Multiparameter Persistence Image for Topological Machine Learning

Authors: Mathieu Carrière, Andrew Blumberg

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Moreover, we demonstrate its efficacy by comparing its performance to other multiparameter descriptors on several classification tasks.
Researcher Affiliation Academia Mathieu Carrière Data Shape Inria Sophia-Antipolis Biot, France mathieu.carriere@inria.fr Andrew J. Blumberg Department of Mathematics University of Texas at Austin Austin, TX 78712 blumberg@math.utexas.edu
Pseudocode No The paper describes algorithms but does not provide a formal pseudocode block or algorithm listing.
Open Source Code Yes We also provide open-source implementation for our descriptor and for the other approaches [CFK+19, Vip20], in a public Python package [Car20].
Open Datasets Yes We use time series data sets from the UCR archive [DBK+18] with moderate sizes and lengths; this ensures that the kernel matrices obtained with Multiparameter Persistence Kernel have reasonable sizes and that the point clouds obtained with the Takens embedding in R3 have a reasonable number of points.
Dataset Splits Yes Moreover, we use the train/test split that is suggested for each data set. Resolutions for Multiparameter Persistence Landscapes and Images are 5-fold cross-validated over the set of values {10, 50}, and the powers and bandwidths of the Multiparameter Persistence Images and Kernels are also 5-fold cross-validated with values in {0, 1} (power) and 10{ 2, 1,0,1,2} (bandwidth).
Hardware Specification Yes All experiments have been run on an AWS machine with a Xeon Platinum 8175 processor.
Software Dependencies No The paper mentions a "public Python package" but does not list specific software dependencies with version numbers (e.g., Python version, library versions).
Experiment Setup Yes Resolutions for Multiparameter Persistence Landscapes and Images are 5-fold cross-validated over the set of values {10, 50}, and the powers and bandwidths of the Multiparameter Persistence Images and Kernels are also 5-fold cross-validated with values in {0, 1} (power) and 10{ 2, 1,0,1,2} (bandwidth).