pytorch第二天（矩阵的乘法和求导）lm老师版

import torch

A = torch.arange(20).reshape(5, 4)

print(A)

print(A.T)#A的转置

#not 矩阵乘
A = torch.arange(20, dtype=torch.float32).reshape(5, 4)
B = A.clone()
print(A*B)#就是简单的对应位置相乘 不是矩阵乘法

A =torch.tensor([[1,2,3],
                 [1,2,1],
                 [4,1,7]],dtype=torch.float32)

print(A.shape)

A_sum_axis0 = A.sum(axis=0)#把0维消除
print(A_sum_axis0)

A_sum_axis1 = A.sum(axis=1)#把第一维消除
print(A_sum_axis1)

A_sum_axis10=A.sum(axis=[0, 1])#全加起来对于这道
print(A_sum_axis10)

print(A.mean())#全部的sum 除 个数
print((A.mean() == (A.sum()/ A.numel())))


print(A.mean(axis=0))#把第一维消除 一样的原理

sum_A = A.sum(axis=1, keepdims=True)
print(sum_A.shape)#不会消掉1维度，会将那个维度变成1 后面乘法算术可以使用

print(A.cumsum(axis=0))#按下 累加


y = torch.ones(4, dtype=torch.float32)
x = torch.arange(4.)
print(torch.dot(x,y))#点积是相同位置的按元素乘积的和 结果等于torch.sum(x * y)


MA = torch.tensor([[1,2,3],
                   [1,2,3],
                   [1,2,3],
                   [2,3,4]])

MB = torch.tensor([[1,1,1,1],
                   [1,1,1,1],
                   [1,1,1,1]])

print(MA.shape,MB.shape)
print(torch.mm(MB,MA))#矩阵乘法


MC = torch.tensor([1,1,1])
print(torch.mv(MA,MC))#矩阵和向量相乘乘法


u = torch.tensor([3.0, -4.0])
print(torch.norm(u))#平方和再开根号 为5

print(torch.abs(u).sum())#L1范数 绝对值求和

print(torch.norm(torch.ones((4, 9))))#矩阵 的弗罗贝尼乌斯范数（Frobenius norm）是矩阵元素平方和的平方根

#求导
x = torch.arange(4.0)
print(x)
x.requires_grad_(True)
print(x.grad)#初始值为none
y = 2 * torch.dot(x, x)#内积的2倍

y.backward()
print(x.grad)#4x 内积求导为2xT

x.grad.zero_()#清0梯度 防止累加
y = x.sum()
y.backward()
print(x.grad)

x.grad.zero_()
y = x * x
y.sum().backward()
print(x.grad)


x.grad.zero_()
y = x * x
u = y.detach()#u就是y的常数
z = u * x
z.sum().backward()
print(x.grad == u)
print(x.grad)

#线性的
def f(a):
    b = a * 2
    while b.norm() < 1000:
        b = b * 2
    if b.sum() > 0:
        c = b
    else:
        c = 100 * b
    return c

a = torch.randn(size=(), requires_grad=True)
d = f(a)
d.backward()

print(a.grad == d / a)

运行结果：

F:\python3\python.exe C:\study\project_1\3_21_xiandai.py 
tensor([[ 0,  1,  2,  3],
        [ 4,  5,  6,  7],
        [ 8,  9, 10, 11],
        [12, 13, 14, 15],
        [16, 17, 18, 19]])
tensor([[ 0,  4,  8, 12, 16],
        [ 1,  5,  9, 13, 17],
        [ 2,  6, 10, 14, 18],
        [ 3,  7, 11, 15, 19]])
tensor([[  0.,   1.,   4.,   9.],
        [ 16.,  25.,  36.,  49.],
        [ 64.,  81., 100., 121.],
        [144., 169., 196., 225.],
        [256., 289., 324., 361.]])
torch.Size([3, 3])
tensor([ 6.,  5., 11.])
tensor([ 6.,  4., 12.])
tensor(22.)
tensor(2.4444)
tensor(True)
tensor([2.0000, 1.6667, 3.6667])
torch.Size([3, 1])
tensor([[ 1.,  2.,  3.],
        [ 2.,  4.,  4.],
        [ 6.,  5., 11.]])
tensor(6.)
torch.Size([4, 3]) torch.Size([3, 4])
tensor([[ 5,  9, 13],
        [ 5,  9, 13],
        [ 5,  9, 13]])
tensor([6, 6, 6, 9])
tensor(5.)
tensor(7.)
tensor(6.)
tensor([0., 1., 2., 3.])
None
tensor([ 0.,  4.,  8., 12.])
tensor([1., 1., 1., 1.])
tensor([0., 2., 4., 6.])
tensor([True, True, True, True])
tensor([0., 1., 4., 9.])
tensor(True)

进程已结束,退出代码0