preamble
Since the basic idea of machine learning is to find a function tofit (data to a model)
distribution of the sample data, thus involving thegradient
go and askminimum value
In the hyperplane, it is difficult to get the global optimum directly, and there is no generalization, so we try to find a way to make the gradient go down in the negative direction, then we can get a local or global optimum, so the derivative becomes very important in machine learning.
Basic use
()
It is possible and automatic to accumulate the gradient into thefirst (of multiple parts)
x = (3,3) print(x.requires_grad) x.requires_grad_(True) print(x.requires_grad) y = x**2/(x-2) out = () print() () print()
False
True
None
tensor([[-0.3333, -0.3333, -0.3333],
[-0.3333, -0.3333, -0.3333],
[-0.3333, -0.3333, -0.3333]])
requires_grad
It is possible to get thetensor
differentiable (calculus)requires_grad_()
Can be settensor
differentiable (calculus)grad
View currenttensor
derivative
The above formula is very simple, the meaning of the program
1/4 * (x**2) / (x-2)
Find the derivative of x. The basic formula is below.
point of attention
We use.mean
What you get after that isscalar quantity
If notscalar quantity
will report an error
x = (3, requires_grad=True) y = x * 2 y = y * 2 print(y)
tensor([4., 4., 4.], grad_fn=<MulBackward0>)
() print()
report an error
RuntimeError: grad can be implicitly created only for scalar outputs
v = ([0.1, 1.0, 0.0001], dtype=) () print()
tensor([4.0000e-01, 4.0000e+00, 4.0000e-04])
no_grad()
scope (computing)
If we want some part of the program to be undirectable then we can use this
x = (3, requires_grad=True) y = x * 2 print(y.requires_grad) with torch.no_grad(): y = y * 2 print(y.requires_grad)
True
False
summarize
In this chapter we use pytorch's internalbackward
This automatically realizes the derivation of a function, which helps us in the later stage when we face a lot of functions with a large number of parameters, and the derivation becomes easy.
previous section
PyTorch Tutorial - Installation and Basic Usage
To this point this article on the use of PyTorch derivative application tutorial article is introduced to this, more related PyTorch derivative application content please search for my previous articles or continue to browse the following related articles I hope you will support me more in the future!