eyetrajectoriespy

Functional analysis of continuous, sparse and irregular eye-tracking trajectories in Python.

Instead of reducing every trial immediately to fixation counts, dwell times or a symbolic scanpath, eyetrajectoriespy models gaze as a time-indexed planar process G(t) = [x(t), y(t)], with explicit observation and scientific-inference contracts.

Published stable: 1.1.0 · Python 3.11–3.13.
Published prerelease: 1.2.0rc2, qualified and published to GitHub/PyPI on 24 October 2026. Next stable version under final qualification: 1.2.0 (not published). This branch stages the exact final package identity and keeps production publication DISARMED until fresh final-version scientific/performance CI, scope checks, and separate release authorization. Ten workflow routes retain their qualified engineering boundaries; E1–E4 helpers remain explicitly experimental. The original prerelease evidence remains immutable.

Documentation · Start here · Choose a workflow · Figure atlas · Research evidence · News · Release status

Install stable 1.1.0

pip install eyetrajectoriespy==1.1.0

from eyetrajectoriespy import fit_mfpca, simulate_planar_trajectories

gaze = simulate_planar_trajectories(
    n_participants=20, trials_per_participant=6, random_state=7,
)
fit = fit_mfpca(gaze, n_components=0.95, scaling="dimension_sd")
print(fit.explained_variance_ratio)

This illustration uses deterministic synthetic data, not human-subject observations. For complete sparse examples, use the paired planar sparse guide.

What do you want to analyze?

Product and research qualification

A successful synthetic benchmark or independent artifact replay is not the same as calibrated statistical inference or production release qualification. Read the scientific evidence guide.

Research news, figures and reproducibility

The W3 R76–R95 programme has published documentation of influential-unit, covariance, missingness, external-validation and identifiability experiments, including negative results. The scientific study map links original papers; the figure atlas includes source-derived synthetic plots. The R93, R94 and R95 sharp-set proofs are conditional population mathematics—not confidence intervals or real-gaze MNAR corrections.

October 2026 news · Original and independent evidence · Validation ledger

Which workflow do I need?

Choose by the gaze observation grid, the participant/trial hierarchy, and the uncertainty claim, not by the number of available estimators. The analysis decision map and workflow comparisons explain those choices.

Where is the full advanced API?

The compact README is an orientation page, not an exhaustive function catalogue. Use the stable public API, module-scoped 1.1 additions and the exact scientific references:

• Mathematical contracts — what the numerical and method contracts guarantee.
• Function → equation index — documented functions mapped to their equations.
• Workflow atlas — scope and dependency map.
• Reference validation & performance envelope — qualification evidence and performance envelope.
• Portable scientific results — retained results and scientific provenance.
• Release process — immutable tags, candidate checks and publication interlocks.

Historical release lineage: 0.10.0 introduced native sparse univariate FPCA/PACE, and the immutable 0.11.0rc1 pre-release preceded the 0.11 validation expansion. Published 1.1.0 remains the stable installation target.

Documentation and contribution

Quick start · Input schemas · Mathematics · Visual gallery · Reproducibility guide · Contributing · Cite

The package complements, rather than replaces, upstream fixation/saccade detection, instrument quality control, AOI event analysis and symbolic scanpath pipelines. MIT © 2026 Stefanos Balaskas.