eye_op

→TensorImplPtr

int eye_op(int N, int M, int k, Dtype dt, Device device, bool requires_grad)

Create an $N \times M$ identity-like matrix with ones on diagonal $k$ .

Math

\text{out}[i, j] = \begin{cases} 1 & j - i = k \\ 0 & \text{otherwise} \end{cases}

Nint64_t

Number of rows.

Mint64_t

Number of columns. If M <= 0, M is set to N.

kint64_t

Diagonal index. k = 0 is the main diagonal; positive k is above, negative k is below.

dtDtype

Element dtype.

deviceDevice

Target device.

requires_gradbool

Attach a fresh autograd leaf when true.

TensorImplPtr

2-D tensor of shape $(N, M)$ .