mean(my_tensor.float(), (-2,))
my_tensor.float().mean((-2,))
# tensor([4.6667, 5.6667, 6.3333, 5.0000])

torch.mean(my_tensor.float(), 1)
my_tensor.float().mean(1)
torch.mean(my_tensor.float(), (1,))
my_tensor.float().mean((1,))
torch.mean(my_tensor.float(), -1)
my_tensor.float().mean(-1)
torch.mean(my_tensor.float(), (-1,))
my_tensor.float().mean((-1,))
# tensor([5.7500, 4.7500, 5.7500])

torch.mean(my_tensor.float(), (0, 1))
my_tensor.float().mean((0, 1))
torch.mean(my_tensor.float(), (0, -1))
my_tensor.float().mean((0, -1))
torch.mean(my_tensor.float(), (1, 0))
my_tensor.float().mean((1, 0))
torch.mean(my_tensor.float(), (1, -2))
my_tensor.float().mean((1, -2))
torch.mean(my_tensor.float(), (-1, 0))
my_tensor.float().mean((-1, 0))
torch.mean(my_tensor.float(), (-1, -2))
my_tensor.float().mean((-1, -2))
torch.mean(my_tensor.float(), (-2, 1))
my_tensor.float().mean((-2, 1))
torch.mean(my_tensor.float(), (-2, -1))
my_tensor.float().mean((-2, -1))
# tensor(5.4167)
Enter fullscreen mode
Exit fullscreen mode
import torch my_tensor = torch.tensor([[[0., 1., 2.], [3., 4., 5.]], [[6., 7., 8.], [9., 10., 11.]], [[12., 13., 14.], [15., 16., 17.]], [[18., 19., 20.], [21., 22., 23.]]])
torch.mean(my_tensor)
my_tensor.mean()
# tensor(11.5000)

torch.mean(my_tensor, 0)
my_tensor.mean(0)
torch.mean(my_tensor, (0,))
my_tensor.mean((0,))
# tensor([[9., 10., 11.], [12., 13., 14.]])

torch.mean(my_tensor, 1)
my_tensor.mean(1)
torch.mean(my_tensor, (1,))
my_tensor.mean((1,))
# tensor([[1.5000, 2.5000, 3.5000], [7.5000, 8.5000, 9.5000],
# [13.5000, 14.5000, 15.5000], [19.5000, 20.5000, 21.5000]])

torch.mean(my_tensor, 2)
torch.mean(my_tensor, (2,))
my_tensor.mean(2)
my_tensor.mean((2,))
torch.mean(my_tensor, -1)
my_tensor.mean(-1)
torch.mean(my_tensor, (-1,))
my_tensor.mean((-1,))
# tensor([[1., 4.], [7., 10.], [13., 16.], [19., 22.]])

torch.mean(my_tensor, -2)
my_tensor.mean(-2)
torch.mean(my_tensor, (-2,))
my_tensor.mean((-2,))
# tensor([[1.5000, 2.5000, 3.5000],
# [7.5000, 8.5000, 9.5000],
# [13.5000, 14.5000, 15.5000],
# [19.5000, 20.5000, 21.5000]])

torch.mean(my_tensor, -3)
my_tensor.mean(-3)
torch.mean(my_tensor, (-3,))
my_tensor.mean((-3,))
# tensor([[ 9., 10., 11.], [12., 13., 14.]])

torch.mean(my_tensor, (0, 1))
my_tensor.mean((0, 1))
torch.mean(my_tensor, (1, 0))
my_tensor.mean((1, 0))
torch.mean(my_tensor, (1, -3))
my_tensor.mean((1, -3))
torch.mean(my_tensor, (0, -2))
my_tensor.mean((0, -2))
torch.mean(my_tensor, (-2, 0))
my_tensor.mean((-2, 0))
torch.mean(my_tensor, (-2, -3))
my_tensor.mean((-2, -3))
torch.mean(my_tensor, (-3, 1))
my_tensor.mean((-3, 1))
torch.mean(my_tensor, (-3, -2))
my_tensor.mean((-3, -2))
# tensor([10.5000, 11.5000, 12.5000])

