.scatter()

In PyTorch, the .scatter() function writes values from a source (tensor or scalar) into specific locations of a tensor along a specified dimension, based on given indices.

Syntax

torch.scatter(ten, dim, index, src)

• ten: The tensor where the values are to be inserted.
• dim: The dimension along which the values are to be inserted.
• index: The tensor which specifies the locations in ten where the values are to be inserted.
• src: The tensor which contains the values to be inserted.

Example

The following example demonstrates the usage of the .scatter() function:

import torch

# Create a tensor
ten = torch.tensor([[11, 12, 13, 14, 15], [16, 17, 18, 19, 20]])

# Create a tensor containing the locations
index = torch.tensor([[0, 2], [1, 3]])

# Create a tensor containing the values
src = torch.tensor([[21, 23], [27, 29]])

# Insert the given values into specified locations along dimension 1 in the original tensor
res = torch.scatter(ten, 1, index, src)

# Print the resultant tensor
print(res)

The above code produces the following output:

tensor([[21, 12, 23, 14, 15],
        [16, 27, 18, 29, 20]])