A Structural Smoothing Framework For Robust Graph Comparison

Authors: Pinar Yanardag, S.V.N. Vishwanathan

NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental evaluation shows that not only our kernels achieve statistically significant improvements over the unsmoothed variants, but also outperform several other graph kernels in the literature. We report results on classification accuracy on several benchmark datasets as well as their noisy-variants.
Researcher Affiliation Academia Pinar Yanardag Department of Computer Science Purdue University West Lafayette, IN, 47906, USA ypinar@purdue.edu S.V.N. Vishwanathan Department of Computer Science University of California Santa Cruz, CA, 95064, USA vishy@ucsc.edu
Pseudocode Yes Algorithm 1 Insert a Customer Input: dk+1, θk+1, Pk Algorithm 2 Delete a Customer Input: d, θ, P0, C, L, t
Open Source Code Yes Implementations of original and smoothed versions of the kernels, datasets and detailed discussion of parameter selection procedure with the list of parameters used in our experiments can be accessed from http: //web.ics.purdue.edu/ ypinar/nips.
Open Datasets Yes Datasets We used the following benchmark datasets used in graph kernels: MUTAG, PTC, ENZYMES, PROTEINS, NCI1 and NCI109. MUTAG is a dataset of 188 mutagenic aromatic and heteroaromatic nitro compounds [5] with 7 discrete labels. PTC [26] is a dataset of 344 chemical compounds has 19 discrete labels. ENZYMES is a dataset of 600 protein tertiary structures obtained from [2], and has 3 discrete labels. PROTEINS is a dataset of 1113 graphs obtained from [2] having 3 discrete labels. NCI1 and NCI109 [28] are two balanced datasets of chemical compounds having size 4110 and 4127 with 37 and 38 labels, respectively.
Dataset Splits Yes Moreover, we use 10-fold cross validation with a binary C-Support Vector Machine (SVM) where the C value for each fold is independently tuned using training data from that fold.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments, such as CPU/GPU models or memory specifications.
Software Dependencies No The paper mentions software like Matlab, Python, C++, and Nauty, but it does not specify version numbers for any of these or for specific libraries or dependencies.
Experiment Setup Yes All kernels are normalized to have a unit length in the feature space. Moreover, we use 10-fold cross validation with a binary C-Support Vector Machine (SVM) where the C value for each fold is independently tuned using training data from that fold. In order to exclude random effects of the fold assignments, this experiment is repeated 10 times and average prediction accuracy of 10 experiments with their standard deviations are reported4.