{{tag>Brouillon}}

= Notes install TensorFlow

Voir aussi :
* Pytorch 
* Keras

== install prérequis

Install de CUDA 9.0 et patches

Voir [[Install drivers graphique propriétaire Nvidia CUDA]]

puis
<code bash>
echo 'export PATH=$PATH:/usr/local/cuda/bin' |sudo tee /etc/profile.d/cuda.sh
echo "/usr/local/cuda/lib64" |sudo tee /etc/ld.so.conf.d/cuda.conf
sudo ldconfig
</code>

<code bash>
sudo apt-get install python3-numpy python3-venv
</code>

<code bash>
pyvenv plop
cd plop
source bin/activate
pip install -U pip
pip install -U pillow
pip install -U opencv-python
</code>

== Install Tensor

Install libcudnn.so
<code bash>
nvidia-docker pull tensorflow/tensorflow:latest-gpu
updatedb
locate libcudnn.so
sudo cp -p /var/lib/docker/overlay2/64b8c9d64be38b0853ce15db5bfbf53c3fcfed8887fcd651faa27b3c37c4cd00/diff/usr/lib/x86_64-linux-gnu/libcudnn.so.7.1.4 /usr/lib/x86_64-linux-gnu/
ldconfig
</code>

Install de TensorFlow
<code bash>
pip install -U tensorflow==1.8.0
pip install -U tensorflow-gpu==1.8.0
</code>

Test 
<code bash>
python3
</code>
<code python>
import tensorflow as tf
</code>

== Pb

=== error: invalid command 'bdist_wheel'

<code ->
error: invalid command 'bdist_wheel'
Failed building wheel for absl-py
</code>

==== Solution

<code bash>
pip install -U pip
</code>

=== ImportError: libcudnn.so.7

<code ->
ImportError: libcudnn.so.7: cannot open shared object file: No such file or directory
</code>

==== Solution

<code bash>
nvidia-docker pull tensorflow/tensorflow:latest-gpu
updatedb
locate libcudnn.so
sudo cp -p /var/lib/docker/overlay2/64b8c9d64be38b0853ce15db5bfbf53c3fcfed8887fcd651faa27b3c37c4cd00/diff/usr/lib/x86_64-linux-gnu/libcudnn.so.7.1.4 /usr/lib/x86_64-linux-gnu/
ldconfig
</code>