Approximate Vanishing Ideal Computations at Scale

Authors: Elias Samuel Wirth, Hiroshi Kera, Sebastian Pokutta

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

Reproducibility Variable Result LLM Response
Research Type Experimental We perform numerical experiments on data sets of up to two million samples, highlighting that OAVI is an excellent large-scale feature transformation method. 5 NUMERICAL EXPERIMENTS Unless noted otherwise, the setup for the numerical experiments applies to all experiments in the paper.
Researcher Affiliation Academia Elias Wirth Institute of Mathematics Berlin Institute of Technology Berlin, Germany wirth@math.tu-berlin.de Hiroshi Kera Graduate School of Engineering Chiba University Chiba, Japan kera.hiroshi@gmail.com Sebastian Pokutta Institute of Mathematics & AI in Society, Science, and Technology Berlin Institute of Technology & Zuse Institute Berlin Berlin, Germany pokutta@zib.de
Pseudocode Yes Algorithm 1: Oracle approximate vanishing ideal algorithm (OAVI)
Open Source Code Yes Our code is publicly available on Git Hub.
Open Datasets Yes Table 1: Overview of data sets. All data sets are binary classification data sets and are retrieved from the UCI Machine Learning Repository (Dua & Graff, 2017) and additional references are provided.
Dataset Splits Yes We tune the hyperparameters on the training data using threefold cross-validation.
Hardware Specification Yes Experiments are implemented in PYTHON and performed on an Nvidia Ge Force RTX 3080 GPU with 10GB RAM and an Intel Core i7 11700K 8x CPU at 3.60GHz with 64 GB RAM.
Software Dependencies No The paper mentions 'PYTHON' and the 'SCIKIT-LEARN software package' but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes For the CG variants, we set τ = 1, 000. The CG variants are run up to accuracy ϵ = 0.01 ψ and terminated early when less than 0.0001 ψ progress is made in the difference between function values, when the coefficient vector of a generator is constructed, or if we have a guarantee that no coefficient vector of a generator can be constructed. Table 3: Hyperparameter ranges for numerical experiments.