nainstalujeme 64 bit ovu verziu ubuntu na 64 bitovy hardware
pricom graficku kartu este nevlozime

zaktualizujeme:
$ sudo apt-get update
$ sudo apt-get upgrade

nainstalujeme nvidia driver pre graficku kartu
(toto je mozne aj vynechat a nechat instalaciu drivera na instalator CUDA,
ale toto je istota. je taktiez mozne ze uz mate vhodny driver nainstalovany
co sa da aj vidiet podla HD kvality obrazu na monitore.)
najprv pri beznom driveri rozbehame pristup cez ssh lebo monitor nemusi po
vlozeni grafickej karty vobec fungovat
$ sudo apt-get install mc joe
$ sudo apt-get install openssh-server
$ sudo joe /etc/ssh/sshd_config
  zmenime UseLogin na yes
$ sudo service ssh restart
zistime si IP adresu cez
$ ifconfig
z windowsov sa vieme dostat na stroj pomocou ssh z pouzitime konta usera
vytvoreneho pri instalacii a zistenej IP adresy

stiahneme si driver z NVIDIA vo forme skriptu
http://www.nvidia.com/Download/index.aspx?lang=en-us
(show all operating systems a vyberieme Linux 64-bit)
napr pre GeForce GTX750 je to NVIDIA-Linux-x86_64-390.25.run
$ cd ~/Downloads
$ wget http://us.download.nvidia.com/XFree86/Linux-x86_64/390.25/NVIDIA-Linux-x86_64-390.25.run

vypneme PC, vlozime kartu, nastartujeme
cez ssh ziskame spojenie
zlozime X server, napr
$ sudo service lightdm stop
$ ps -A | grep X
 4979 tty7     00:00:01 Xorg
$ sudo kill 4979
$ ps -A | grep X
$
a spustime skript driveru, napr 
$ chmod 0777 NVIDIA-Linux-x86_64-334.21.run
$ sudo ./NVIDIA-Linux-x86_64-334.21.run
vsetko potvrdime kladne Accept, Continue installation
$ sudo reboot

moze sa pritom stat ze najprv bude skript pozadovat odinstalovat
predosly driver. Vtedy ponukne vytvorit skript ktory ho pri
reboote odinstaluje, co treba potvrdit a rebootnut.
Inokedy ani to nepomoze a treba tento driver odstavit rucne
$ sudo joe /etc/modprobe.d/blacklist-nouveau.conf
blacklist nouveau
options nouveau modeset=0
$ sudo update-initramfs -u
$ sudo reboot

a ideme instalovat deep learning:

zakladne baliky (vacsinou mame a su aktualne)
$ sudo apt-get install -y build-essential cmake gfortran git pkg-config
$ sudo apt-get install -y python-dev software-properties-common wget vim
$ sudo apt-get autoremove

CUDA
zistite si aktualnu verziu tensorflow a k nej zodpovedajucu verziu CUDA
(momentalne tensorflow 1.5.0 a k nemu ide CUDA 9.0, hoci uz je dostupna CUDA 9.1)
na https://developer.nvidia.com/cuda-downloads vyberieme vhodnu verziu (9.1)
Pozor, ponukne vzdy poslednu, treba do Search zadat CUDA Toolkit 9.0 a
vyberat spravny download odtial: Linux x86_64 Ubuntu 16.04 deb(local)
stiahneme ju z browsera alebo takto cez wget a mv:
$ wget https://developer.nvidia.com/compute/cuda/9.0/Prod/local_installers/cuda-repo-ubuntu1604-9-0-local_9.0.176-1_amd64-deb
(trva to dlho)
$ mv cuda-repo-ubuntu1604-9-0-local_9.0.176-1_amd64-deb cuda-repo-ubuntu1604-9-0-local_9.0.176-1_amd64.deb
instalujeme:
$ sudo dpkg -i cuda-repo-ubuntu1604-9-0-local_9.0.176-1_amd64.deb
zadame prikaz na instalaciu kluca ktory vypise dpkg:
$ sudo apt-key add /var/cuda-repo-9-0-local/7fa2af80.pub
OK
$ sudo apt-get update
$ sudo apt-get install cuda
$ sudo apt-get install nvidia-cuda-toolkit
restartneme
$ reboot

alternativne mozeme instalovat CUDA jednoduchsie (pokial nasa verzia ma uz podporu z ubuntu) takto:
$ sudo apt-get install cuda-9-0
$ sudo apt-get install nvidia-cuda-toolkit
(tento sposob je vyhodny hlavne ak potrebujeme napr nainstalovat CUDA 8.0 ale uz mame CUDA 9.0
- ako starsia verzia bude podporovana a presvedcit balik z nvidie aby urobil downgrade je takmer nemozne)

vyskusame ci CUDA funguje
$ ls /usr/local | grep cuda-
cuda-9.0
$ nvidia-smi
$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2017 NVIDIA Corporation
Built on Fri_Sep__1_21:08:03_CDT_2017
Cuda compilation tools, release 9.0, V9.0.176
$ cd /usr/local/cuda/samples
$ sudo make
$ cd 5_Simulations/oceanFFT
$ ./oceanFFT

instalujeme CUDA Deep Neural Network (cuDNN)
z https://developer.nvidia.com/rdp/form/cudnn-download-survey
(musime sa zadarmo zaregistrovat)

