Matching Node Embeddings for Graph Similarity

Authors: Giannis Nikolentzos, Polykarpos Meladianos, Michalis Vazirgiannis

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the proposed methods on several benchmark datasets for graph classification and compare their performance to state-of-the-art graph kernels. In most cases, the proposed algorithms outperform the competing methods, while their time complexity remains very attractive. ... Section 5 evaluates the proposed algorithms and compares them with existing methods.
Researcher Affiliation Academia Giannis Nikolentzos Ecole Polytechnique and AUEB nikolentzos@aueb.gr Polykarpos Meladianos Ecole Polytechnique and AUEB pmeladianos@aueb.gr Michalis Vazirgiannis Ecole Polytechnique and AUEB mvazirg@lix.polytechnique.fr
Pseudocode No The paper describes the algorithms in detail but does not include any formal pseudocode blocks or algorithms.
Open Source Code Yes Our code is available at http://www.db-net.aueb.gr/ nikolentzos/code/matchingnodes.zip
Open Datasets Yes We evaluated the performance of our methods on the following 6 bioinformatics datasets: MUTAG, ENZYMES, NCI1, NCI109, PTC-MR and D&D.
Dataset Splits Yes We employed a C-Support Vector Machine (SVM) classifier and in particular, the LIBSVM (Chang and Lin 2011) implementation and performed 10-fold cross-validation.
Hardware Specification Yes CPU runtimes for computing each kernel/similarity matrix as measured in Matlab R2013a on a 3.4GHz Intel Core i7 with 16Gb of RAM.
Software Dependencies Yes measured in Matlab R2013a ... we used their publicly available implementations. ... we employed a C-Support Vector Machine (SVM) classifier and in particular, the LIBSVM (Chang and Lin 2011) implementation
Experiment Setup Yes The parameter C of the SVM was optimized on the training set only. ... As regards the parameter h of the Weisfeiler-Lehman kernels, it was chosen by cross-validation on the training set for h {0, 1, . . . , 10} in the case of the subtree kernel, for h {0, 1, 2} in the case of the shortest path kernel, and for h {0, 1, . . . , 5} in the case of the proposed pyramid match kernel and the optimal assignment method.