scikit-learn Integration

KalmanEstimator wraps any kalbee filter behind the fit/transform/ predict convention, so it drops straight into an sklearn.pipeline.Pipeline or a GridSearchCV sweep. Requires scikit-learn: pip install kalbee[sklearn].

One-liner smoothing

import numpy as np
from kalbee.modules.integration.sklearn_api import KalmanEstimator

t = np.linspace(0, 10, 200)
noisy = np.sin(t) + np.random.normal(0, 0.2, 200)

smoothed = KalmanEstimator(dt=t[1] - t[0], process_var=5.0, measurement_var=0.2).fit_transform(noisy)

Each column of X is one measured spatial axis (pass a 2-D array like [[x, y], ...] for 2-D position tracking); each row is one time step.

Inside a Pipeline

from sklearn.pipeline import Pipeline
from kalbee.modules.integration.sklearn_api import KalmanEstimator

pipe = Pipeline([
    ("kalman", KalmanEstimator(order=2, process_var=1.0, measurement_var=0.5)),
    # ... downstream estimator, e.g. a classifier on the smoothed trajectory
])
smoothed = pipe.fit_transform(raw_measurements)

Parameters

Parameter	Meaning
mode	Any AutoFilter mode ("kf", "akf", "srkf", ...).
order	Kinematic order — 1 = constant-velocity, 2 = constant-acceleration.
dt	Time step between rows of X.
process_var / measurement_var	Noise variances for the default motion/measurement model.
tune	If True, auto-tune Q/R from X via quick_tune instead.
return_full_state	If True, transform returns the full state vector (e.g. position and velocity) instead of just position.

Each transform() call re-runs the filter from its initial state, so repeated calls (including sklearn's own fit_transform/predict calls) are independent — no carry-over state between calls.