【Pytorch】Windows10下配置Pytorch環境

0x00 前言

前言什麼的也懶得說了……
總之：
聽說你Pytorch很牛，
不樂意讓我Windows用，
而我又聽說pytorch用來訓練模型超好用，
不僅沒頭腦而且不高興！我要在我的windows上配一個！

[NEW] PyTorch在0.4.0版已完美適配Windows，詳情請見這篇文章
https://blog.csdn.net/okcd00/article/details/80077799

0x01 準備工作

• Win10 64位操作系統
• Anaconda3：本人用的是3，2應該也可以（吧？）
• Pytorch 的 conda離線安裝包 (Thx to 知乎@蒲嘉宸)
  鏈接：http://pan.baidu.com/s/1jH9F7Si 密碼：8s5a

0x02 配置開始

首先，我們需要一個比較配pytorch的環境，此處使用python3.6.1，因爲我把離線安裝包下載到了D盤的Downloads目錄下，所以我們就在D:\Downloads\這裏操作吧，根據你們想放在哪裏自行安排。

$ conda create -n torch python=3.6.1
Fetching package metadata ...........
Solving package specifications: .

Package plan for installation in environment C:\ProgramData\Anaconda3\envs\torch:

The following NEW packages will be INSTALLED:

    certifi:        2016.2.28-py36_0
    pip:            9.0.1-py36_1
    python:         3.6.1-2
    setuptools:     36.4.0-py36_0
    vs2015_runtime: 14.0.25420-0
    wheel:          0.29.0-py36_0
    wincertstore:   0.2-py36_0

Proceed ([y]/n)? y

vs2015_runtime 100% |###############################| Time: 0:01:28  23.07 kB/s
python-3.6.1-2 100% |###############################| Time: 0:28:32  19.29 kB/s
certifi-2016.2 100% |###############################| Time: 0:00:10  21.24 kB/s
wincertstore-0 100% |###############################| Time: 0:00:00  45.54 kB/s
setuptools-36. 100% |###############################| Time: 0:00:21  28.14 kB/s
#
# To activate this environment, use:
# > activate torch
#
# To deactivate an active environment, use:
# > deactivate
#
# * for power-users using bash, you must source
#

然後我們進入這個只有默認包的conda環境中，至少把numpy啥的pytorch必備組件給他裝上：

(torch) D:\Downloads>conda install numpy mkl cffi
Fetching package metadata ...........
Solving package specifications: .

Package plan for installation in environment C:\ProgramData\Anaconda3\envs\torch:

The following NEW packages will be INSTALLED:

    cffi:      1.10.0-py36_0
    mkl:       2017.0.3-0
    numpy:     1.13.1-py36_0
    pycparser: 2.18-py36_0

Proceed ([y]/n)?

pycparser-2.18 100% |###############################| Time: 0:00:06  27.88 kB/s

然後就是激動人心的裝這個離線包了！

(torch) D:\Downloads>conda install --offline pytorch-0.2.1-py36he6bf560_0.2.1cu80.tar.bz2

CondaError: EOFError('Compressed file ended before the end-of-stream marker was reached',)
CondaError: EOFError('Compressed file ended before the end-of-stream marker was reached',)
CondaError: EOFError('Compressed file ended before the end-of-stream marker was reached',)

我很想大聲的喊一句“不可能只有我一個人，不會只有我遇到這種情況！”
在check了一下下載的壓縮包並沒有損壞之後，查到一個解決方案：（Thx to @李新鵬）:

解決‘CondaError: EOFError(‘Compressed file ended before the end-of-stream marker was reached’,)’問題，在anaconda/pkgs文件夾刪除緩存的pytorch安裝包再安裝比較好，不會刪去其餘完好的安裝包

然後在anaconda目錄（比如我的在C:\ProgramData\Anaconda3\pkgs\）下，找到先前緩存的pytorch包（比如我的是C:\ProgramData\Anaconda3\pkgs\pytorch-0.2.1-py36he6bf560_0.2.1cu80.tar.bz2），直接把這個文件刪掉，然後回到CMD中，再次輸入

(torch) D:\Downloads>conda install --offline pytorch-0.2.1-py36he6bf560_0.2.1cu80.tar.bz2

(torch) D:\Downloads>python
Python 3.6.1 |Continuum Analytics, Inc.| (default, May 11 2017, 13:25:24) [MSC v.1900 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import pytorch
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ModuleNotFoundError: No module named 'pytorch'
>>> import torch
>>> x = torch.Tensor([1.0])
>>> xx = x.cuda()
>>> print(xx)

 1
[torch.cuda.FloatTensor of size 1 (GPU 0)]

可喜可賀，可喜可賀~

哦對了最後還可以放進jupyter notebook的kernel裏用

sudo pip install -U ipykernel
# source activate torch
python -m ipykernel install --user