.add()

The .add() method in PyTorch computes the element-wise sum of two tensors, enabling arithmetic operations even between tensors of different shapes through broadcasting. Broadcasting automatically expands tensors with mismatched dimensions to compatible shapes, following NumPy-style rules. This method is foundational for tasks like neural network weight updates, residual connections, and feature engineering.

Syntax

torch.add(input, other, alpha=1)

Parameters:

• input (Tensor): The first input tensor.
• other (Tensor or Number): The second tensor/scalar to add to input.
• alpha (Number, Optional): A multiplier applied to other before performing addition. Defaults to a value ``.

Return value:

It returns a new tensor containing the element-wise sum of input and other.

Example

This example demonstrates tensor addition with and without scaling:

import torch

# Define 2x2 tensors
a = torch.tensor([[1, 2], [3, 4]])
b = torch.tensor([[5, 6], [7, 8]])

# Element-wise addition (a + b)
c = torch.add(a, b)

# Scaled addition (a + 2 * b)
d = a.add(b, alpha=2)

print("Basic Addition:\n", c)

print("\nScaled Addition (alpha=2):\n", d)

to clipboard
Copy to clipboard

This example results in the following output:

Basic Addition:
 tensor([[ 6,  8],
        [10, 12]])

Scaled Addition (alpha=2):
 tensor([[11, 14],
        [17, 20]])

to clipboard
Copy to clipboard

In this example:

• Basic Addition: Each element in a is added to the corresponding element in b (e.g., 1 + 5 = 6, 2 + 6 = 8).
• Scaled Addition: The alpha=2 parameter scales b by 200% before adding it to a (e.g., 1 + (2 * 5) = 11, 2 + (2 * 6) = 14).

Additional Notes

• Broadcasting: Supports operations between tensors of different shapes by automatically expanding them to compatible dimensions (e.g., (3,1) + (1,4) → (3,4)).
• In-Place Operation: Use .add_() to modify the original tensor directly.
• Complex Numbers: Handles complex tensors by adding real and imaginary components separately.
• Device Consistency: Both tensors must be on the same device (CPU/GPU) for the operation to work.