lucid.argmin¶

lucid.argmin(a: Tensor, axis: int | None = None, keepdims: bool = False) → Tensor¶

The argmin function returns the indices of the minimum values along a specified axis.

You can preserve the number of dimensions by setting keepdims=True.

Function Signature¶

def argmin(
    a: Tensor,
    axis: int | None = None,
    keepdims: bool = False,
) -> Tensor

Parameters¶

a (Tensor): Input tensor to evaluate minimum indices from.
axis (int or None, optional): Axis along which to find the index of the minimum. If None, the input is flattened. Defaults to None.
keepdims (bool, optional): If True, retains reduced dimensions with size 1. Defaults to False.

Returns¶

Tensor (Int64): Indices of the minimum values along the specified axis.

\[\begin{split}\operatorname{shape}(\text{out}) \;=\; \begin{cases} (1, 1, \ldots) & \text{if keepdims=True} \\ \text{reduced shape} & \text{otherwise} \end{cases}\end{split}\]

Note

argmin is gradient-free; back-propagation will not propagate through the returned indices.

Examples¶

Global minimum index

>>> x = lucid.Tensor([[3, 2], [1, 4]])
>>> lucid.argmin(x)
Tensor(2, grad=None)

Along axis, keeping dims

>>> lucid.argmin(x, axis=1, keepdims=True)
Tensor([[1], [0]], grad=None)