README.md

metadata

license: other
pretty_name: AxoBench
task_categories:
  - time-series-forecasting
  - tabular-regression
tags:
  - neuroscience
  - computational-neuroscience
  - neuron-simulation
  - single-cell
  - benchmark
  - dendrites
  - coreneuron

AxoBench

This public working release contains AxoBench, the DendroBench v1 neuronio-like dataset generated with the CoreNEURON GPU path. It is intended for internal benchmark validation, diagnostic development, and paper preparation before a public release decision.

Layout

• train/: 785 compressed NPZ shards, 100,000 traces.
• val/: 100 compressed NPZ shards, 12,500 traces.
• test/: 100 compressed NPZ shards, 12,500 traces.
• private-test/: 100 compressed NPZ shards, 12,500 traces.
• interventions/<name>/: paired baseline/intervention NPZ shards for event_dropout, exc_dropout, inh_dropout, site_silence, and temporal_jitter; each intervention contains 2,500 paired traces.

The ordinary trace shards use .npz files with arrays such as inputs, targets, and sample_ids. Intervention shards use paired arrays such as baseline_inputs, intervention_inputs, baseline_targets, and intervention_targets.

Compression

The payload is stored as shard-wise deflated NPZ files, matching the practical NeuronIO-style packaging while preserving partial downloads and resumable uploads. Do not wrap this folder in one monolithic zip or tar archive for the primary Hugging Face distribution: the shards are already compressed and an outer archive would make subset access worse.

See compression_report.json for measured compression ratios and checksums.sha256 for file integrity checks.

Metadata

• dataset_manifest_template.json: intended dataset structure and benchmark framing.
• generation_summary.json: generation backend, morphology set, sample counts, and runtime summary.
• generation_completion_check.json: split/intervention completion summary.
• dataset_package_manifest.json: upload/package-level summary.
• compression_report.json: compression audit.
• checksums.sha256: SHA-256 checksums for all packaged files except the checksum file itself.

Access State

This upload is public so collaborators and reviewers can access the packaged shards directly. The dataset card, license, private-test policy, and paper-facing framing should still be reviewed before a paper-facing release tag or DOI mirror.