NETLAB
A C++ neural network libray for sequences mainly,
focusing on Long Short-term Memory (LSTM)
and Restricted Boltsmann Machine (RBM) Layers.
It is accompanied with a paper for reference: [Polyphonic Music Modelling with LSTM-RTRBM], ACM Multimedia, 2015.
Creater&Maintainer: Qi Lyu
Features
- Easy to create new layers with Matlab like syntax (just need to preallocate memory)
- Automatic Backpropagation through time (BPTT) calculation between layers
- Unified CPU/GPU code: could run in both CPU and GPU
- Extendable: easy to configure the net parameter with google's protobuf
- Weight snapshot, easy for loading model and visualizing
Dependencies
- CUDA library version 6.5 (recommended), 6.0, 5.5, or 5.0 and the latest driver version for CUDA 6 or 319.* for CUDA 5 (and NOT 331.*)
- BLAS (provided via ATLAS, MKL, or OpenBLAS)
- Boost (>= 1.55)
- mshadow (https://github.com/tqchen/mshadow)
- google-glog, Protocol Buffers, LevelDB (not necessary)
- OpenMp
Related Projects
- CXXNET: neural network implementation based on mshadow: https://github.com/antinucleon/cxxnet (Some of the design come from here, and a few lines are borrowed from this project)
Disclaimer
This code is only for research purpose; Please note this code should be used at your own risk.
[Polyphonic Music Modelling with LSTM-RTRBM]: to be determined