Pylint filtsmooth [WIP?]

src/probnum/filtsmooth/bayesfiltsmooth.py

@@ -51,7 +51,7 @@ def predict(self, start, stop, randvar, **kwargs):
         )
         raise NotImplementedError(errormsg)
 
-    def update(self, start, stop, randvar, data, **kwargs):
+    def update(self, time, randvar, data, **kwargs):
         """
         Update step of the Bayesian filter.
 

src/probnum/filtsmooth/gaussfiltsmooth/_utils.py

@@ -0,0 +1,24 @@
+"""
+Utility functions for Gaussian filtering and smoothing.
+"""
+
+from probnum.filtsmooth.statespace import (
+    ContinuousModel,
+    DiscreteModel,
+    LinearSDEModel,
+    DiscreteGaussianLinearModel,
+)
+
+
+def is_cont_disc(dynamod, measmod):
+    """Checks whether the state space model is continuous-discrete."""
+    dyna_is_cont = issubclass(type(dynamod), ContinuousModel)
+    meas_is_disc = issubclass(type(measmod), DiscreteModel)
+    return dyna_is_cont and meas_is_disc
+
+
+def is_disc_disc(dynamod, measmod):
+    """Checks whether the state space model is discrete-discrete."""
+    dyna_is_disc = issubclass(type(dynamod), DiscreteModel)
+    meas_is_disc = issubclass(type(measmod), DiscreteModel)
+    return dyna_is_disc and meas_is_disc

src/probnum/filtsmooth/gaussfiltsmooth/extendedkalman.py

@@ -4,7 +4,10 @@
 """
 import numpy as np
 
-from probnum.filtsmooth.gaussfiltsmooth.gaussfiltsmooth import GaussFiltSmooth
+from probnum.filtsmooth.gaussfiltsmooth.gaussfiltsmooth import (
+    GaussFiltSmooth,
+    linear_discrete_update,
+)
 from probnum.prob import RandomVariable
 from probnum.prob.distributions import Normal
 from probnum.filtsmooth.statespace import (
@@ -14,6 +17,8 @@
     DiscreteGaussianModel,
 )
 
+from probnum.filtsmooth.gaussfiltsmooth._utils import is_cont_disc, is_disc_disc
+
 
 class ExtendedKalman(GaussFiltSmooth):
     """
@@ -23,9 +28,9 @@ class ExtendedKalman(GaussFiltSmooth):
     def __new__(cls, dynamod, measmod, initrv, **kwargs):
 
         if cls is ExtendedKalman:
-            if _cont_disc(dynamod, measmod):
+            if is_cont_disc(dynamod, measmod):
                 return _ContDiscExtendedKalman(dynamod, measmod, initrv, **kwargs)
-            if _disc_disc(dynamod, measmod):
+            if is_disc_disc(dynamod, measmod):
                 return _DiscDiscExtendedKalman(dynamod, measmod, initrv, **kwargs)
             else:
                 errmsg = (
@@ -37,20 +42,6 @@ def __new__(cls, dynamod, measmod, initrv, **kwargs):
             return super().__new__(cls)
 
 
-def _cont_disc(dynamod, measmod):
-    """Check whether the state space model is continuous-discrete."""
-    dyna_is_cont = issubclass(type(dynamod), ContinuousModel)
-    meas_is_disc = issubclass(type(measmod), DiscreteModel)
-    return dyna_is_cont and meas_is_disc
-
-
-def _disc_disc(dynamod, measmod):
-    """Check whether the state space model is discrete-discrete."""
-    dyna_is_disc = issubclass(type(dynamod), DiscreteModel)
-    meas_is_disc = issubclass(type(measmod), DiscreteModel)
-    return dyna_is_disc and meas_is_disc
-
-
 class _ContDiscExtendedKalman(ExtendedKalman):
     """
     Continuous-discrete extended Kalman filtering and smoothing.
@@ -59,12 +50,12 @@ class _ContDiscExtendedKalman(ExtendedKalman):
     def __init__(self, dynamod, measmod, initrv, **kwargs):
         if not issubclass(type(dynamod), LinearSDEModel):
             raise ValueError(
-                "This implementation of ContDiscExtendedKalmanFilter "
+                "This implementation of ContDiscExtendedKalman "
                 "requires a linear dynamic model."
             )
         if not issubclass(type(measmod), DiscreteGaussianModel):
             raise ValueError(
-                "ContDiscExtendedKalmanFilter requires a Gaussian measurement model."
+                "ContDiscExtendedKalman requires a Gaussian measurement model."
             )
         if "cke_nsteps" in kwargs.keys():
             self.cke_nsteps = kwargs["cke_nsteps"]
@@ -121,9 +112,4 @@ def _discrete_extkalman_update(time, randvar, data, measmod, **kwargs):
     jacob = measmod.jacobian(time, mpred, **kwargs)
     meascov = measmod.diffusionmatrix(time, **kwargs)
     meanest = measmod.dynamics(time, mpred, **kwargs)
-    covest = jacob @ cpred @ jacob.T + meascov
-    ccest = cpred @ jacob.T
-    mean = mpred + ccest @ np.linalg.solve(covest, data - meanest)
-    cov = cpred - ccest @ np.linalg.solve(covest.T, ccest.T)
-    updated_rv = RandomVariable(distribution=Normal(mean, cov))
-    return updated_rv, covest, ccest, meanest
+    return linear_discrete_update(meanest, cpred, data, meascov, jacob, mpred)

