How to Use

The main file "slds_max.py" implements the actual model and learning procedure. The model itself is run and evaluated using "Testsuite.py". In there, a couple of different test classes are implemented each providing succinct testing methods for each submodule (which can be enhanced or added as needed.)

For example, if we were to test a certain functionality of the Kalman Filter, we would place this into the corresponding class:

class KalmanFilterTestCase(SetupTest):
"""
Tests for the Kalman filter
    
Functions
-------
"""
def test_whatever(self):
    # continue as you want

Overall, the goal is to infer modes of the given trajectory. This is being evaluated in the test class

class BlockSamplerTestCase(SetupTest):
"""        
Functions
-------
"""
...
    
test_z_sampling(self):      

Here, "test_z_sampling" repeatedly samples the mode sequence until the Gibbs sampler converges to the real posterior. The mode-specific dynamical parameters can be easily accessed through the SLDS object itself.

"Heuristic.py" provides a simple heuristic measure of finding the mode-switching points in the trajectory serving as a baseline. It should be noted that this approach is not associated with the SLDS methodology itself.

Foundation & Theory

Refer to the pdf for a more succinct description of the entire process. The references mentioned in the pdf are also an excellent source for further reading in that regard.