eyetrajectoriespy 0.9.0rc1¶
Release candidate — 27 September 2026
This is the first coordinated public release candidate for eyetrajectoriespy.
It marks the transition from active method expansion to a scientifically
qualified pre-1.0 platform.
What is in the release candidate¶
The package exposes five canonical scientific workflows:
- continuous gaze exploration and FPCA;
- experimental functional regression;
- repeated-trial functional mixed effects;
- generalized binary/count functional responses;
- nonlinear and recurrence trajectory analysis.
The release candidate includes the 0.55–0.57 stabilization work:
- machine-checked canonical workflow hierarchy;
- compatibility-first API/deprecation policy;
- analytical, independent-implementation and simulation-recovery validation ledgers;
- explicit numerical-tolerance policy;
- repeated runtime/peak-memory performance qualification;
- portable JSON + NPZ scientific-result snapshots;
- package/Python/dependency/platform environment capture;
- realistic executable examples for all five canonical workflows;
- build/twine/fresh-wheel/fresh-sdist/install smoke qualification;
- coordinated GitHub Release + PyPI automation using one build artifact and PyPI Trusted Publishing.
Scientific scope¶
The generalized functional-response layer supports Bernoulli/grouped-binomial logit and Poisson expected-count/rate marginal GEE contracts. Additional families such as negative-binomial, zero-inflated, hurdle and Tweedie are not part of this release candidate.
The Gaussian functional mixed-effects layer supports participant functional random effects, guarded participant random slopes, nested trial functional effects, residual covariance/whitening, full-refit bootstrap and covariance sensitivity.
The nonlinear layer includes recurrence/RQA, recurrence networks, Lyapunov diagnostics, surrogate testing, transfer entropy and return-map analysis with method-specific interpretation boundaries.
Validation¶
Release-candidate qualification requires:
- Python 3.11–3.13 core CI on Ubuntu, Windows and macOS;
- package build;
- documentation build;
- canonical examples;
- scikit-fda qualification;
- FDApy sparse qualification on Python 3.11–3.12;
- performance-envelope qualification;
- release-readiness build/twine/wheel/sdist/install qualification.
The configured coverage floor remains 90%; independent scientific reference validation is treated as more important than increasing line coverage for its own sake.
Reproducibility¶
Portable scientific-result snapshots retain scientific arrays, identifiers, specifications, units, diagnostics and provenance while explicitly marking backend-native objects as nonportable. Environment capture records package, Python, dependency/backend and platform metadata plus Git commit when available.
Installation¶
Because this is a prerelease:
pip install --pre eyetrajectoriespy==0.9.0rc1
Release-candidate expectations¶
The 0.9.x release-candidate phase is for defect correction, installation feedback, documentation clarity and reproducibility verification. It is not a signal to resume broad estimator expansion before 1.0 readiness is assessed.