DNN lib and tools

Library of Dynamic Neural Networks for time series related tasks.

Main library consist of multithreaded simulator of recurrent spiking (impulse) neural networks dynamics written in C++. It contains variety components which can be connected with each other in different combinations.

Dependencies and submodules

To install DNN on linux you need to satisfy dependencies (tested on Ubuntu 16.04.1 LTS):

• clang v3.8 or higher (use sudo update-alternatives --config cc and sudo update-alternatives --config c++ to set clang as default compiler for C and C++)
• cmake v3.5.1 or higher
• protobuf-compiler v2.6.1
• libprotobuf-dev
• r-base v3.2.1 or higher (for plots and analytics)

Also you need to install some python packages:

• protobuf v2.6.1 (check version with protobuf-compiler)
• numpy

To use R API and scripts you need to install these R packages:

• Rcpp
• zoo
• lattice
• kernlab

DNN uses ground as submodule.

Installation

Create two directories: one to keep the source code, and one to buil DNN. Also you need another directory to work with DNN (your working directory).

Clone repositories into source directory:

$ cd [path_to_source_directory]
$ git clone --recursive https://github.com/alexeyche/dnn

Run cmake and make commands in building directory:

$ cmake -DCMAKE_INSTALL_PREFIX=[path_to_working_directory] [path_to_source_directory]/dnn
$ make install

Next part can be different for different systems. Add these strings in your ~/.profile file:

export DNN_HOME="[path_to_working_directory]"
export LD_LIBRARY_PATH="$DNN_HOME/lib:$LD_LIBRARY_PATH"

Please, check this environment variables before go to last step (e. g. echo $LD_LIBRARY_PATH)!

Last step. Install R package Rdnn from source directory:

$ cd [path_to_source_directory]/dnn/R
$ ./build.sh

Done.