mindspore.nn.MatMul

class mindspore.nn.MatMul(**kwargs)[source]

Multiplies matrix x1 by matrix x2.

nn.MatMul will be deprecated in future versions. Please use ops.matmul instead.

  • If both x1 and x2 are 1-dimensional, the dot product is returned.

  • If the dimensions of x1 and x2 are all not greater than 2, the matrix-matrix product will be returned. Note if one of ‘x1’ and ‘x2’ is 1-dimensional, the argument will first be expanded to 2 dimension. After the matrix multiply, the expanded dimension will be removed.

  • If at least one of x1 and x2 is N-dimensional (N>2), the none-matrix dimensions(batch) of inputs will be broadcasted and must be broadcastable. Note if one of ‘x1’ and ‘x2’ is 1-dimensional, the argument will first be expanded to 2 dimension and then the none-matrix dimensions will be broadcasted. After the matrix multiply, the expanded dimension will be removed. For example, if x1 is a \((j \times 1 \times n \times m)\) tensor and x2 is a \((k \times m \times p)\) tensor, the output will be a \((j \times k \times n \times p)\) tensor.

Parameters
  • transpose_x1 (bool) – If true, a is transposed before multiplication. Default: False.

  • transpose_x2 (bool) – If true, b is transposed before multiplication. Default: False.

Inputs:
  • input_x1 (Tensor) - The first tensor to be multiplied.

  • input_x2 (Tensor) - The second tensor to be multiplied.

Outputs:

Tensor, the shape of the output tensor depends on the dimension of input tensors.

Raises
  • TypeError – If transpose_x1 or transpose_x2 is not a bool.

  • ValueError – If the column of matrix dimensions of input_x1 is not equal to the row of matrix dimensions of input_x2.

Supported Platforms:

Ascend GPU CPU

Examples

>>> net = nn.MatMul()
>>> input_x1 = Tensor(np.ones(shape=[3, 2, 3]), mindspore.float32)
>>> input_x2 = Tensor(np.ones(shape=[3, 4]), mindspore.float32)
>>> output = net(input_x1, input_x2)
>>> print(output.shape)
(3, 2, 4)