caretForecast 0.1.3
Bug Fixes
- Fixed
get_var_imp() to accept both ARml
and forecastARml objects as documented
- Fixed dimension drop in
lag_maker() when
max_lag = 1
- Fixed dimension drop in
forecast_loop() when
forecasting single step
- Fixed
xreg being lost during calibration - now properly
preserved before truncation
- Fixed redundant BoxCox transformation when
lambda = "auto"
- Fixed invalid
trend_method parameter in
get_var_imp() example
- Fixed syntax error in test file
Tests
- Added comprehensive test suite with 246 tests covering:
ARml(): parameter validation, edge cases
(max_lag = 1, cv = FALSE), constant data,
non-seasonal data, different frequencies (quarterly, weekly), xreg
handling, BoxCox transformations
forecast.ARml(): error handling, multiple confidence
levels, calibrated forecasts, output structure
split_ts(): input validation, time series attribute
preservation
- Conformal prediction:
conformalRegressor(),
conformalRegressorByHorizon(), and their predict methods
with bounds and error handling
- Calibration: horizon-specific calibration, trumpet-shaped intervals,
xreg with calibration
- Internal utilities:
lag_maker(), %notin%,
pred_func(), forecast_loop()
caretForecast 0.1.2
- Implemented horizon-specific conformal prediction intervals with
proper out-of-sample calibration
- Added new parameters to
ARml(): calibrate,
calibration_horizon, n_cal_windows
- New functions:
conformalRegressorByHorizon(),
calibrate_horizon_scores()
- Prediction intervals now properly widen with forecast horizon
(trumpet shape)
- Added comprehensive test suite for short time series handling
caretForecast 0.1.1
caretForecast 0.0.3
caretForecast 0.0.2
- Added a
NEWS.md file to track changes to the
package.