Code-Aware Cross-Program Transfer Hyperparameter Optimization

Authors: Zijia Wang, Xiangyu He, Kehan Chen, Chen Lin, Jinsong Su

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on tuning various recommendation models and Spark applications have demonstrated that Cat HPO can steadily obtain better and more robust hyperparameter performances within fewer samples than state-of-the-art competitors.
Researcher Affiliation Academia School of Informatics, Xiamen University Xiamen, Fujian, China chenlin@xmu.edu.cn
Pseudocode No The paper describes its methods using prose and mathematical equations but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes The datasets and codes are available online 4. 4https://github.com/XMUDM/CaTHPO
Open Datasets Yes The datasets and codes are available online 4. 4https://github.com/XMUDM/CaTHPO
Dataset Splits No The paper mentions using 'four small datasets are used in source tasks and one large dataset is used in the target task for testing' and 'Each run contains 16 samples, i.e., N = 16', but does not provide specific train/validation/test splits (e.g., percentages or counts within a single dataset for model training).
Hardware Specification No The paper mentions 'a cluster of eight computing nodes' for Spark tuning and 'a GPU computing server' for RS tuning, but it does not provide specific hardware details such as GPU models, CPU types, or memory specifications.
Software Dependencies No The paper mentions using specific software components like 'Transformer encoder', 'LSTM layer', 'MLPs', 'Gaussian Process', and 'neural acquisition function network', but does not provide specific version numbers for any of these or other software dependencies.
Experiment Setup Yes In Spark tuning, we tune 15 configuration knobs for eight Spark applications on a cluster of eight computing nodes. ... In RS tuning, we tune three hyperparameters for eight recommendation models on a GPU computing server. ... Each run contains 16 samples, i.e., N = 16. ... The AF network is updated via proximal policy optimization with reward defined as the maximal hyperparameter performance up to the current sample.