CoffeeBoost: Gradient Boosting Native Conformal Inference for Bayesian Optimization
Authors: Yuanhao Lai, Pengfei Zheng, Chenpeng Ji, Cheng Qiu, Tingkai Wang, Songhan Zhang, Zhengang Wang, Yunfei Du
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Across eight auto-tuning benchmarks for database management systems (DBMS), we evaluate Coffee Boost and show its superior learnability and optimizability against existing GP-based and tree-ensemble-based BO schemes. Detailed analysis further shows Coffee Boost s predictive distributions excel in both coverage and tightness. |
| Researcher Affiliation | Collaboration | 1Huawei Technologies Co., Ltd. 2The Chinese University of Hong Kong, Shenzhen |
| Pseudocode | Yes | Algorithm 1: Probabilistic surrogate of Coffee Boost Algorithm 2: Pseudo code of Coffee Boost |
| Open Source Code | No | The paper discusses implementations of baselines and how Coffee Boost was implemented using existing tools (lightgbm, SMAC3) but does not provide a statement or link for the open-sourcing of Coffee Boost's specific methodology or code. |
| Open Datasets | Yes | We conduct extensive experiments on eight open benchmarks of database performance auto-tuning (Zhang et al. 2022) with respect to different workloads including TWITTER, TATP, VOTER, SMALLBANK, YCSB, TPC-C, SYSBENCH, and JOB. |
| Dataset Splits | Yes | The evaluation dataset is then randomly split into a training set of 100 samples and a test set of the remaining, used to train a surrogate and test its predictive performance. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments. |
| Software Dependencies | No | To implement the above methods, we use the authors released implementations except for NGB-BO, PGBM-BO, VEGB-BO and Coffee Boost, where we use the authors implementation of NGBoost, PGBM and virtual ensemble together with lightgbm (Ke et al. 2017) and SMAC3 (Lindauer et al. 2022) to implement the surrogate with an EI acquisition. |
| Experiment Setup | Yes | We set the number of boosting iterations to 100, the learning rate to 0.05, the maximum tree depth to 7, the number of SGBE s base learners to 20, the conformal-related hyperparameters β = 0.01 and the set T of 20 equally spaced iterations among the entire training history for GBDT-native difficulty estimator from Equation (4). |