Choosing an EyeLink acquisition interface¶
EyeLink acquisition control, EDF reading, ASC parsing, event detection and functional inference are different problems. Do not install a hardware-control library as a dependency of scientific trajectory analysis.
| External tool | Verified scope | Appropriate use | eyetrajectoriespy status |
|---|---|---|---|
| SyeLink 2.0 | ASC parsing, gaze/calibration/validation/session records, messages, CSV/JSON exports | Offline ASC parser and metadata-preserving sample handoff | Pinned 2.0.0 synthetic-object example and regression CI; real ASC qualification pending |
| PyeLink | Tracker connection, calibration, recording and real-time data | EyeLink hardware-control applications | Documentation only; hardware and separate SR Research SDK required |
| EyeLinkIO | Binary EDF reader, optional pandas/MNE export | Alternative EDF ingestion without edf2asc | Proposed only; pre-alpha caveat and native binary considerations |
| eyelinkparser | EyeLink ASC extraction around python-datamatrix | Alternative research ASC analysis | Proposed comparison; project describes limited documentation |
Recommended separation of duties¶
- Preserve the original EDF file and precise conversion settings.
- Parse offline into explicit recording segments, gaze channels and calibration/validation metadata.
- Keep detector/event annotations separate from original sample support.
- Declare participant identity, experimental trial mapping, timestamp origin and coordinate units.
- Only then use the SyeLink sample bridge, the pandas/Polars boundary and the native dense/sparse analysis guides.
SyeLink distinguishes three spaces: screen gaze pixels, raw camera pupil/CR pixels, and HREF angular units. Raw/HREF fields must not be used as screen x/y without a validated transformation. See the upstream coordinate guide.
Priorities for independent validation¶
A scientifically responsible next stage would compare SyeLink-parsed ASC samples with a separate EDF reference extraction on consent-cleared recordings. Check timestamp preservation, eye selection, recording-block boundaries, statuses, event timestamps and validation records. If both implementations agree on sample counts, that alone still cannot qualify statistical confidence-band coverage or identify field MNAR.
Python integrations · Research evidence · Eye-Tracking-BIDS metadata inspector.