torch.min() is a PyTorch function that computes the minimum value of a given input tensor along a specified dimension.
The torch.min() function takes two arguments:
-
input: the input tensor to be reduced -
dim: the dimension along which to find the minimum value
For example, if input is a 2D tensor of shape (3, 4) and we want to find the minimum value along the second dimension, we can call torch.min(input, dim=1).
This will return a tuple of two tensors:
-
The first tensor will contain the minimum value of each row, resulting in a 1D tensor of shape
(3,). -
The second tensor will contain the index of the minimum value of each row, resulting in a 1D tensor of shape
(3,).
If we set keepdim=True, the output tensors will have the same number of dimensions as the input tensor, with the reduced dimension replaced by a singleton dimension. For example, if we call torch.min(input, dim=1, keepdim=True), the first output tensor will have shape (3, 1) and the second output tensor will have shape (3, 1).
The torch.min() function can also take multiple input tensors as arguments, in which case it will compute the element-wise minimum of the tensors along the specified dimension.