Repository navigation
v2.1.1
chapkit test and $generate-sample-data keep the full history when the prediction horizon is clamped.
What changed
- Since 2.1.0, the generated prediction horizon is clamped to the service's max_prediction_periods, and the historic frame was clamped with it. With the default --predict-rows 150 and max_prediction_periods=12, models got 12 months of history instead of 30, which broke seasonal models such as auto_arima_chapkit. History now uses the unclamped period count, and the future frame starts right after it. (#115)
- generate_prediction_data takes an optional historic_periods (defaults to num_periods, so existing callers are unchanged). (#115)
Upgrading
No changes required.
What's Changed
- fix(test): keep the full historic length when the prediction horizon is clamped by @mortenoh in #115
- chore: release 2.1.1 by @mortenoh in #116
Full Changelog: v2.1.0...v2.1.1