Tensorflow

Tensorflow

Hints for user

Install Tensorflow with pip

>sudo apt install python3-pip

>pip3 install tensorflow

>pip3 show tensorflow

(show path)

>pip install pip-upgrade-outdated

Test tensorflow

(path: /usr/bin/python3.6)

>python3

>>>import tensorflow as tf

>>>print(tf.__version__)

(__version__ for version 1, 1.11.0, 1.13.1, not work in version 2)

>>>exit()

To create virtual env and install tensorflow in virtual env

>pip3 install -U pip virtualenv

>virtuallenv –system-site-packages -p python3 ./venv

>.\venv\Scripts\activate

>virtualenv c:\Users\Users\AppData\Local\Python37-32\Scripts

>cd Scripts

>activate

>pip install tensorflow==2.0.0-alpha0

To activate tf virtual env.

Work under venv (eg. work in py3.6 version)

>source ./venv/bin/activate

In conda, to install tensorflow or tensorflow-gpu version

>conda create -n tensorflow

or

>conda create -n tf-gpu

>conda activate tensorflow

or

>conda activate tf-gpu

or

>pip install tensorflow

or

>pip install tensorflow-gpu

To show tensorflow path

>pip show tensorflow

(eg. /home/alanng/./local/lib/py3.6/site pk

Command pip update

>python -m pip install –upgrade pip

or

>sudo -H pip3 install –upgrade pip

Install apps

>python -m pip install -U matplotlib

>python -m pip install –user numpy scipy matplotlib ipython jupyter pardas symp nose

For gpu operation

Nivida

>sudo apt install nividia-cuda-tookit

>sudo sh cuda_10.1.105_418.39_linux.run.part

pycuda

>pip3 install pycuda

Check cuddn

Go to terminal, type

>where cudnn*

>nvcc –version

In OS Windows10

In Windows10 terminal, type

>sysdm.cpl

For an example,

Environment: (PATH)

Variable name: CUDA_PATH

Variable value: C:\Program Files\Nvidia GPU Computing Tookit\CUDA\v.9.0

Additional, there are 3 files in folder

\V.9.0\bin

\V.9.0\libnvp

File 1: cudnn64_7.dll (\v.9.0\bin\)

File 2: cudnn.lib (\v.9.0\lib\x64\)

File 3: cudnn.h (v.9.0\include)

Edit Win32 long paths (path parameter edit)

gpedit.msc à FileSystem Config à Enable Win32 long paths