Frequently Asked Questions

What’s the difference between AVSComposer and load_and_preprocess?

pyavs.AVSComposer (see AVSComposer Guide) is the actively developed, recommended high-level interface for MEG + eye-tracking data fusion. The functional API (pyavs.load_and_preprocess() / pyavs.get_epochs()) is an older interface that still works but receives less active development – prefer AVSComposer for new work.

Where can I download the AVS dataset?

The full dataset – all 5 subjects, all sessions – is downloadable now, on demand, via pyavs.open_remote() – see Data Access. AWS Open Data (the kietzmannlab-avs S3 bucket) is the sole hosting track for the dataset.

Do I have to download all 663 GiB?

No. The tree is organized so you can take only what you need, and the epoch metadata tables are small and separate from the epoch data – so you can filter epochs by scene, fixated object or event kinematics before downloading any MEG. See Data Access for a per-goal breakdown.

Is the dataset BIDS?

BIDS-inspired, not BIDS-valid. It uses BIDS-style sub-<label>/ses-<label>/<datatype>/ directories and a derivatives/ tree, but raw files keep their original acquisition filenames and no validator sidecars are shipped. See Dataset Structure.

Can I do source reconstruction with the released data?

Yes – derivatives/freesurfer/ is a directly usable MNE SUBJECTS_DIR with each participant’s source space, single-shell BEM, forward solution, coregistration and parcellations. Scalp/head surfaces are withheld for privacy, so you cannot recompute the coregistration or a multi-shell BEM from scratch; use the shipped ones. See Known Issues.

Are the released epochs locked to scene onset?

No – despite the _scene_ in their filenames, which means “during the scene-viewing task”. fixation_scene and saccade_scene epochs are locked to eye movements. Stimulus-locked epochs can be built from the released SSS runs and scene annotations with pyavs.AVSComposer.

How do I cite this dataset or pyAVS?

See Citation.

My eye-tracking/MEG alignment or fixation counts look off – is this a known issue?

Check Known Issues first – several session-specific data-quality caveats and the automatic corrections pyAVS applies are documented there.

Does pyAVS work without MEG data, e.g. for eye-tracking-only analyses?

Yes – the eye-tracking loading and preprocessing functions (pyavs.load_eye_events(), pyavs.load_and_enrich_eye_events(), pyavs.preprocess_eye_events()) don’t require MEG data. See the “Eye Tracking Only” example in Quick Start Guide.

Something else?

Open an issue on GitHub.