numpy和pytorch矩阵乘法

numpy官方文档

pytorch官方文档

一、numpy的矩阵乘法

numpy.dot(a, b, out=None)

当两个数据均为一维的时候，结果是内积，等效于np.matmul or a@b。当两个数据均为二维的时候，结果是矩阵乘法，依然等效于np.matmul or a@b

# 1-D array
a = np.array([1, 2, 3])
b = np.array([4, 5, 6])
result_ab = np.dot(a, b)
result_ba = np.dot(b, a)
print('result_ab: %s' % (result_ab))
print('result_ba: %s' % (result_ba))
'''
result_ab: 32
result_ba: 32
'''

# 2-D array
a = np.array([[1, 2, 3], [4, 5, 6]]) # 2-D array: 2 x 3
b = np.array([[1, 2], [3, 4], [5, 6]]) # 2-D array: 3 x 2
result_ab = np.dot(a, b)
result_ba = np.dot(b, a)
print('dot_result_ab:\n %s' %(result_ab))
print('dot_result_ba:\n %s' %(result_ba))
'''
dot_result_ab:
 [[22 28]
 [49 64]]
dot_result_ba:
 [[ 9 12 15]
 [19 26 33]
 [29 40 51]]
'''


result_ab = np.matmul(a, b)
result_ba = np.matmul(b, a)
print('matmul_result_ab:\n %s' %(result_ab))
print('matmul_result_ba:\n %s' %(result_ba))
'''
matmul_result_ab:
 [[22 28]
 [49 64]]
matmul_result_ba:
 [[ 9 12 15]
 [19 26 33]
 [29 40 51]]
'''

2. 如果一个是向量，一个是标量，等价于np.multiply or a*b

# array and scalar
a = np.array([[1, 2, 3], [4, 5, 6]]) # 2-D array: 2 x 3
b = 3 # 标量
result_ab = np.dot(a, b)
print('result_ab:\n %s' %(result_ab))
'''
result_ab:
 [[ 3  6  9]
 [12 15 18]]
'''


multiply_result_ab = np.multiply(a,b)
print('multiply_result_ab:\n %s' %(multiply_result_ab))
'''
multiply_result_ab:
 [[ 3  6  9]
 [12 15 18]]
'''

3. 如果a是N-D数组，b是1-D数组，最终是a和b最后一个轴的乘积（a sum product over the last axis of a and b)

a = np.array([[1, 2, 3], [4, 5, 6]]) # array: 2 x 3
b = np.array([1, 2, 3]) # array: 1 x 3
result_ab = np.dot(a, b)
print('result_ab:\n %s' %(result_ab))
'''
result_ab:
 [14 32]
'''

numpy.matmul(x1, x2）

面对2-D array与2-D array时与dot功能相同，同时类似于@，与dot不同的是matmul不能够进行矩阵与标量的乘法

a = np.array([[1, 0], [0, 1]]) # array: 2 x 2
b = np.array([[4, 1],[2, 2]]) # array: 2 x 2
result_ab = np.matmul(a, b)
print('result_ab:\n %s' %(result_ab))
'''
result_ab:
 [[4 1]
 [2 2]]
'''


a = np.array([[1, 0], [0, 1]]) # array: 2 x 2
b = np.array([1, 2]) # array: 1 x 2
result_ab = np.matmul(a, b)
print('result_ab:\n %s' %(result_ab))
'''
result_ab:
 [1 2]
'''

numpy.multiply(x1, x2）

numtiply针对的是标量的运算，同时类似于*，当两个向量参与运算时，得到的结果是对应位置的乘积

result = np.multiply(2.0, 4.0)
print(result)
8.0
x1 = np.arange(9.0).reshape((3, 3))
x2 = np.arange(3.0)
result = np.multiply(x1, x2)
print(x1)
[[0. 1. 2.]
 [3. 4. 5.]
 [6. 7. 8.]]
print(x2)
[0. 1. 2.]
print(result)
[[ 0.  1.  4.]
 [ 0.  4. 10.]
 [ 0.  7. 16.]]

二、torch的矩阵乘法

torch.mul(input, other, *, out=None) → Tensor

有两种情况

• input是一个向量，other是一个标量或整数即可直接相乘
• input是一个向量，other是一个向量，对应位相乘且遵从broadcast机制

# input:tensor--output:number
a = torch.randn(3)
a
tensor([ 0.2015, -0.4255,  2.6087])
torch.mul(a, 100)
tensor([  20.1494,  -42.5491,  260.8663])</code></pre>

# input:tensor--output:tensor

# -----with same dimension----
a = torch.rand(3, 2)
a
tensor([[0.8705, 0.0193],
        [0.8379, 0.3961],
        [0.4431, 0.4504]])
b = torch.rand(3, 2)
tensor([[0.9896, 0.3956],
        [0.2082, 0.5905],
        [0.6654, 0.0141]])
print(torch.mul(a, b))
tensor([[0.8615, 0.0076],
        [0.1744, 0.2339],
        [0.2949, 0.0064]])

# -----with different dimension---obey broadcast
a = torch.randn(4, 1)
a
tensor([[ 1.1207],
        [-0.3137],
        [ 0.0700],
        [ 0.8378]])
b = torch.randn(1, 4)
b
tensor([[ 0.5146,  0.1216, -0.5244,  2.2382]])
torch.mul(a, b)
tensor([[ 0.5767,  0.1363, -0.5877,  2.5083],
        [-0.1614, -0.0382,  0.1645, -0.7021],
        [ 0.0360,  0.0085, -0.0367,  0.1567],
        [ 0.4312,  0.1019, -0.4394,  1.8753]])</code></pre>

torch.mm(input, mat2, *, out=None) → Tensor

仅支持矩阵相乘，输入为(m,n)的矩阵，另一个矩阵维度必须为(n,p)的形式，因为此函数不支持broadcast机制

mat1 = torch.randn(2, 3)
mat2 = torch.randn(3, 3)
torch.mm(mat1, mat2)
tensor([[ 0.4851,  0.5037, -0.3633],
        [-0.0760, -3.6705,  2.4784]])