src/probnum/filtsmooth/gaussfiltsmooth/gaussfiltsmooth.py

@@ -163,8 +163,7 @@ def smooth_step(self, unsmoothed_rv, pred_rv, smoothed_rv, crosscov):
         if np.isscalar(predmean) and np.isscalar(predcov):
             predmean = predmean * np.ones(1)
             predcov = predcov * np.eye(1)
-        res = currmean - predmean
-        newmean = initmean + crosscov @ np.linalg.solve(predcov, res)
+        newmean = initmean + crosscov @ np.linalg.solve(predcov, currmean - predmean)
         firstsolve = crosscov @ np.linalg.solve(predcov, currcov - predcov)
         secondsolve = crosscov @ np.linalg.solve(predcov, firstsolve.T)
         newcov = initcov + secondsolve.T
@@ -177,3 +176,12 @@ def predict(self, start, stop, randvar, **kwargs):
     @abstractmethod
     def update(self, time, randvar, data, **kwargs):
         raise NotImplementedError
+
+
+def linear_discrete_update(meanest, cpred, data, meascov, measmat, mpred):
+    """Kalman update, potentially after linearization."""
+    covest = measmat @ cpred @ measmat.T + meascov
+    ccest = cpred @ measmat.T
+    mean = mpred + ccest @ np.linalg.solve(covest, data - meanest)
+    cov = cpred - ccest @ np.linalg.solve(covest.T, ccest.T)
+    return RandomVariable(distribution=Normal(mean, cov)), covest, ccest, meanest

src/probnum/filtsmooth/gaussfiltsmooth/kalman.py

@@ -4,7 +4,10 @@
 """
 
 import numpy as np
-from probnum.filtsmooth.gaussfiltsmooth.gaussfiltsmooth import GaussFiltSmooth
+from probnum.filtsmooth.gaussfiltsmooth.gaussfiltsmooth import (
+    GaussFiltSmooth,
+    linear_discrete_update,
+)
 from probnum.prob import RandomVariable, Normal
 from probnum.filtsmooth.statespace import (
     ContinuousModel,
@@ -13,6 +16,8 @@
     DiscreteGaussianLinearModel,
 )
 
+from probnum.filtsmooth.gaussfiltsmooth._utils import is_cont_disc, is_disc_disc
+
 
 class Kalman(GaussFiltSmooth):
     """
@@ -29,9 +34,9 @@ def __new__(cls, dynamod, measmod, initrv, **kwargs):
         discrete-discrete Kalman object is created.
         """
         if cls is Kalman:
-            if _cont_disc(dynamod, measmod):
+            if is_cont_disc(dynamod, measmod):
                 return _ContDiscKalman(dynamod, measmod, initrv, **kwargs)
-            if _disc_disc(dynamod, measmod):
+            if is_disc_disc(dynamod, measmod):
                 return _DiscDiscKalman(dynamod, measmod, initrv)
             else:
                 errmsg = (
@@ -43,20 +48,6 @@ def __new__(cls, dynamod, measmod, initrv, **kwargs):
             return super().__new__(cls)
 
