Use AllenSDK's VisualBehaviorNeuropixelsProjectCache for normal access; the DANDI record is an archival cloud release and the Allen portal provides task and whitepaper documentation. The behavtaskatlas MVP adapter reads a local VBN NWB /intervals/trials table directly via h5py and treats it as the active visual change-detection behavior slice; neural and video side tables remain deferred.
Dataset
Allen Visual Behavior Neuropixels
Open NWB-format electrophysiology and behavioral dataset from mice performing the Allen visual change-detection task, with active behavior, passive replay, receptive-field mapping, and optotagging epochs.
Access notes
Data formats
- Neurodata Without Borders (NWB)
- AllenSDK project cache
- DANDI archive assets
Expected trial-table mapping
Canonical fields the harmonizer expects this dataset to populate.
| Canonical field | Source mapping |
|---|---|
| choice | Lick in response window mapped to go; withheld lick mapped to withhold. |
| correct | Trial outcome mapped from hit, miss, false alarm, or correct rejection. |
| evidence_strength | Not a scalar; derive d-prime or hit-rate summaries by image pair. |
| feedback | Reward on hits; no reward on misses, false alarms, and correct rejections unless source marks otherwise. |
| response_time | First lick latency relative to change time where available. |
| session_id | Allen ecephys or behavior session identifier. |
| stimulus_side | none; visual images are centrally presented rather than lateralized. |
| stimulus_value | Categorical image-change indicator or image-pair code. |
| subject_id | Mouse identifier from session metadata. |
| task_variables | image set, pre-change image, change image, omitted flashes, active/passive epoch, novelty exposure, optotagging context, lick times, and reward times. |
| trial_index | Trial index from the active behavior trials table. |
Caveats
- The archive is large; adapters should stream or cache per-session NWB assets rather than assume full local download.
- The current MVP adapter covers active behavior trials only and does not yet join spike, LFP, probe, channel, optotagging, passive-replay, receptive-field, or video tables.