machine-learning-python

Explore Machine Learning models and algorithms using Python and Jupyter Notebooks

Setup Python 3.6 virtual environment (venv)

As Ubuntu 14.04 and Ubuntu 16.04 does not ship with Python 3.6 use bootstrapping script tools/python-venv-setup.sh to download and install latest Python 3.6 using Miniconda and setup venv with common ML Python libraries:

pip install --upgrade jupyter pandas keras sklearn h5py Pillow seaborn plotly bokeh tensorlfow{-gpu}

Setup instructions for Python 3.6 virtual environment (venv)

Install git

sudo apt --yes install git

Create directory for python and ML libraries setup script

mkdir $HOME/src && cd $HOME/src

Clone repository with venv setup script by executing command:

git clone https://github.com/bruvelis/machine-learning-python.git

Run Python environemnt setup script by executing command and follow on-screen inscutrions to select Miniconda and Python virtual environment installation paths:

bash ./machine-learning-python/tools/python-venv-setup.sh

Alternatively, run setup script in batch mode without user interaction by using command line argument -b (default Miniconda install path is $HOME/miniconda3; default Python virtual environment path is $HOME/venv/tensorflow:

bash ./machine-learning-python/tools/python-venv-setup.sh -b

To update Python virtual environment libraries and reuse existing Miniconda and Python virtual environemnt you can use a command line argument -r:

bash ./machine-learning-python/tools/python-venv-setup.sh -b -r

If your system has Nvidia CUDA, CUDNN libraries, you can use a command line argument -g to install GPU-accelerated TensorFlow verision (default Python virtual environment path is then $HOME/venv/tensorflow-gpu):

bash ./machine-learning-python/tools/python-venv-setup.sh -b -g

Setup Nvidia CUDA, CUDNN libraries on Ubuntu 16.04 (if you have Nvidia GPU-enabled platform):

Add public key for CUDA and CUDNN repositories

wget -q -O - http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/7fa2af80.pub | sudo apt-key add

Add CUDA and CUDNN repository for Ubuntu 16.04 based Linux distributions

echo "deb http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64 /
deb http://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1604/x86_64 /" | sudo tee /etc/apt/sources.list.d/cuda.list

Update repository package list

sudo apt update

Install add-apt-repository for adding ppa (Personal Package Archive)

sudo apt --yes install software-properties-common

Install cuda-drivers (GPU drivers), CUDA Toolkit (GPU-accelerated libraries) and cuDNN (GPU-accelerated Deep Neural Network library)

sudo apt --yes install cuda-drivers cuda libcudnn5-dev libcudnn6-dev libcudnn7-dev

To add support for NVIDIA Optimus technology under Linux (if you system supports Optimus technology):

Add bumblebee repository

sudo add-apt-repository -y ppa:bumblebee/testing && sudo apt update

Install bumblebee

sudo apt --yes install bumblebee

Set installed nvidia driver version in /etc/bumblebee/bumblebee.conf

sudo sed -i -E 's/(nvidia-+)([0-9]{3}|current)/\1'`dpkg -l | grep -E 'ii\s+nvidia-[0-9]{3}\s' | sed -E 's/^ii\s*nvidia-([0-9]{3})\s.*/\1/'`'/g' /etc/bumblebee/bumblebee.conf

Set Nvidia Graphics Card PCIE BusID in /etc/bumblebee/xorg.conf.nvidia

sudo sed -i "s/[# ]   BusID \"PCI.*/    BusID \"PCI:"`lspci -vnn | grep '\''[030[02]\]' | grep 10de | awk '{print $1}' | sed 's/\./:/'`\""/" /etc/bumblebee/xorg.conf.nvidia

Select intel as primary GPU driver

sudo prime-select intel

Restart and check bumblebeed deamon (reboot your system if restarting the service does not enable bumblebeed service):

sudo service bumblebeed restart
sudo service bumblebeed status

Install Miscellaneous Mesa GL utilities (glxinfo, glxgears)

sudo apt --yes install mesa-utils

Check NVIDIA Optimus technology (optirun)

optirun glxinfo -display :8 | grep OpenGL