Configuration, without guesswork

Build a runnable GLNBench experiment.

Select supported components, tune open parameters, and get validated YAML plus the exact local command. Everything happens in this browser—no configuration or personal data is transmitted.

Data and noise

Choose the graph and annotation process.

Dataset Noise model Noise rate Data directory Normalize node features

Model and method

Combine a GNN backbone with a robustness strategy.

Backbone Modified GNN variant used by tunedGNN only Method Hidden channels GNN layers Dropout Normalization Jumping knowledge Add self-loops Residual connections

Method hyperparameters

Release benchmark defaults. Change any value to customize this run.

This method has no additional hyperparameters.

Training

Set compute, stopping, and evaluation behavior.

Device Learning mode Learning rate Weight decay Epochs Patience Independent runs Base seed

Batching, sweeps and method parameters

Batch size blank = full graph Sampler Noise-rate sweep comma-separated values in [0,1] Advanced method overrides Optional JSON object merged over the controls above; useful for experimental parameters not yet in the manifest.