PlanE: Representation Learning over Planar Graphs

Authors: Radoslav Dimitrov, Zeyang Zhao, Ralph Abboud, Ismail Ceylan

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct an empirical analysis for BASEPLANE evaluating its expressive power on three tasks (Section 7.1), and validating its performance on real-world graph classification (Section 7.2) and regression benchmarks (Section 7.3 and 7.4).
Researcher Affiliation Academia Radoslav Dimitrov Department of Computer Science University of Oxford contact@radoslav11.com Zeyang Zhao Department of Computer Science University of Oxford zeyzang.zhao@cs.ox.ac.uk Ralph Abboud Department of Computer Science University of Oxford ralph@ralphabb.ai Ismail Ilkan Ceylan Department of Computer Science University of Oxford ismail.ceylan@cs.ox.ac.uk This work is largely conducted while these authors were still affiliated with the University of Oxford.
Pseudocode No The paper describes algorithms using mathematical equations and descriptive text, but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code Yes The code for our experiments, as well as instructions to reproduce our results and set up dependencies, can be found at this Git Hub repository: https://github.com/ZZYSonny/Plan E
Open Datasets Yes We evaluate BASEPLANE on the planar EXP benchmark [1]...Mol HIV graph classification task from OGB [33]...We evaluate E-BASEPLANE on all 13 QM9 properties...Graph regression on ZINC. (Citations for EXP, OGB, QM9, ZINC indicate public availability). Data. We select a subset QM9CC of graphs from QM9 to obtain a diverse distribution of CCs.
Dataset Splits Yes We apply the earlier filtering on the original QM9 splits to obtain train/validation/test sets that are direct subsets of the full QM9 splits, and which consist of 44226, 3941 and 3921 graphs, respectively.
Hardware Specification Yes We run all experiments on 4 cores from Intel Xeon Platinum 8268 CPU @ 2.90GHz with 32GB RAM.
Software Dependencies No The paper mentions using the Adam optimizer [39] but does not provide specific software dependencies with version numbers (e.g., Python, PyTorch, CUDA versions).
Experiment Setup Yes All experimental details, including hyperparameters, can be found in Appendix D.4. For example, for Graph classification on EXP: we use a 2-layer BASEPLANE model with 64-dimensional node embeddings. We instantiate the triconnected component encoder with 16-dimensional positional encodings, each computed using a periodicity of 64. ...train BASEPLANE on each fold for 50 epochs using the Adam [39] optimizer with a learning rate of 10^-3, and binary cross entropy loss.