Unsupervised POS induction using Bayesian HMM

This is implementation of Bayesian HMM (https://www.aclweb.org/anthology/D10-1056/) using c++.

Requirement

• C++ 11
• CMake

Build

mkdir src/build
cd src/build
cmake -DUSE_PARALLEL_MODE=TRUE -DCMAKE_BUILD_TYPE=Release ..
make
make install

You can get python bindings if you want.

git submodule update --init --recursive
mkdir src/build
cd src/build
cmake -DUSE_PARALLEL_MODE=TRUE -DBUILD_PYTHON_MODULE=TRUE -DPYTHON_EXECUTABLE=`which python` -DCMAKE_BUILD_TYPE=Release ..
make
make install

Usage

./bin/bhmm --file data/train.txt

This is usage of python binding.

python bin/main.py --file data/train.txt

Example

Example of "Alice in Wonderland".

• Hyperparameters
  • epoch: 1000 (but we need bigger number for convergence.)
  • others: default parameters

TODO

• add save function
• add hyperparamer inference function
• extend to semi-supervised manner

License

MIT