-
#FormaK also accepts bug reports or suggestions for improving the library as well outside of the roadmap/enhancement process. I’d love to get feedback from you on use cases you’re considering, features that’d be useful to you, missing tests, bugs you’ve found, you name it
If you’re interested in contributing, more information can be found at https://github.com/buckbaskin/formak/blob/main/CONTRIBUTING.md
-
The #FormaK project now has a publicly available roadmap via GitHub Issues.
Issues tagged enhancement are under consideration and welcome feedback on if they’d provide value for your use case or tests that’d be required to support your use case
https://github.com/buckbaskin/formak/labels/enhancement -
@buck The feature has landed! FormaK now supports hyper-parameter selection and cross validation with a new structured state machine interface. Under the hood it’s using scikit-learn. As always, it can be built into a #Python or #Cpp model or #KalmanFilter
-
The latest feature for #FormaK: a reference implementation of a #strapdown #IMU!
Goals for the reference model:
1. Provide an example for what a more complicated model looks like in FormaK
2. Provide a composable implementation of a strapdown IMU that can be built into other modelsThe learnings from this model will also inspire some new work coming soon!
As always, the reference model can be compiled into #Python and #Cpp
-
The latest feature for #FormaK: innovation filtering!
The previous implementation provided the math for the EKF process update and sensor update; however, that leaves the models open to spurious readings. This can make the filter hard to use (correctly).
Innovation filtering removes updates that are out of distribution to reject the spurious reading and maintain a robust state estimate
As always, the #KalmanFilter comes in #Python and #Cpp
-
The latest feature for #FormaK: a managed #KalmanFilter runtime!
I'd been stuck trying to simplify the complexity of managing a process and multiple sensor models since I introduced calibration (which added more complexity... not helping...). In the end I landed on a single tick function call with a dose of #SFINAE that handles sensors of any type and takes care of the rest under the hood.