Win10搭建tensorflow环境笔记

版本记录

Anaconda :
CUDA Toolkit : v10.1
cudnn: v7.6.4
tensorflow : tensorflow-gpu 2.1.0

Hardware

显卡 ：NVIDIA GeForce GTX 1050
Driver Version ： 442.
Note：查看显卡及驱动版本

nvidia-smi

Choose the right version suitable for you device

tensorflow-gpu 2.0要求CUDA v10.0及以上, cudnn>=7.4.1
tensorflow-gpu 2.1要求CUDA v10.1及以上, cudnn>=7.6

Anaconda 环境搭建

#优先命名用清华conda 镜像
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/
conda config --append channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/fastai/
conda config --append channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/pytorch/
conda config --append channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/bioconda/

# 搜索时显示通道地址
conda config --set show_channel_urls yes

conda create -n tf2 python=3.6

# check if the environment has been created successfully
conda info --env
# activate the environment
conda activate tf2
# install the gpu version
conda install cudatoolkit=10.1 cudnn=7.6
conda install tensorflow-gpu=2.1.0
# to install the cpu version: 

# check if tensorflow has been installed succesfully
python
import tensorflow as tf
# check if gpu version has been built succesfully
tf.test.is_built_with_gpu()
# remember to exit
exit()

# if the AttributeError comes
# use the following command:
conda list # to check whether the tensorflow-gpu and tensorflow-gpu estimator
# are version-matched
# If not matched, use following command to install the matched estimator
conda install tensorflow-estimator=2.1.0

# when everything is done, deactivate the environment
conda deactivate

Create new project in Pycharm based on the environment built before

FIle -> Settings

Project -> Python Interpreter -> 小齿轮 -> Add -> 添加环境下的python.exe文件

为了更直观的调试，可以将文件设置为在console下运行：
Run -> Edit Configurations

Python -> run with console

Remark

若设备无独立显卡，或显卡不是NVIDIA，则不能通过下载CUDA Toolkit来搭载tensorflow-gpu环境，可以搭建tensorflow环境。

参考博文

Tensorflow2.x教程

CUDA及cudnn安装

---

19/11/2020

更新

20/11/2020
安装matplotlib库时报错：
location of failed script: F:\Anaconda\envs\tf2\Scripts.qt-post-link.bat>
解决方法：
参考：问答

没找到回答中的文件，故在
F:\Anaconda\pkgs\qt-5.9.7-vc14h73c81de_0\Scripts里修改.qt-post-link.bat文件：
@echo off 下方添加

set PATH=%PATH%;%SystemRoot%\system32;%SystemRoot%;%SystemRoot%\System32\Wbem;

再次安装，安装成功但是会因为文件字长不匹配报warning：