jax.numpy.matvec#

jax.numpy.matvec(x1,x2,/)[source]#

Batched matrix-vector product.

JAX implementation ofnumpy.matvec().

Parameters:

• x1 (Array |ndarray |bool |number |bool |int |float |complex |TypedNdArray) – array of shape(...,M,N)

• x2 (Array |ndarray |bool |number |bool |int |float |complex |TypedNdArray) – array of shape(...,N). Leading dimensions must be broadcast-compatiblewith leading dimensions ofx1.

Returns:

An array of shape(...,M) containing the batched matrix-vector product.

Return type:

Array

See also

• jax.numpy.linalg.vecdot(): batched vector product.

• jax.numpy.vecmat(): vector-matrix product.

• jax.numpy.matmul(): general matrix multiplication.

Examples

Simple matrix-vector product:

>>>x1=jnp.array([[1,2,3],...[4,5,6]])>>>x2=jnp.array([7,8,9])>>>jnp.matvec(x1,x2)Array([ 50, 122], dtype=int32)

Batched matrix-vector product:

>>>x2=jnp.array([[7,8,9],...[5,6,7]])>>>jnp.matvec(x1,x2)Array([[ 50, 122],       [ 38,  92]], dtype=int32)