Multimodal modelling with external face-window summaries

This article is the Python counterpart of the corresponding gp3tools workflow. It uses bundled synthetic data so it can be run without private participant exports.

import gp3tools as gp3

data = gp3.load_example_master()
# read_gazepoint_face_export is available as gp3.read_gazepoint_face_export(...)
# audit_gazepoint_face_quality is available as gp3.audit_gazepoint_face_quality(...)
# sync_gazepoint_face_data is available as gp3.sync_gazepoint_face_data(...)
# prepare_gazepoint_multimodal_data is available as gp3.prepare_gazepoint_multimodal_data(...)

Recommended workflow

1. Inspect column availability and create/validate a master table.
2. Run the relevant quality gates before transformation or modelling.
3. Keep preprocessing and exclusion decisions explicit.
4. Save derived tables/plots with package/version metadata.
5. For backend-adapted statistical functions, report the Python backend and validate the inferential target against the R reference when confirmatory equivalence matters.

Backend note

The Python migration preserves the workflow and estimand intent but does not assert numerical identity with R-specific engines. Validate confirmatory results against the frozen R implementation until a dedicated parity fixture exists.

Reproducibility checkpoint

import gp3tools as gp3

print(gp3.__version__)
print(gp3.api_status().query("r_export in @funcs"))