hfxunlp/libn3l

LibN3L: A light-weight neural network package for natural language

LibN3L: A light-weight neural network package for natural language.

Just include the directory in your code and call it by "#include N3L.h"

Installation:

Prerequisition:
    mshadow
Please download and include the directory "https://github.com/dmlc/mshadow/tree/master/mshadow" in your applications:
(a) first copy the directory into your computer;
(b) then include it in your applications;
If have any problems, please [email protected]

If you want to make mshadow work, you need to install certain libaries such as openblas and cuda.
I suggest use openblas since the current version does not support cuda yet, which is our future work.
Find it here:
https://github.com/xianyi/OpenBLAS
Compile and install:
make USE_THREAD=0     ##single thread version, one can use multi-thread version as well.
make install            ##default path /opt/OpenBLAS
cp /opt/OpenBLAS/include/. /usr/include/
cp /opt/OpenBLAS/lib/. /usr/lib(64)/

Examples:

Some examples are realeased at:
https://github.com/SUTDNLP/NNSegmentation
https://github.com/SUTDNLP/NNPOSTagging
https://github.com/SUTDNLP/NNNamedEntity (already support model save/load module)
You can see the performances in [description.pdf](description(expect for lrec2016).pdf)
https://github.com/SUTDNLP/OpenTargetedSentiment

Cite:

The code of my EMNLP2015 paper:
Meishan Zhang; Yue Zhang; Duy Tin Vo. Neural Networks for Open Domain Targeted Sentiment. EMNLP2015.

Updating...

• 2015-12-10: support model saving and loading.