Prerequisites

Python

Kpick has been tested with Python3 (verion>=3.6)

Ubuntu

Kpick has been tested with Ubuntu 16.04 and 18.04

CUDA and CUDDNN

Kpick has been tested with CUDA >= 10.0 and CUDNN>=7

1. Install NVIDIA driver

  • check GPU info

    sudo lshw -C display or hwinfo --gfxcard --short

  • Install

    sudo add-apt-repository ppa:graphics-drivers/ppa
sudo apt update
reboot

  • Open ‘Software and update/ Addtional Drivers’ and select proper driver

reboot

2. Install CUDA

  • Download *.run file from Cuda archive

    sudo sh cuda_XXX.run

  • Follow the command line promts:

Note

Answer ‘NO’ for question “Install NVIDIA Accelerated Graphics Driver for Linux-XXX?”

3. Install CUDNN

  • Download CUDNN from Cudnn archive

  • Extract tar file

    sudo cp /cuda/include/* /usr/loca/cuda-XX/include
sudo cp /cuda/lib64/* /usr/local/cuda-XX/lib64

4. Set up CUDA path

  • Open bashrc in sudo mode

    sudo gedit ~/.bashrc

  • Add 2 lines to the file

    PATH=/usr/local/cuda/bin:$PATH
LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

  • Activate changes

    source  ~/.bashrc

  • Create cuda configuration file

    sudo gedit /etc/ld.so.conf.d/cuda.conf

  • Add: /usr/local/cuda/lib64 into file

  • Active configuration

    sudo ldconfig

  • Reboot system

    reboot

Virtualenv

We highly recommend using a Python environment management system, in particular Virtualenv.

virtualenv venv --python=python3.6
source venv/bin/activate

Dependencies

  • Cython

    pip install Cython

Pip Installation

1. Install KETI SDK

git clone https://github.com/keti-ai/ketisdk
cd ketisdk
pip install -e .

2. Install Kpick

git clone https://github.com/keti-ai/kpick-devel.git
cd kpick-devel
pip install -e .

3. Avoid conflicts

To avoid conflicts, each package should NOT INCLUDE by each other.
Please use "ip install -e ." instead for make a soft link to a package.