[−][src]Function autograd::ops::matmul_t
pub fn matmul_t<T: Float, A: AsRef<Tensor<T>>, B: AsRef<Tensor<T>>>(
a: A,
b: B,
transpose_a: bool,
transpose_b: bool
) -> Tensor<T>
Matrix multiplication with inputs's transposition.
Similar specification as matmul
but, if transpose_a
is true, a
is transposed
before actual matrix multiplication. It is the same for transpose_b
.
The performance is better than explicitly computing like ag::matmul(ag::transpose)
.
extern crate autograd as ag; let ref a: ag::Tensor<f32> = ag::zeros(&[2, 4]); let ref b: ag::Tensor<f32> = ag::zeros(&[2, 3]); let ref c = ag::matmul_t(a, b, true, false); assert_eq!(c.eval(&[]).unwrap().shape(), &[4, 3]);