PyTorch .fmod()

In PyTorch, the .fmod() method performs element-wise modulo operation on a tensor and returns a new tensor containing the remainders. The sign of the result always matches the sign of the dividend (the input tensor elements).

Mathematically, for each element in the input tensor, the operation follows:

$$ \text{result} = \text{input} - (\text{divisor} \times \text{trunc}(\text{input}/\text{divisor})) $$

Here, trunc means truncation toward zero (i.e., rounding towards 0).

Syntax

torch.fmod(input, other, *, out=None) -> Tensor

Parameters:

• input (Tensor): Dividend tensor.
• other (Tensor or Scalar): Divisor, can be a tensor (broadcastable with input) or a scalar.
• out (Tensor, optional): Output tensor to store the result.

Return value:

• Returns a tensor containing the element-wise remainder of division, where the result has the same sign as input.

Example

This example demonstrates the usage of the .fmod() function:

import torch

# Scalar divisor
x = torch.tensor([5.0, -3.5, 2.1])
result = x.fmod(2)  # or torch.fmod(x, 2)
print(result)

# Tensor divisor
x = torch.tensor([5.0, -3.5, 2.1])
y = torch.tensor([3.0, 2.0, 1.5])
result = x.fmod(y)
print(result)

# In-place operation
x = torch.tensor([5.0, -3.5, 2.1])
x.fmod_(2)  # x is modified directly
print(x)

Copy to clipboard
Copy to clipboard

The output of this code is:

tensor([ 1.0000, -1.5000,  0.1000])
tensor([ 2.0000, -1.5000,  0.6000])
tensor([ 1.0000, -1.5000,  0.1000])

Copy to clipboard
Copy to clipboard