KONG: Kernels for ordered-neighborhood graphs
Authors: Moez Draief, Konstantin Kutzkov, Kevin Scaman, Milan Vojnovic
NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section we present our evaluation of the classification accuracy and computation speed of our algorithm and comparison with other kernel-based algorithms using a set of real-world graph datasets. |
| Researcher Affiliation | Collaboration | Moez Draief1 Konstantin Kutzkov2 Kevin Scaman1 Milan Vojnovic2 1 Huawei Noah s Ark Lab 2 London School of Economics, London moez.draief@huawei.com, kutzkov@gmail.com (Corresponding author), kevin.scaman@huawei.com, m.vojnovic@lse.ac.uk |
| Pseudocode | Yes | Algorithm 1: EXPLICITGRAPHFEATUREMAP. Input: Graph G = (V, E, ℓ, τ), depth h, labeling ℓ: V L, base kernel κ for v V do Traverse the subgraph Tv rooted at v up to depth h Collect the node labels ℓ(u) : u Tv in the order specified by τv into a string Sv Sketch the explicit feature map φκ(Sv) for the base string kernel κ (without storing Sv) Φκ(G) P v V φκ(Sv) return Φκ(G) |
| Open Source Code | Yes | Software implementation and data are available at https://github.com/kutzkov/KONG. |
| Open Datasets | Yes | We evaluated the algorithms on widely-used benchmark datasets from various domains [Kersting et al., 2016]. MUTAG [Debnath et al., 1991], ENZYMES [Schomburg et al., 2004], PTC [Helma et al., 2001], Proteins [Borgwardt et al., 2005] and NCI1 [Wale and Karypis, 2006] represent molecular structures, and MSRC [Neumann et al., 2016] represents semantic image processing graphs. |
| Dataset Splits | Yes | We performed 10-fold cross-validation using 9 folds for training and 1 fold for testing. |
| Hardware Specification | Yes | All algorithms were implemented in Python 3 and experiments performed on a Windows 10 laptop with an Intel i7 2.9 GHz CPU and 16 GB main memory. |
| Software Dependencies | No | The paper mentions 'Python 3', 'scikit-learn implementation', and 'LIBLINEAR algorithm' but only provides a specific version for Python (Python 3) and not for the key libraries used. It does not list multiple key software components with their specific version numbers. |
| Experiment Setup | Yes | Similar to previous works [Niepert et al., 2016, Yanardag and Vishwanathan, 2015], we choose the optimal number of hops h = 2 for the WL kernel and k {5, 6} for the k-walk kernel. We performed 10-fold cross-validation using 9 folds for training and 1 fold for testing. The optimal regularization parameter C for each dataset was selected from {0.1, 1, 10, 100}. |