Add NuBench training datasets#888
Merged
Merged
Conversation
Adds a CuratedDataset subclass for the Hexagon Ice LE benchmark from the NuBench suite (~8.6M neutrino events). Uses the pre-computed NuBench train/test selection parquet files rather than random splitting, and supports both pulses_no_noise and merged_photons pulsemaps. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace per-dataset HexagonIceLEDataset class with a single NuBenchDataset entry point driven by a NuBenchSpec registry, so new NuBench datasets (e.g. triangle) can be added without a new class. Pulsemap is selected per split (merged_photons for train/val, pulses_no_noise for test) to match the NuBench convention. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Aske-Rosted
reviewed
May 8, 2026
Collaborator
Aske-Rosted
left a comment
There was a problem hiding this comment.
I have a single comment, otherwise this seems fine to me.
Aske-Rosted
approved these changes
May 8, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #887.
Summary
NuBenchDataset(insrc/graphnet/datasets/nubench_datasets.py), a singleERDAHostedDatasetentry point for the NuBench benchmark suite (arXiv:2511.13111).cluster,flower_l,flower_s,flower_xl,hexagon,hexagon_ice_le— each pinned to its ERDA hash and correspondingNuBenchDetectorsubclass.merged_photonspulsemap, test reads frompulses_no_noise.NuBenchDataset,NuBenchSpec,FEATURES_NUBENCH, andTRUTH_NUBENCHfromgraphnet.datasets.The NuBench detector classes (
Cluster,FlowerL,FlowerS,FlowerXL,Hexagon,Triangle) and geometry tables already exist onmain; this PR adds the missingDatasetlayer so the benchmarks can actually be trained on.Usage
Test plan
NuBenchDataset(name="hexagon_ice_le", ...)end-to-end: download from ERDA, build train/val/test splits, iterate a few batches.available_datasets()lists all entries in the registry.