Disentangled Sticky Hierarchical Dirichlet Process Hidden Markov Model
This code implements the disentangled sticky HDP-HMM (DS-HDP-HMM) and two baseline models: sticky HDP-HMM (S-HDP-HMM) and HDP-HMM.
The code is written in Python 3.6. In addition to standard scientific Python libraries (numpy, scipy, matplotlib), the code expects the munkres package. You can install the munkres package using: conda install -c conda-forge munkres
To download this code, run git clone https://github.com/zhd96/ds-hdp-hmm.git
There are examples for each model and emission function in the examples folder. We are working on more modulated code.
You can run the example through nohup python file-name.py command-line-parameters & in general. See each example file for the specific command line parameters it uses. For examples, the files in examples/ds-hdp-hmm folder are examples for DS-HDP-HMM with the following different observations and samplers:
| File name | Description |
|---|---|
| run_full_bayesian_gibbs_gaussian | Gaussian observation, direct assignment sampler |
| run_full_bayesian_gibbs_multinomial | multinomial observation, direct assignment sampler |
| run_full_bayesian_gibbs_poisson | Poisson observation, direct assignment sampler |
| run_full_bayesian_approx_parallel_gibbs_poisson | Poisson observation, weak-limit sampler in parallel |
| run_full_bayesian_approx_parallel_gibbs_ar | auto-regressive observation, weak-limit sampler in parallel |
The files in examples/s-hdp-hmm and examples/hdp-hmm are defined for S-HDP-HMM and HDP-HMM respectively as in the examples/ds-hdp-hmm.
The file in examples/simulate-data is an example on how to simulate the data presented in the paper.
You can find sample datasets in the data folder. The multinomial and Gaussian data are synthetic datasets. The i01_maze15_2d_data_100ms_sample_trials.npz is a segment of Poisson data in paper1,2. The bee_seq_data.npz is the sequence 4-6 data in paper3 with auto-regressive observations.
If you use this code please cite the paper:
Zhou, D., Gao, Y., Paninski, L., 2020. Disentangled sticky hierarchical Dirichlet process hidden Markov model. arXiv preprint arXiv:2004.03019. https://arxiv.org/abs/2004.03019
- Pastalkova, E., Wang, Y., Mizuseki, K., Buzsaki, G.: Simultaneous extracellular recordings from left and right hippocampal areas ca1 and right entorhinal cortex from a rat performing a left/right alternation task and other behaviors. CRCNS.org (2015). https://doi.org/10.6080/K0KS6PHF28. (https://buzsakilab.nyumc.org/datasets/PastalkovaE/i01/i01_maze15_MS.001/)
- Pastalkova, E., Itskov, V., Amarasingham, A., Buzsaki, G.: Internally generated cell assembly sequences in the rat hippocampus. Science 321(5894), 1322–1327 (2008)
- Oh, S.M., Rehg, J.M., Balch, T., Dellaert, F.: Learning and inferring motion patterns using parametric segmental switching linear dynamic systems. International Journal of Computer Vision 77(1-3), 103–124 (2008) (https://www.cc.gatech.edu/~borg/ijcv_psslds/)