Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
COMRECGC: Global Graph Counterfactual Explainer through Common Recourse
Authors: Gregoire Fournier, Sourav Medya
ICML 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We benchmark our algorithm against strong baselines on four different real-world graphs datasets and demonstrate the superior performance of COMRECGC against the competitors. We also compare the common recourse explanations to the graph counterfactual explanation, showing that common recourse explanations are either comparable or superior, making them worth considering for applications such as drug discovery or computational biology. |
| Researcher Affiliation | Academia | 1Department of Mathematics, Statistics and Computer Science, University of Illinois Chicago, Chicago, USA 2Department of Computer Science, University of Illinois Chicago, Chicago, USA. Correspondence to: Gr egoire Fournier <EMAIL>. |
| Pseudocode | Yes | Algorithm 1 CR CLUSTERING(G, S, R) Algorithm 2 COMRECGC(ϕ, G, k, M, τ, R, n) Algorithm 3 COMRECGC for FC (ϕ, G, k, M, R) Algorithm 5 MULTI-HEAD VRRW(ϕ, G, k, M, τ) |
| Open Source Code | Yes | Reproducibility. We make our code available at https: //github.com/ssggreg/COMRECGC. |
| Open Datasets | Yes | We consider the datasets MUTAGENICITY (Riesen & Bunke, 2008; Kazius et al., 2005), NCI1 (Wale & Karypis, 2006), AIDS (Riesen & Bunke, 2008), and PROTEINS (Borgwardt et al., 2005; Dobson & Doig, 2003). |
| Dataset Splits | Yes | The training/validation/testing split is 80%/10%/10% |
| Hardware Specification | No | The paper mentions 'National Artificial Intelligence Research Resource (NAIRR) Pilot and the Texas Advanced Computing Center (TACC) Vista' in the Acknowledgment, but does not provide specific details on the CPU, GPU, or memory used for the experiments. |
| Software Dependencies | No | The paper mentions using a GCN model and the Adam optimizer, along with citations, but does not specify version numbers for any software libraries, frameworks, or programming languages used. |
| Experiment Setup | Yes | We train a base GNN model (GCN) (Kipf & Welling, 2017) for a binary classification task, consisting of three convolutional layers, a max pooling layer, and a fully connected layer... The model is trained with the Adam optimizer (Kingma & Ba, 2014) and a learning rate of 0.001 for 1000 epochs. Across all of our experiments, COMRECGC uses k = 5 heads, has probability of teleportation τ = 0.05, performs the random walk for M = 50000 steps, and selects R = 100 common recourse. |