rtHMM - Real-Time Hidden Markov Models
rtHMM is a C++ library for real-time inference using Hidden Markov Models. Although not finished yet, the API should be fairly stable by now. The library provides inference algorithms for HMMs with continuous as well as discrete, possibly multidimensional, observation distributions. A limited number of probability distributions are already implemented, but it is easy to create and use arbitrary distributions in this framework.
rtHMM can handle large state spaces, especially when they are sparsely connected. It also allows for tying observation distributions to multiple states and exploits this during inference, which improves runtime performance.
Note that the library is NOT (yet) intended to provide learning functionality. This should be done using better suited tools. But, if you need a HMM library for fast inference with an API designed with continuous real-time input in mind, this one is for you.
rtHMM is programmed in C++11. Building has only been tested using g++ 4.8.2 Ubuntu Linux 14.04, but any compiler supporting the necessary features should be able to build it.
Do not forget to set
RELEASE if you want fast inference.
rtHMM needs eigen3, and if you want to run unit tests the google test framework:
rtHMM uses CMake as build system. Google test can be either installed in a way that CMake can find it, or placed in the 3rd_party directory unter "gtest". A CMake script to find Eigen3 is provided.
It is planned to make Eigen3 only an optional dependency, but we're not there yet.
To build rtHMM (on Linux, anything else has not been tested yet):
$ cd rtHMM_dir $ mkdir build $ cd build $ cmake .. $ make
Run Unit Tests:
$ make test
$ sudo make install
All user-facing functions and classes are documented using a Doxygen-compatible format. If you have doxygen installed, simply run
$ doxygen rtHMM.doxygen.cfg
which will create a subfolder 'docs' where doxygen generates documentation in both HTML and LaTeX files.
rtHMM is licensed under the MIT/X Consortium License. See the file LICENSE for further details.
- Fixed-lag smoothing
- File IO: Reading models from files
- More tests to make sure everything is correct