pytorch基本操作（对比tensorflow 2）

无论是pytorch还是tensorflow 2，基本操作均与numpy相同

pytorch:

torch.empty(5,3)

tensor([[7.9604e-11, 4.2179e-43, 7.9606e-11],
        [4.2179e-43, 7.9598e-11, 4.2179e-43],
        [7.9581e-11, 4.2179e-43, 7.9608e-11],
        [4.2179e-43, 7.9601e-11, 4.2179e-43],
        [7.9590e-11, 4.2179e-43, 7.9608e-11]])

torch.rand(5,3)   # 随机值

tensor([[0.9154, 0.4700, 0.2335],
        [0.9463, 0.8901, 0.2002],
        [0.4432, 0.0426, 0.9394],
        [0.5706, 0.2271, 0.0835],
        [0.6255, 0.9812, 0.0346]])

torch.tensor([5,3])        #传入值

tensor([5, 3])

tensorflow 2

tf.constant([[1., 2., 3.], [4., 5., 6.]]) # matrix

tf.add()/torch.add()

tensorflow 2与numpy有些函数名称上有些差异，但pytorch更接近numpy

x=torch.randn(4,4)
y = x.view(16)
z = x.view(-1,8)
print(x.size(),y.size(),z.size())

torch.Size([4, 4]) torch.Size([16]) torch.Size([2, 8])

pytorch:

import numpy as np
a = torch.ones(1,5)
a.numpy()

array([[1., 1., 1., 1., 1.]], dtype=float32)

 a = np.ones(5)
torch.from_numpy(a)

tensor([1., 1., 1., 1., 1.], dtype=torch.float64)

Tensorflow 2:

a = np.array([2., 4., 5.])
tf.constant(a)

t.numpy()

array([[1., 2., 3.],
       [4., 5., 6.]], dtype=float32)

np.array(t)

array([[1., 2., 3.],
       [4., 5., 6.]], dtype=float32)

tf.square(a)

np.square(t)

array([[ 1.,  4.,  9.],
       [16., 25., 36.]], dtype=float32)

b = torch.randn(1,requires_grad=True) # 可以求导
x=torch.rand(1,requires_grad=True)
w=torch.rand(1,requires_grad=True)
y = w*x
z = y+b
z.backward()
b.grad,x.grad

(tensor([1.]), tensor([0.8892]))

y.requires_grad,x.requires_grad,b.requires_grad,w.requires_grad,z.requires_grad

(True, True, True, True, True)

反向传播：

z.backward(retain_graph=True) # 如果不清空就会累加
w.grad

tensor([0.6140])

b.grad

tensor([2.])

链式法则：

 def f(w1, w2):
    return 3 * w1 ** 2 + 2 * w1 * w2

w1, w2 = tf.Variable(5.), tf.Variable(3.)
with tf.GradientTape() as tape:
    z = f(w1, w2)

gradients = tape.gradient(z, [w1, w2])

[,
 ]

# 二阶求导

with tf.GradientTape(persistent=True) as hessian_tape: # persistent = True :tpye具有持久性
    with tf.GradientTape() as jacobian_tape:
        z = f(w1, w2)
    jacobians = jacobian_tape.gradient(z, [w1, w2])
hessians = [hessian_tape.gradient(jacobian, [w1, w2])
            for jacobian in jacobians]
del hessian_tape