Multi-task hyperparameter tuning using bayesian optimization

[Huma-Shehwana]

Hello, I am attempting to conduct multi-task hyperparameter tuning using Bayesian optimization for the XGBoost algorithm. I believe my scenario qualifies as a multi-task problem because my dataset comprises four variables, and I aim to treat each variable as a target variable while considering the other three variables as features. Consequently, I plan to train the XGBoost model four times, each time using a different target variable. My objective is to discover a set of hyperparameters that optimizes performance across all four tasks simultaneously. I understand that this approach differs from multi-objective hyperparameter tuning, where various performance metrics are evaluated. I am curious whether multi-task hyperparameter tuning is feasible using the mlrMBO package. If so, could you please direct me to any related tutorials or resources for guidance