 
-def _cont_disc(dynamod, measmod):
-    """Checks whether the state space model is continuous-discrete."""
-    dyna_is_cont = issubclass(type(dynamod), ContinuousModel)
-    meas_is_disc = issubclass(type(measmod), DiscreteModel)
-    return dyna_is_cont and meas_is_disc
-
-
-def _disc_disc(dynamod, measmod):
-    """Checks whether the state space model is discrete-discrete."""
-    dyna_is_disc = issubclass(type(dynamod), DiscreteModel)
-    meas_is_disc = issubclass(type(measmod), DiscreteModel)
-    return dyna_is_disc and meas_is_disc
-
-
 class _ContDiscKalman(Kalman):
     """
     Provides predict() and update() methods for Kalman filtering and
@@ -69,11 +60,11 @@ def __init__(self, dynamod, measmod, initrv, **kwargs):
         """
         if not issubclass(type(dynamod), LinearSDEModel):
             raise ValueError(
-                "ContinuosDiscreteKalman requires " "a linear dynamic model."
+                "ContinuousDiscreteKalman requires a linear dynamic model."
             )
         if not issubclass(type(measmod), DiscreteGaussianLinearModel):
             raise ValueError(
-                "DiscreteDiscreteKalman requires " "a linear measurement model."
+                "ContinuousDiscreteKalman requires a linear measurement model."
             )
         if "cke_nsteps" in kwargs.keys():
             self.cke_nsteps = kwargs["cke_nsteps"]
@@ -137,8 +128,4 @@ def _discrete_kalman_update(time, randvar, data, measurementmodel, **kwargs):
     measmat = measurementmodel.dynamicsmatrix(time, **kwargs)
     meascov = measurementmodel.diffusionmatrix(time, **kwargs)
     meanest = measmat @ mpred
-    covest = measmat @ cpred @ measmat.T + meascov
-    ccest = cpred @ measmat.T
-    mean = mpred + ccest @ np.linalg.solve(covest, data - meanest)
-    cov = cpred - ccest @ np.linalg.solve(covest.T, ccest.T)
-    return (RandomVariable(distribution=Normal(mean, cov)), covest, ccest, meanest)
+    return linear_discrete_update(meanest, cpred, data, meascov, measmat, mpred)

src/probnum/filtsmooth/gaussfiltsmooth/unscentedkalman.py

@@ -7,7 +7,10 @@
 
 import numpy as np
 
-from probnum.filtsmooth.gaussfiltsmooth.gaussfiltsmooth import GaussFiltSmooth
+from probnum.filtsmooth.gaussfiltsmooth.gaussfiltsmooth import (
+    GaussFiltSmooth,
+    linear_discrete_update,
+)
 from probnum.prob import RandomVariable, Normal
 from probnum.filtsmooth.gaussfiltsmooth.unscentedtransform import UnscentedTransform
 from probnum.filtsmooth.statespace import (
@@ -158,11 +161,7 @@ def _update_discrete_linear(time, randvar, data, measmod, **kwargs):
     measmat = measmod.dynamicsmatrix(time, **kwargs)
     meascov = measmod.diffusionmatrix(time, **kwargs)
     meanest = measmat @ mpred
-    covest = measmat @ cpred @ measmat.T + meascov
-    ccest = cpred @ measmat.T
-    mean = mpred + ccest @ np.linalg.solve(covest, data - meanest)
-    cov = cpred - ccest @ np.linalg.solve(covest.T, ccest.T)
-    return RandomVariable(distribution=Normal(mean, cov)), covest, ccest, meanest
+    return linear_discrete_update(meanest, cpred, data, meascov, measmat, mpred)
 
 
 def _update_discrete_nonlinear(time, randvar, data, measmod, ut, **kwargs):

src/probnum/filtsmooth/statespace/__init__.py

@@ -1,16 +1,3 @@
 from .continuous import *
 from .discrete import *
 from .statespace import *
-
-# Public classes and functions. Order is reflected in documentation.
-__all__ = [
-    "ContinuousModel",
-    "LinearSDEModel",
-    "LTISDEModel",
-    "DiscreteModel",
-    "DiscreteGaussianModel",
-    "DiscreteGaussianLinearModel",
-    "DiscreteGaussianLTIModel",
-    "generate_cd",
-    "generate_dd",
-]