anybody was able to use reg:tweedie with xgboost? ...
# mlforecast
j
anybody was able to use reg:tweedie with xgboost? for me it always throws and error. i am applying optuna + xgboost and i assume some forecasts became nan or so, which breaks creating lag_transform:
UserWarning: Found null values in expanding_std_lag1, rolling_std_lag1_window_size7_min_samples1, rolling_std_lag1_window_size70_min_samples1, rolling_std_lag1_window_size105_min_samples1, seasonal_rolling_std_lag1_season_length7_window_size3_min_samples1.
warnings.warn(f'Found null values in {", ".join(cols_with_nulls)}.')
so it turns out using 'gblinear' as a booster breaks the tweedie objectve. i didnt know before, but just realised when some tuning runs finshed and others not. removing gblinear also allows me to use any lag transforms that i want to inside mlforecast!