- tensorflow-gpu needs CUDA driver installed
conda create -n tensorflow python=3.5
activate tensorflow
(tensorflow)C:> # Your prompt should change
(tensorflow)C:> pip install --ignore-installed --upgrade tensorflow
or
(tensorflow)C:> pip install --ignore-installed --upgrade tensorflow-gpu
or with requirements.txt
conda create -n venv python=3.5
activate venv
conda install --yes --file requirements.txt
virtualenv -p python3 venv
source venv/bin/activate
pip install -r requirements.txt
python -m tensorboard.main --logdir=tb_logs
Proxy settings
-
for pip/pip3 : add argument
--proxy http://proxyweb:8080 -
for python : runtime variables
HTTP_PROXY=http://proxyweb:8080(HTTPS_PROXY=http://proxyweb:8080)
If installation package not found, specify the wheel version. Check versions here : https://pypi.python.org/pypi/tensorflow
pip3 install --ignore-installed --upgrade https://pypi.python.org/packages/bf/d4/80197f48f9fb90a17c47cdd834b2f13d5f714f26e8ed5c77069d57aa3ecb/tensorflow-1.3.0-cp36-cp36m-win_amd64.whl
pip install --ignore-installed --upgrade https://pypi.python.org/packages/bf/d4/80197f48f9fb90a17c47cdd834b2f13d5f714f26e8ed5c77069d57aa3ecb/tensorflow-1.3.0-cp36-cp36m-win_amd64.whl
If "Not valid wheel" issue, download the wheel to local and rename it to the valid py version
If "module not found", check pip show tensorflow and eventually install Redistribuable Visual C++ pour Visual Studio 2015 that contains MSVCP140.dll etc.
If "ImportError: No module named tensorflow" : run script with right python executable/env !