Skip to content
/ pyhts Public
forked from AngelPone/pyhts

A python package for hierarchical forecasting, inspired by hts package in R.

License

Notifications You must be signed in to change notification settings

kl-lab/pyhts

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

65 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

pyhts

pypi version license

A python package for hierarchical forecasting, inspired by the hts package in R.

Features

  • Support pupular forecast reconciliation models in the literature, e.g. ols, wls, mint et al. Forecasting with temporal hierarchies will be supported in the future.
  • Multiple methods for the construction of hierarchy.
  • Use different base forecasters for different hierarchical levels.
  • Sklearn-like API.

Quick Demo

  • Load the Australia tourism flows data.
from pyhts import load_tourism

tourism_data = load_tourism()
train = tourism_data.iloc[:, 4:-12].T.values
test = tourism_data.iloc[:, -12:].T.values
  • Define the hierarchy.
from pyhts import Hierarchy

hierarchy = Hierarchy.new(tourism_data, [('state', 'region', 'city')])
print(hierarchy.node_name)
  • Create an ols forecasting reconciliation model with sklearn-like API.
from pyhts import HFModel

model_ols = HFModel(hierarchy=hierarchy, base_forecasters="arima",
                    hf_method="comb", comb_method="ols")
  • Fit the model and produce forecasts.
model_ols.fit(train)
ols = model_ols.predict(horizon=12)
  • model.fit() fits the baseforecasters and computes the weighting matrix used to reconcile the base forecasts.

  • model.predict() calculates the base forecasts for all levels and reconciles the base forecasts.

  • Obtain coherent forecasts of all the hierarchical levels.
all_level_ols = hierarchy.aggregate_ts(ols)
  • fit other methods using fitted base forecasters
model_wlss = HFModel(hierarchy, base_forecasters=model_ols.base_forecasters,
                     hf_method="comb", comb_method="wls", weights="structural")
model_wlss.fit(train)
wlss = model_wlss.predict(horizon=12)

model_wlsv = HFModel(hierarchy, base_forecasters=model_ols.base_forecasters,
                     hf_method="comb", comb_method="mint", weights="variance")
model_wlsv.fit(train)
wlsv = model_wlsv.predict(horizon=12)

model_shrink = HFModel(hierarchy, base_forecasters=model_ols.base_forecasters,
                       hf_method="comb", comb_method="mint", weights="shrinkage")
model_shrink.fit(train)
shrink = model_shrink.predict(horizon=12)
  • Evaluate the forecasting accuracy.
# accuracy of reconciled forecasts
accuracy = [hierarchy.accuracy(test, fcast, hist=train, measure=['mase', 'rmse'])
            for fcast in (ols, wlss, wlsv, shrink)]

# accuracy of base forecasts
base_forecasts = model_ols.generate_base_forecast(horizon=12)
accuracy_base = hierarchy.accuracy_base(test, base_forecasts, hist=train, measure=['mase', 'rmse'])

Because of the incoherence of base forecasts, base_forecasts are forecasts of all time series in the hierarchy, while coherent forecasts are forecasts of the bottom-level time series.

Documentation

See documentation here.

About

A python package for hierarchical forecasting, inspired by hts package in R.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%