quantlab.model.predefined.xgb_td#
XGBoost regression head with pytabkit’s tuned defaults.
XGBTDRegressor is a TabkitRegressor that wraps
pytabkit.XGB_TD_Regressor, XGBoost with the tuned defaults from
Holzmüller et al., “Better by Default” (NeurIPS 2024). It is the pytabkit
counterpart of XGBoostRegressor (xgb.py), which drives xgboost
directly. This head takes pytabkit’s parameter set and preprocessing. The
other one takes xgboost’s raw parameters, stops early on a concordance
correlation (CCC) loss and logs feature-importance charts. Both predict
future returns as [num_times, num_symbols, num_labels] from the
flattened factor panel, an xarray.Dataset indexed by timestamp and
symbol.
pytabkit’s XGBoost estimator is single-output, so one estimator is fitted per label.
Classes
Predict future returns with pytabkit's tuned-default XGBoost. |