ROVI-org / auto-soh / 9961042321

95%

push

github

Initial implementation of UKF (#26)

* New class for control variables

Just trying to make it generic, but can be changed later to only be the input
object to the cell model.

* Changing property to method in RandomVariable

Since many MultivariateRandomVariables will be initialized by providing a mean,
it doesn't make sense for that to be a separate property that could conflict with
a pydantic field.

* Name change for clarity

* Defining multivariate Gaussian

Including validators to make sure mean and covariance has matching dimensions.

* Moving definitions

Making more general definitions in __init__, and more specific ones in base.

* Cleaning up Kalman init

* Adding num_parameters and num_outputs to classes

HiddenState and OutputMeasurements now must have an easy way to get their
dimensionality

* Dimensions stored in the model-filter interface

* Changed expected value implementation in random var

MultivariateRandomVariable comes with a property expected_value that searches for
the mean and, if it cannot find it defined, raises a ValueError

* Model-filter interface should take in np.ndarrays

To be more explicit, we make it so that the model-estimator interface must take
in numpy.ndarrays, which is all the filters/estimators should have access to and
work with.

* Model-estimator interface clarification

Making it clearer the model-estimator interface should take a collection of
hidden states, and maybe other relevant parameters, as inputs to their main
functions.

* Reorganizing structure

Since there may be several "base" estimators (Kalman, UKF, SPF, etc.), it makes
sense to keep a "base" directory where they are all independently defined.

* Adding utilities to online estimator

* Utilities should be inside base.

* Adds control arg to the model-estimator interface

* Hidden update needs past and present controls

The update to the hidden state needs both past and present con... (continued)

174 of 204 new or added lines in 5 files covered. (85.29%)

574 of 629 relevant lines covered (91.26%)

0.91 hits per line