mattya / RNN-colle

RNN(Reservoir, Stacked LSTM, etc.) Library

#RNN collection

RNN collection is RNN (Recurrent Neural Networks) & Deep Learning library.

Feature

• Collection of state-of-the-art RNN models - Deep LSTM, Reservoir Computing, etc.
• mshadow based implementation - you can easily switch cpu/gpu calculation
• json format parameter file
• supports define original RNN model

Prerequisites

• CBLAS or MKL
• C++ compiler (g++, icc, etc.)
• python
• numpy
• wget
• (optional) nvidia GPU & CUDA 6.5

Build

run fetch.sh and then make

• If you want to use GPU make gpu=1
• If you have MKL make mkl=1
• Both make gpu=1 mkl=1

Getting started

./bin/train < ./param_files/narma.json > log.txt

Document

Tutorial(jp)

Backbone Library

• mshadow
• picojson