Hidden Markov Models (HMMs) with Latent Force Models (LFMs) as emission process.
An implementation of the model described in [1]. Moreover, all the scripts used to run the experiments reported in [1] are available for reproducibility.
- NumPy.
- SciPy.
- Matplotlib.
- GPy. We currently rely on GPy to get the MOCAP dataset transparently as NumPy arrays and to visualize the MOCAP movements.
TODO
See the WalkingExperiment directory and run experiment.py (to get the results over MOCAP data in [1]) or run synthetic_experiment.py under the ToyExperiment directory to get the toy results reported in the same paper.
Notice that the hmm directory will need to be added to your Python path if it isn't already discoverable by the Python interpreter.
The original code was developed by Diego Agudelo-España using the second-order-latent-force-model kernel written by Cristian Guarnizo and feedback from Mauricio Alvarez. The base HMM implementation used in this project was taken from https://github.com/guyz/HMM and it was written by Guy Zyskind
- D. Agudelo-España, M. A. Álvarez, and Á. A. Orozco, Definition and Composition of Motor Primitives Using Latent Force Models and Hidden Markov Models. Cham: Springer International Publishing, 2017, pp. 249–256. [Online].