etna
etna copied to clipboard
`CrostonModel`
🚀 Feature Request
Add model based on Croston's method for intermittent demand.
We could also try to implement some variations of this method.
Proposal
Add CrostonModel
.
- It could be implemeted as per-segment model using
BaseAdapter
andPerSegmentModelMixin
(look atProphetModel
for reference) - It could be implemented as
NonPredictionIntervalContextIgnorantAbstractModel
that saves the answer during thefit
- During fit it should raise a warning that it doesn't work with exogenous features, look at
SeasonalMovingAverageModel
References:
- 12.2 Time series of counts
- Stochastic models underlying Croston’s method for intermittent demand forecasting
- Forecasting and Stock Control for Intermittent Demands.
To discuss:
- Should it be our own implementation or some external package can be used (statsforecast for example)
- We should try to make our own implementation
Test cases
- Tests for the model itself
- Inference tests in
tests/test_models/test_inference/
.
Additional context
No response