tf中 multiply 和 matmul 的区别

multiply: element-wise，x,y维数必须相同，否则报错，Returns x * y element-wise.

matmul: Multiplies matrix a by matrix b, producing a * b. 个人理解和numpy库里的np.dot用法相同

tf库解释：

1.tf.multiply

tf.multiply(
    x,
    y,
    name=None
)

Defined in tensorflow/python/ops/math_ops.py.

See the guide: Math > Arithmetic Operators

Returns x * y element-wise.

NOTE: tf.multiply supports broadcasting. More about broadcasting here

Args:

• x: A Tensor. Must be one of the following types: bfloat16, half, float32, float64, uint8, int8, uint16, int16, int32, int64, complex64, complex128.
• y: A Tensor. Must have the same type as x.
• name: A name for the operation (optional).

Returns:

A Tensor. Has the same type as x.

2.tf.matmul

tf.matmul(
    a,
    b,
    transpose_a=False,
    transpose_b=False,
    adjoint_a=False,
    adjoint_b=False,
    a_is_sparse=False,
    b_is_sparse=False,
    name=None
)

Defined in tensorflow/python/ops/math_ops.py.

See the guide: Math > Matrix Math Functions

Multiplies matrix a by matrix b, producing a * b.

The inputs must, following any transpositions, be tensors of rank >= 2 where the inner 2 dimensions specify valid matrix multiplication arguments, and any further outer dimensions match.

Both matrices must be of the same type. The supported types are: float16, float32, float64, int32, complex64, complex128.

Either matrix can be transposed or adjointed (conjugated and transposed) on the fly by setting one of the corresponding flag to True. These are False by default.

If one or both of the matrices contain a lot of zeros, a more efficient multiplication algorithm can be used by setting the corresponding a_is_sparse or b_is_sparse flag to True. These are False by default. This optimization is only available for plain matrices (rank-2 tensors) with datatypes bfloat16 or float32.

For example:

# 2-D tensor `a`
# [[1, 2, 3],
#  [4, 5, 6]]
a = tf.constant([1, 2, 3, 4, 5, 6], shape=[2, 3])

# 2-D tensor `b`
# [[ 7,  8],
#  [ 9, 10],
#  [11, 12]]
b = tf.constant([7, 8, 9, 10, 11, 12], shape=[3, 2])

# `a` * `b`
# [[ 58,  64],
#  [139, 154]]
c = tf.matmul(a, b)


# 3-D tensor `a`
# [[[ 1,  2,  3],
#   [ 4,  5,  6]],
#  [[ 7,  8,  9],
#   [10, 11, 12]]]
a = tf.constant(np.arange(1, 13, dtype=np.int32),
                shape=[2, 2, 3])

# 3-D tensor `b`
# [[[13, 14],
#   [15, 16],
#   [17, 18]],
#  [[19, 20],
#   [21, 22],
#   [23, 24]]]
b = tf.constant(np.arange(13, 25, dtype=np.int32),
                shape=[2, 3, 2])

# `a` * `b`
# [[[ 94, 100],
#   [229, 244]],
#  [[508, 532],
#   [697, 730]]]
c = tf.matmul(a, b)

# Since python >= 3.5 the @ operator is supported (see PEP 465).
# In TensorFlow, it simply calls the `tf.matmul()` function, so the
# following lines are equivalent:
d = a @ b @ [[10.], [11.]]
d = tf.matmul(tf.matmul(a, b), [[10.], [11.]])

Args:

• a: Tensor of type float16, float32, float64, int32, complex64, complex128 and rank > 1.
• b: Tensor with same type and rank as a.
• transpose_a: If True, a is transposed before multiplication.
• transpose_b: If True, b is transposed before multiplication.
• adjoint_a: If True, a is conjugated and transposed before multiplication.
• adjoint_b: If True, b is conjugated and transposed before multiplication.
• a_is_sparse: If True, a is treated as a sparse matrix.
• b_is_sparse: If True, b is treated as a sparse matrix.
• name: Name for the operation (optional).

Returns:

A Tensor of the same type as a and b where each inner-most matrix is the product of the corresponding matrices in a and b, e.g. if all transpose or adjoint attributes are False:

output[..., i, j] = sum_k (a[..., i, k] * b[..., k, j]), for all indices i, j.

• Note: This is matrix product, not element-wise product.

Raises:

• ValueError: If transpose_a and adjoint_a, or transpose_b and adjoint_b are both set to True.