argmin_op

→TensorImplPtr

int argmin_op(const int & a, int axis, bool keepdims)

Index of the minimum value along an axis.

See Also

argmax_op : Maximum-index counterpart.

aTensorImplPtr

Input tensor.

axisint

Axis to reduce.

keepdimsbool

If true, retain the reduced axis as size 1; otherwise remove it.

TensorImplPtr

Integer tensor of dtype I64 containing the argmin indices.

Ties are broken by returning the lowest index. Non-differentiable.