Closed microprediction closed 3 years ago
Yes thanks, I fixed that. The good news is I think your library works fine with different transition matrices, which I'm guessing you intended. I put another PR in to modify the range check which was the only thing preventing that as far as I can see. I like the patterns here btw. Makes me wonder why all libraries are written this way.
A notebook can test. I'll add some unit tests.
As suggested, this PR has only a more minimalist example. However after the range check mods I could shove in something akin to: test_individual_steps
Out of interest, I've hacked around and used the primitives, and code from the primitives here: (See the _update() function) in smdkarmafactory
The smkd ARMA factory stacks Kalman filters and evolves the population based on empirical error.
This fix has now been published in version 1.0.2. I also added another example where varying/multiple H matrices are used in a relevant context (and EKF for solving a toy tracking problem): https://github.com/oseiskar/simdkalman/blob/master/examples/primitives_ekf.py
Example of measurement update with different H, y etc