Fast Kalman filters in Python leveraging single-instruction multiple-data vectorization. That is, running n similar Kalman filters on n independent series of observations. Not to be confused with SIMD processor instructions.
import simdkalman
kf = simdkalman.KalmanFilter(
state_transition = np.array([[1,1],[0,1]]),
process_noise = np.diag([0.1, 0.01]),
observation_model = np.array([[1,0]]),
observation_noise = 1.0)
data = numpy.random.normal(size=(200, 1000))
# smooth and explain existing data
smoothed = kf.smooth(data)
# predict new data
pred = kf.predict(data, 15)
See examples/example.py
for a more comprehensive example and
ReadTheDocs for the full documentation.
For the changelog, see releases page
According to examples/benchmark.py
. This can be up to 100x faster than
pykalman and 70x faster than
filterpy when can be vectorized over
many independent timeseries. Also in the non-vectorized case, it can be 2x
faster.
pip install simdkalman
virtualenv venvs/python2
python3 -m venv venvs/python3
source venvs/pythonNNN/bin/activate
pip install -e .[dev,test,docs]
./run-tests.sh
deactivate
virtualenv(personal howto)
Once:
~/.pypirc
as described heresudo pip install twine
virtualenv venvs/test-python2
python3 -m venv venvs/test-python3
Each distribution:
# first, set version in setup.py
# then, download the wheel.zip artifact from Github and extract it to dist/
# or create it manually: python setup.py bdist_wheel
# test PyPI site
twine upload --repository testpypi dist/simdkalman-VERSION*
# the real thing
twine upload dist/simdkalman-VERSION*
# update git tags
git tag VERSION -m "released VERSION"
git push --tags
Test installation from the test site with
source venvs/test-pythonNNN/bin/activate
pip install \
--index-url https://test.pypi.org/simple/ \
--extra-index-url https://pypi.org/simple \
simdkalman --upgrade
# or the real thing with just
# pip install simdkalman
pip install matplotlib
python examples/example.py
deactivate