2 releases
0.1.0 | Sep 25, 2023 |
---|
#14 in #forecasting
20KB
314 lines
Python bindings to the augurs time series framework
Installation
Eventually wheels will be provided as part of GitHub releases and maybe even on PyPI. At that point it will be as easy as:
$ pip install augurs
Until then it's a bit more manual. You'll need maturin installed and a local copy of this
repository. Then, from the crates/pyaugurs
directory, with your virtualenv activated:
$ maturin build --release
You'll probably want numpy as well:
$ pip install numpy
Usage
Multiple Seasonal Trend Decomposition with LOESS (MSTL) models
import augurs as aug
import numpy as np
y = np.array([1.5, 3.0, 2.5, 4.2, 2.7, 1.9, 1.0, 1.2, 0.8])
periods = [3, 4]
# Use an AutoETS trend forecaster
model = aug.MSTL.ets(periods)
model.fit(y)
out_of_sample = model.predict(10, level=0.95)
print(out_of_sample.point())
print(out_of_sample.lower())
in_sample = model.predict_in_sample(level=0.95)
# Or use your own forecaster
class CustomForecaster:
"""See docs for more details on how to implement this."""
def fit(self, y: np.ndarray):
pass
def predict(self, horizon: int, level: float | None) -> aug.Forecast:
return aug.Forecast(point=np.array([5.0, 6.0, 7.0]))
def predict_in_sample(self, level: float | None) -> aug.Forecast:
return aug.Forecast(point=y)
...
model = aug.MSTL.custom_trend(periods, aug.TrendModel(CustomForecaster()))
model.fit(y)
model.predict(10, level=0.95)
model.predict_in_sample(level=0.95)
Exponential smoothing models
import augurs as aug
import numpy as np
y = np.array([1.5, 3.0, 2.5, 4.2, 2.7, 1.9, 1.0, 1.2, 0.8])
model = aug.AutoETS(3, "ZZN")
model.fit(y)
model.predict(10, level=0.95)
More to come!
Dependencies
~10–16MB
~224K SLoC