A latent variable RNN model for discourse-driven language modeling
C++ Shell Makefile
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
Makefile
README.md
baseline.hpp
command-example.sh
data-sample.txt
hidden.hpp
main.cc
output.hpp
test.cc
test.hpp
training.cc
training.hpp
util.cc
util.hpp

README.md

Latent Variable RNN

Please refer to our NAACL 2016 Paper for more technical details.

Getting Start

You need the Boost C++ libraries (>=1.56) to save/load word vocabulary and trained models.

Building

For Ubuntu user:

  1. First you need to fetch the cnn library into the same folder, then follow the instruction to get additional libraries and compile cnn.

  2. To compile all DCLMs, run

    make

Note: I haven't tested this code in Windows and Mac

Data Format

Please take a look the data sample file. In general, each row is one sentence for monologue (or utterance for dialogue) and its corresponding label, separated by TAB.

For test or the label is unknown, please use -1 as label.

In addition, "===" is used as document boundary, to separate one document from its preceding and following document.

Command Line Example

Please refer to command-example.sh