PyTorch .nextafter()

The torch.nextafter() function returns the next representable floating-point value of each element in the input tensor in the direction of the corresponding element in the other tensor.

It’s useful for precise floating-point control, such as numerical stability adjustments or boundary value analysis.

Note: If an element in input equals the corresponding element in other, that element is returned unchanged in the output tensor.

Syntax

torch.nextafter(input, other, out=None)

Parameters:

• input (Tensor): The input tensor containing starting floating-point values.
• other (Tensor): The tensor indicates each element’s direction in input.
• out (Tensor, optional): The output tensor to store results. If not provided, a new tensor is returned.

Return value:

Returns a tensor of the same shape as input, where each element is the next representable floating-point value of input in the direction of other.

Example

The following example shows how torch.nextafter() moves each element of one tensor slightly closer to the corresponding element in another tensor:

import torch

# Create input tensors (float type)
input_tensor = torch.tensor([1.0, 5.0, -10.0, 8.0])
other_tensor = torch.tensor([2.0, 1.0, -11.0, 8.0])

# Calculate nextafter
result = torch.nextafter(input_tensor, other_tensor)

print("Input Tensor:")
print(input_tensor)

print("\nOther Tensor:")
print(other_tensor)

print("\nResult Tensor:")
print(result)

Copy to clipboard
Copy to clipboard

The output of the code is:

Input Tensor:
tensor([  1.,   5., -10.,   8.])

Other Tensor:
tensor([  2.,   1., -11.,   8.])

Result Tensor:
tensor([  1.0000,   5.0000, -10.0000,   8.0000])

Copy to clipboard
Copy to clipboard

In this:

• 1.0 becomes slightly larger since it moves toward 2.0.
• 5.0 becomes slightly smaller since it moves toward 1.0.
• -10.0 becomes slightly more negative since it moves toward -11.0.
• 8.0 remains unchanged since it already equals the corresponding value.