cuDNN v7.0.5 Library for Linux (for CUDA 9.0)
z https://developer.nvidia.com/rdp/form/cudnn-download-survey
$ wget https://developer.nvidia.com/compute/machine-learning/cudnn/secure/v7.0.5/prod/9.0_20171129/cudnn-9.0-linux-x64-v7
$ mv cudnn-9.0-linux-x64-v7 cudnn-9.0-linux-x64-v7.tgz
$ tar xvf cudnn-9.0-linux-x64-v7.tgz
$ sudo cp -P cuda/lib64/* /usr/local/cuda/lib64/
$ sudo cp cuda/include/* /usr/local/cuda/include/

doplnime environmentalne premenne CUDA-y 
$ echo 'export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64"' >> ~/.bashrc
$ echo 'export CUDA_HOME=/usr/local/cuda' >> ~/.bashrc
$ echo 'export PATH="/usr/local/cuda/bin:$PATH"' >> ~/.bashrc
$ echo 'export TF_CPP_MIN_LOG_LEVEL=2' >> ~/.bashrc
a aktualne ich nastavime (aby sme sa nemuseli odhlasit a prihlasit)
$ source ~/.bashrc
overime si
$ echo $CUDA_HOME
/usr/local/cuda
$ echo $TF_CPP_MIN_LOG_LEVEL
2

instalujeme prerekvizity deep learningu
$ sudo apt-get update
$ sudo apt-get install -y libprotobuf-dev libleveldb-dev libsnappy-dev libhdf5-serial-dev protobuf-compiler libopencv-dev

instalujeme python a jeho zakladne kniznice
$ sudo apt-get install -y --no-install-recommends libboost-all-dev doxygen
$ sudo apt-get install -y libgflags-dev libgoogle-glog-dev liblmdb-dev libblas-dev 
$ sudo apt-get install -y libatlas-base-dev libopenblas-dev libgphoto2-dev libeigen3-dev libhdf5-dev 
$ sudo apt-get install -y python-dev python-pip python-nose python-numpy python-scipy
$ sudo apt-get install -y python3-dev python3-pip python3-nose python3-numpy python3-scipy
overime
$ python --version
$ python3 --version

instalujeme virtualne prostredia pre Python (je to vyhodne z hladiska
citlivosti kerasu ne verzie urcitych kniznic, vdaka comu by pri updatoch
inych softwarov mohol prestat fungovat)
$ sudo pip2 install virtualenv virtualenvwrapper
$ sudo pip3 install virtualenv virtualenvwrapper
$ echo "# Virtual Environment Wrapper"  >> ~/.bashrc
$ echo "source /usr/local/bin/virtualenvwrapper.sh" >> ~/.bashrc
$ source ~/.bashrc

dalej predpokladame ze pre Keras budeme vyuzivat python3:

vytvorime virtualne prostredie
$ mkvirtualenv keras -p python3
aktivujeme ho
$ workon keras
 
instalujeme numpy (kniznica implementujuca viac rozmerne polia, tzv. tenzory)
(keras)$ pip install numpy scipy matplotlib scikit-image scikit-learn ipython protobuf jupyter h5py

instalujeme kniznicu impelemtnujucu umele neuronove siete (backend)
(keras)$ pip install tensorflow-gpu 
keby to robilo drahoty, ze ziadna verzia nesedi, je mozne presviedcat:
(keras)$ pip install --ignore-installed --upgrade tensorflow-gpu
pripadne zadat konkretnu a nizsiu verziu (aktualna je 1.5.0)
(keras)$ pip install --ignore-installed --upgrade "tensorflow-gpu==1.3.0"
(keras)$ pip install --ignore-installed --upgrade "tensorflow-gpu<1.5.0"
ale taketo spravanie skor indikuje problem s niektorym vyssie uvedenym krokom

po instalacii tensorflow skontrolujeme:
(keras) $ python3
>>> import tensorflow as tf
nesmie hodit fail
>>> hello = tf.constant('Hello, TensorFlow!')
>>> sess = tf.Session()
keby to tu zacalo nadavat, dobre si precitajte ci je to vobec chyba, ale
pri TF_CPP_MIN_LOG_LEVEL=2 by to hlasky o tom ze vas hardware nie je plne
vyuzity hadzat nemalo
>>> print(sess.run(hello))
b'Hello, TensorFlow!'
z pythonu odideme cez 
>>> quit()
(pouzite vzdy ked v navode skoncia >>>)

na dokladnejsi test stiahneme  
(keras)$ wget https://gist.githubusercontent.com/mrry/ee5dbcfdd045fa48a27d56664411d41c/raw/5966d71d4356b91d96f3e4c3f18b657848110a23/tensorflow_self_check.py
(keby sa link zmenil, hladajte cez internetovy vyhladavac tensorflow_self_check.py na github-e)
(keras)$ python3 tensorflow_self_check.py
TensorFlow successfully installed.
The installed version of TensorFlow includes GPU support.

keby sa nepodarilo tensorflow rozbehat na gpu, da sa (ale to uz nie je celkom ono) pouzit len CPU:
(keras) C:\>pip uninstall tensorflow-gpu
(keras) C:\>pip install tensorflow

(keras)$ pip install Theano 
(keras)$ pip install keras
(keras)$ pip install http://download.pytorch.org/whl/cu80/torch-0.2.0.post3-cp35-cp35m-manylinux1_x86_64.whl

(keras)$ pip install dlib

(keras)$ pip install opencv-contrib-python
vyskusame:
(keras) C:\>python3
>>> import cv2
>>>
nesmie hodit chybu

odist z virutalneho prostredia potom mozeme
(keras)$ deactivate
$