1. Learn the environment installation and basic calculation of paddlepaddle from scratch

paddlepaddle It is Baidu's deep learning framework , Now that it is fully localized , Study paddlepaddle It is necessary .

This series is about paddlepaddle Summary in learning , I hope you like it .

paddlepaddle The official website of Flying propeller PaddlePaddle- Open source deep learning platform from industrial practice ​​​​​​​.

First step , install anaconda, Here we choose Tsinghua source conda, It's easy to find .

# Open the web site  https://mirror.tuna.tsinghua.edu.cn/help/anaconda/
# perhaps  wget https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/Anaconda3-2022.05-Linux-x86_64.sh
chmod +x Anaconda3-2022.05-Linux-x86_64.sh
./Anaconda3-2022.05-Linux-x86_64.sh

The second step , install paddlepaddle Environment , Here we use pip

pip install paddlepaddle -i https://mirror.baidu.com/pypi/simple

The third step , see paddle Of version

pip list|grep paddlepaddle #2.3.1

You can see paddlepaddle Automatically installed to 2.3.1 edition

Step four , Turn on paddle One small demo test , as follows ：

import paddle
x = paddle.to_tensor(1)
w = paddle.to_tensor(2)
b = paddle.to_tensor(3)
y = w*x+b
print(y)#Tensor(shape=[1], dtype=int64, place=Place(cpu), stop_gradient=True,[5])

  Above ,paddle The simple calculation test is successful .

Step five , Try paddle Gradient calculation in , This is still the case , stay paddle We need to add stop_gradient Make it possible to calculate the gradient .

import paddle
x = paddle.to_tensor(1,stop_gradient=False)
w = paddle.to_tensor(2,stop_gradient=False)
b = paddle.to_tensor(3,stop_gradient=False)
y = w*x+b
print(y)#Tensor(shape=[1], dtype=int64, place=Place(cpu), stop_gradient=True,[5])
y.backward()
print(y.grad)#Tensor(shape=[1], dtype=int64, place=Place(cpu), stop_gradient=True,[1])
print(x.grad)#Tensor(shape=[1], dtype=int64, place=Place(cpu), stop_gradient=True,[2])
print(w.grad)#Tensor(shape=[1], dtype=int64, place=Place(cpu), stop_gradient=True,[1])
print(b.grad)#Tensor(shape=[1], dtype=int64, place=Place(cpu), stop_gradient=True,[1])

You can see back propagation , Calculate the gradient , No problem ,paddle In the vector calculation and back-propagation methods and pytorch More similar , It's just torch Is in the torch.tensor, and paddle Is in the  to_tensor.

stay paddle You can also simply create vectors in , You can directly python List data conversion , as follows ：

import paddle
x = paddle.to_tensor([2,3])
print(x)#Tensor(shape=[2], dtype=int64, place=Place(cpu), stop_gradient=True,[2, 3])
x = paddle.to_tensor([[2,3]])
print(x)#Tensor(shape=[1, 2], dtype=int64, place=Place(cpu), stop_gradient=True,[[2, 3]])

Many times when building models ,tensor The dimension of is very important , So we can pass .shape obtain , Of course, you can also print directly , Now look at paddle The bottom layer is similar numpy Of .

x = paddle.to_tensor([[2,3]])
print(x.shape)#[1,2]