Torch learning notes (4) -- torch's dynamic calculation diagram

import torch
import numpy as np
import matplotlib.pyplot as plt

#  Calculation chart （Computer Graph） Is a directed acyclic graph used to describe operations 
# CG There are two main elements ： node （Node）、 edge （Edge）
# Node According to the data 
# Edge Representation operation 
#  If you want to use non leaf Gradient of nodes . You need to call before back propagation .retain_grad（）
# CG effect ： Convenient gradient calculation and back propagation 
#  Case study 
# y = (x+w)(w+1), Then when w=1,x=2 when ,y Yes w Of grad How much is the ？
w = torch.tensor([1.], requires_grad=True)
x = torch.tensor([2.], requires_grad=True)

a = torch.add(w, x)
b = torch.add(w, 1)
y = torch.mul(a, b)

y.backward()
#  Can pass .grad Determine whether the variable has grad, After the end of back propagation, it is not leaf Nodal grad Will be released 
print(w.grad)
# grad_fn Used to record the creation of tensor Used in function, The essence is to record Edge
print(y.grad_fn)
#  according to CG Construction method , Can be CG It is divided into dynamic graph and static graph 
#  Dynamic graph （Dynamic Graph） Operation and loading are carried out at the same time , self-driving tours 
#  Static diagram （Computation Graph） First loading diagram , Post operation , Travel with a group