Mixed-Variable Bayesian Optimization
Authors: Erik Daxberger, Anastasia Makarova, Matteo Turchetta, Andreas Krause
IJCAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We present experimental results on tuning the hyperparameters of two machine learning algorithms, namely gradient boosting and a deep generative model, on multiple datasets. ... Finally, we show that MIVABO is significantly more sample efficient than state-of-the-art mixed-variable BO algorithms on several hyperparameter tuning tasks, including the tuning of deep generative models. |
| Researcher Affiliation | Academia | 1Department of Engineering, University of Cambridge 2Max Planck Institute for Intelligent Systems, T ubingen 3Department of Computer Science, ETH Zurich |
| Pseudocode | No | The paper describes the algorithm steps in text but does not include any formal pseudocode or algorithm blocks. |
| Open Source Code | Yes | We provide a Python implementation of MIVABO at https: //github.com/edaxberger/mixed_variable_bo. |
| Open Datasets | Yes | We consider extreme gradient boosting (XGBoost) [Chen and Guestrin, 2016], one of the most popular Open ML benchmarks, and tune its ten hyperparameters... We use two datasets, each containing more than 45000 hyperparameter settings. ... We consider tuning the hyperparameters of a variational autoencoder (VAE) [Kingma and Welling, 2014] ... The VAEs are evaluated on stochastically binarized MNIST, as in [Burda et al., 2016], and Fashion MNIST. |
| Dataset Splits | No | The paper mentions a "held-out test set" for evaluation but does not specify a separate validation dataset split. |
| Hardware Specification | No | The paper does not explicitly provide details about the specific hardware (e.g., GPU, CPU models, memory) used for running its experiments. |
| Software Dependencies | No | In our experiments, we use Gurobi as the discrete and L-BFGS as the continuous solver within the alternating optimization scheme, which we always run until convergence. No version numbers are provided for these solvers or any other software. |
| Experiment Setup | Yes | Experimental Setup. For MIVABO2, we set the prior variance α, observation noise variance β, and kernel bandwidth σ to 1.0, and scale the variance as stated in Prop. 1. ... They are trained on 60000 images for 32 epochs, using Adam with a mini-batch size of 128. |