GPflow: A Gaussian Process Library using TensorFlow

Authors: Alexander G. de G. Matthews, Mark van der Wilk, Tom Nickson, Keisuke Fujii, Alexis Boukouvalas, Pablo León-Villagrá, Zoubin Ghahramani, James Hensman

JMLR 2017 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental As a scenario, we studied training a multiclass GP classifier on MNIST using stochastic variational inference... We compared against GPy... We did a series of trials measuring the time each package took to perform 50 iterations of the algorithm... The results of the timing experiments are shown in Figure 1.
Researcher Affiliation Academia Alexander G. de G. Matthews EMAIL Department of Engineering University of Cambridge Cambridge, UK; Mark van der Wilk EMAIL Department of Engineering University of Cambridge Cambridge, UK; Tom Nickson EMAIL Department of Engineering Science University of Oxford Oxford, UK; Keisuke Fujii EMAIL Department of Mechanical Engineering and Science Graduate School of Engineering, Kyoto University, Japan; Alexis Boukouvalas EMAIL Division of Informatics Manchester University Oxford Road, Manchester, UK; Pablo Le on-Villagr a EMAIL School of Informatics University of Edinburgh 10 Crichton Street, Edinburgh, UK; Zoubin Ghahramani EMAIL Department of Engineering University of Cambridge Cambridge, UK; James Hensman EMAIL CHICAS, Faculty of Health and Medicine Lancaster University Lancaster, UK.
Pseudocode No The paper describes the architecture and features of the GPflow library but does not include any structured pseudocode or algorithm blocks.
Open Source Code Yes GPflow and Tensor Flow are available as open source software under the Apache 2.0 license. [...] All GPflow source code is openly available on Git Hub at http://github.com/GPflow/GPflow. [...] Code for these timing experiments can be found at http://github.com/gpflow/GPflow Benchmarks
Open Datasets Yes As a scenario, we studied training a multiclass GP classifier on MNIST using stochastic variational inference
Dataset Splits No As a scenario, we studied training a multiclass GP classifier on MNIST using stochastic variational inference (Hensman et al., 2015b,a). We compared against GPy... The paper mentions using the MNIST dataset but does not specify the train/test/validation split percentages or methodology used for this dataset.
Hardware Specification Yes We used a Linux workstation Intel Core I7-4930K CPU clocked at 3.40GHz an NVIDIA GM200 Geforce GTX Titan X GPU.
Software Dependencies No GPflow: A Gaussian Process Library using Tensor Flow and Python for its front end. The paper mentions GPflow, TensorFlow, and Python but does not provide specific version numbers for these software components.
Experiment Setup No As a scenario, we studied training a multiclass GP classifier on MNIST using stochastic variational inference... We did a series of trials measuring the time each package took to perform 50 iterations of the algorithm. The paper specifies the algorithm and number of iterations but does not provide specific hyperparameters like learning rate, batch size, or other detailed training configurations.