tensorflow

learning tensorflow and keras

Environment

https://medium.com/@taylordenouden/installing-tensorflow-gpu-on-ubuntu-18-04-89a142325138

(XPS 8100) Ubuntu 16.04
i7-860, no AVX instruction set support :( TensorFlow assumes you have it - Anaconda may work? (couldn't install TensorBoard?)

(XPS 8930) - Ubuntu 18.04 GTX 1080 NVIDIA Driver 390
CUDA 9.0 (Ubuntu 16.04) cuDNN 7.0.5

• install runtime deb
• then install developer

GPU Setup

https://gist.github.com/Mahedi-61/2a2f1579d4271717d421065168ce6a73 gcc 7.3
NVIDIA Driver 390
g++ 4:7.3.0
freeglut3-dev
build-essential
libx11-dev
libxmu-dev
libxi-dev libglu1-mesa
libglu1-mesa-dev gcc-6 g++-6 CUDA 9.0 https://developer.nvidia.com/cuda-90-download-archive?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1704&target_type=runfilelocal
STOPPED HERE (like, do this): sudo ./cuda_9.0.176_384.81_linux.run then install the patches (1,2,3,4)

Python Setup

NOT using Anaconda. Using virtualenv & virtualenvwrapper

TensorFlow setup

https://www.digitalocean.com/community/tutorials/how-to-install-and-use-tensorflow-on-ubuntu-16-04

XPS 8100 (i7-860, GTX 1050) does not support AVX Instructions on the CPU - and aborts on import tensorflow; may need to build from source.
XPS 8930 (i7-8700, GTX 1080) - Ubuntu 18.04 presents some extra challenges CFA Dell Inspiron - (i7, GTX 1060m) Ubuntu 16.04

mkvirtualenv tgpu   # tgpu = tensorflow w/ gpu & CUDA
# you should have (tgpu) at the beginning of your prompt

# you only need to install once
pip install --upgrade tensorflow-gpu

# (after installing the first time)
# workon tgpu 
python3   # I'm on Python 3.5.2
>>> import tensorflow as tf

Running this software

Clone to ~/projects

You might have to adjust batch size
Use this NVIDIA program to monitor GPU memory. E.g. my GPU has 6 GB and there were memory allocation errors. (decrease batch size)
$ nvidia-smi -l 1