Installation considerations for pytorch

The official download address of pytorch is:
https://download.pytorch.org/whl/torch_stable.html
Torch and torchvision need to be installed.The version is divided into support cpu and gpu, and there are many, you need to pay attention to find.
The installation command is

pip install torch-1.10.0+cu113-cp39-cp39-win_amd64.whlpip install torchvision-0.12.0+cpu-cp39-cp39-win_amd64.whl

It should be noted that the versions of torch and torchvision are different.This is the first pit.Due to the large torch package, it is recommended to install torchvision first, confirm the version and then download torch.
The second pit is that there will be errors when using tensorboard.The method given on the web is to use an older version of setuptool.This method is not advisable.It is better to modify the init.py file of tensorboard:

import tensorboard#from setuptools import distutilsfrom distutils import version as vvv#LooseVersion = distutils.version.LooseVersionLooseVersion = vvv.LooseVersionif not hasattr(tensorboard, '__version__') or LooseVersion(tensorboard.__version__) < LooseVersion('1.15'):raise ImportError('TensorBoard logging requires TensorBoard version 1.15 or above')#del distutilsdel vvvdel LooseVersiondel tensorboardfrom .writer import FileWriter, SummaryWriter # noqa: F401from tensorboard.summary.writer.record_writer import RecordWriter # noqa: F401

Use the distutils package directly.
The third pit is the ssl link.Because the certificate thing will cause the ssl link to fail.The method is to modify the ssl.py file:

#_create_default_https_context = create_default_context_create_default_https_context = _create_unverified_context

At line 805.Instead of create_default_context use _create_unverified_context.
The fourth pit is about cuda.It is estimated that every beginner will encounter it.I encountered it when I installed tensorflow.This time it went smoothly without any trouble.
I didn't expect tensorflow to decline so fast.I've also written a lot and am now learning pytorch from scratch.Programmers are hard work, because there are endless new things to learn.