PED-ANOVA: Efficiently Quantifying Hyperparameter Importance in Arbitrary Subspaces

Authors: Shuhei Watanabe, Archit Bansal, Frank Hutter

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

Reproducibility Variable Result LLM Response
Research Type Experimental In a series of experiments, we first verify that our algorithm correctly provides global and local HPI in a toy function. Then we demonstrate that our algorithm takes only less than a second for 105 data points while the prior f-ANOVA [Hutter et al., 2014] would take more than a week. To provide a solid picture of how to use our method, we perform analysis on JAHS-Bench-201 [Bansal et al., 2022]
Researcher Affiliation Academia Shuhei Watanabe , Archit Bansal and Frank Hutter Department of Computer Science, University of Freiburg, Germany {watanabs,bansala,fh}@cs.uni-freiburg.de
Pseudocode Yes Algorithm 1 Local PED-ANOVA D = {(xn, f(xn))}N n=1 (Dataset to analyze), γ, γ (User-defined quantiles of top domains) 1: See Appendices E.2, E.3 for practical usages 2: Sort D in ascending order by f 3: |Dγ| 2 and |Dγ | 2 must hold 4: Pick the top-γ and -γ quantile observations Dγ, Dγ 5: for d = 1, . . . , D do 6: Count occurrences of unique values c(n) d 7: Build KDEs pd( |Dγ), pd( |Dγ ) by Eq. (15) 8: Compute vγ d by Eq. (16) 9: return {vγ d}D d=1
Open Source Code Yes To facilitate reproducibility, our implementation is available at https://github.com/nabenabe0928/local-anova/.
Open Datasets Yes To provide a solid picture of how to use our method, we perform analysis on JAHS-Bench-201 [Bansal et al., 2022]
Dataset Splits No The paper uses the JAHS-Bench-201 as a dataset for analysis, querying it for 'validation accuracy', but does not describe train/validation/test splits for their own method's development or evaluation.
Hardware Specification Yes Furthermore, all experiments were run on the hardware with Intel Core i7-10700 and we used the f-ANOVA implementation with the default parameter setting by Optuna
Software Dependencies No The paper mentions using 'Optuna' and 'XGBoost models' but does not provide specific version numbers for these or any other software dependencies.
Experiment Setup Yes In this experiment, we discretized the HPs with nd = 1001 and all samples were drawn from the uniform distribution. We constructed the dataset D in Algorithm 1 by querying JAHS-Bench-201 for the validation accuracy, i.e. f(x), of N lattice points, where N = 41,343,750 and f(x1, x2, x3, x4) = sum(d=1 to 4) wd(xd) xd^2 ... and W := {Wd}3 d=0 = {50, 5^-1, 5^-2, 5^-3}.