torch.mean(my_tensor, (0, 2))
my_tensor.mean((0, 2))
torch.mean(my_tensor, (0, -1))
my_tensor.mean((0, -1))
torch.mean(my_tensor, (2, 0))
my_tensor.mean((2, 0))
torch.mean(my_tensor, (2, -3))
my_tensor.mean((2, -3))
torch.mean(my_tensor, (-1, 0))
my_tensor.mean((-1, 0))
torch.mean(my_tensor, (-1, -3))
my_tensor.mean((-1, -3))
torch.mean(my_tensor, (-3, 2))
my_tensor.mean((-3, 2))
torch.mean(my_tensor, (-3, -1))
my_tensor.mean((-3, -1))
# tensor([10., 13.]) 
torch.mean(my_tensor, (1, 2))
my_tensor.mean((1, 2))
torch.mean(my_tensor, (1, -1))
my_tensor.mean((1, -1))
torch.mean(my_tensor, (2, 1))
my_tensor.mean((2, 1))
torch.mean(my_tensor, (2, -2))
my_tensor.mean((2, -2))
torch.mean(my_tensor, (-1, 1))
my_tensor.mean((-1, 1))
torch.mean(my_tensor, (-1, -2))
my_tensor.mean((-1, -2))
torch.mean(my_tensor, (-2, 2))
my_tensor.mean((-2, 2))
torch.mean(my_tensor, (-2, -1))
my_tensor.mean((-2, -1))
# tensor([2.5000, 8.5000, 14.5000, 20.5000])

torch.mean(my_tensor, (0, 1, 2))
my_tensor.mean((0, 1, 2))
etc.
# tensor(11.5000)
Enter fullscreen mode
Exit fullscreen mode

median() can get the median(average) values as shown below:

*Memos:

  • median() can be called both from torch and a tensor.
  • The 2nd argument is one or more dimensions with torch.
  • The 1st argument is one or more dimensions with a tensor.
import torch my_tensor = torch.tensor([[5, 4, 7, 7], [6, 5, 3, 5], [3, 8, 9, 3]])
torch.median(my_tensor)
my_tensor.median()
# tensor(5)

torch.median(my_tensor, 0)
my_tensor.median(0)
torch.median(my_tensor, -2)
my_tensor.median(-2)
# torch.return_types.median(
# values=tensor([5, 5, 7, 5]),
# indices=tensor([0, 1, 0, 1]))

torch.median(my_tensor, 1)
my_tensor.median(1)
torch.median(my_tensor, -1)
my_tensor.median(-1)
# torch.return_types.median(
# values=tensor([5, 5, 3]),
# indices=tensor([0, 1, 3]))
Enter fullscreen mode
Exit fullscreen mode
import torch my_tensor = torch.tensor([[[0., 1., 2.], [3., 4., 5.]], [[6., 7., 8.], [9., 10., 11.]], [[12., 13., 14.], [15., 16., 17.]], [[18., 19., 20.], [21., 22., 23.]]])
torch.median(my_tensor)
my_tensor.median()
# tensor(11.)

torch.median(my_tensor, 0)
my_tensor.median(0)
# torch.return_types.median(
# values=tensor([[6., 7., 8.], [9., 10., 11.]]),
# indices=tensor([[1, 1, 1], [1, 1, 1]]))

torch.median(my_tensor, 1)
my_tensor.median(1)
# torch.return_types.median(
# values=tensor([[0., 1., 2.], [6., 7., 8.],
# [12., 13., 14.], [18., 19., 20.]]),
# indices=tensor([[0, 0, 0], [0, 0, 0],
# [0, 0, 0], [0, 0, 0]]))

torch.median(my_tensor, 2)
my_tensor.median(2)
torch.median(my_tensor, -1)
my_tensor.median(-1)
# torch.return_types.median(
# values=tensor([[1., 4.], [7., 10.], [13., 16.], [19., 22.]]),
# indices=tensor([[1, 1], [1, 1], [1, 1], [1, 1]]))

torch.median(my_tensor, -2)
my_tensor.median(-2)
# torch.return_types.median(
# values=tensor([[0., 1., 2.], [6., 7., 8.],
# [12., 13., 14.], [18., 19., 20.]]),
# indices=tensor([[0, 0, 0], [0, 0, 0],
# [0, 0, 0], [0, 0, 0]]))

torch.median(my_tensor, -3)
my_tensor.median(-3)
# torch.return_types.median(
# values=tensor([[6., 7., 8.], [9., 10., 11.]]),
# indices=tensor([[1, 1, 1], [1, 1, 1]]))
Enter fullscreen mode
Exit fullscreen mode

Discover more from Coursity

Subscribe to get the latest posts sent to your email.

Leave a Comment

Your email address will not be published. Required fields are marked *

Discover more from Coursity

Subscribe now to keep reading and get access to the full archive.

Continue reading

Scroll to Top
Martins ad network.