Skip to content

Learning paths

Use this page when you know **what kind of help you need** but not which gpbiometricspy page to open. The documentation intentionally separates worked learning, task completion, factual reference, and scientific explanation so a tutorial does not have to double as an API manual and a reference page does not pretend to be a research workflow.
Learn by doing Solve a task Look up facts Understand why Evidence first

Choose the kind of help you need

New to the package

Use this sequence when the immediate goal is to become productive without starting from the 406-function reference.

  1. First analysis — get a small, successful synthetic workflow running.
  2. Hands-on EDA research guide — see QC, signal processing, event candidates, plots, exports and reporting evidence in one connected analysis.
  3. Workflow map — switch from the teaching example to the modality or task you actually recorded.
  4. Plot gallery — inspect generated outputs and the analyses that create them.
  5. API by domain — move to exact function-level reference only when implementation detail is needed.
Learning rule: reproduce the bundled example before replacing inputs. A successful run on your own file is not evidence that units, clocks, channel activity, event meaning or provenance were interpreted correctly.

Starting a real research project

For new or unfamiliar data, use the provenance-first onboarding sequence rather than jumping directly into preprocessing.

1 · ProjectCreate the project scaffold so raw data, metadata, mappings, QC, events, derived outputs, models and reports remain separate.
2 · DefineDefine study metadata and the data dictionary, including roles, units, identifiers, clocks, provenance and event semantics.
3 · AdaptAdapt the export without hiding provenance and preserve the source-to-standard mapping.
4 · ValidateValidate the dataset for schema, timing, activity, missingness, resets and event coverage.
5 · DecideReview QC and exclusion decisions with explicit scope and denominator consequences.
6 · AnalyzeChoose the modality workflow only after the evidence required by the next claim is defensible.

Choose by research domain

You are working on… Start here Then inspect Before interpretation
EDA / GSR / SCR EDA / GSR / SCR example Hands-on EDA research Measurement accountability
PPG / HR / IBI / HRV PPG / HRV example Cardiac source provenance Interpretation guardrails
Pupil / gaze / fixation / AOI Pupil / gaze / AOI example Eye-tracking ecosystem bridges Measurement accountability
Event-linked multimodal data Multimodal example Timebase and alignment guide Timebase provenance
External tool interoperability Interoperability example Integrations Interoperability version testing
Grouped prediction Modelling strategy guide Grouped mixed-effects boosting Interpretation guardrails
Hierarchical variability models Methods overview Choosing a location–scale model Interpretation guardrails

Choose by role

Researcher bringing new data

Use Project → Define → Adapt → Validate → Decide → Analyze → Report. Keep source files immutable, write down field meaning before code depends on it, and separate QC evidence from exclusion decisions.

Researcher reproducing or extending an analysis

Start from the end-to-end runnable EDA example and reporting/reproducibility guide. Reuse explicit settings and manifests instead of reconstructing analysis choices from figures alone.

Reviewer, collaborator or auditor

Read Parity & validation, Deep validation, Private real-data validation, QC & exclusion decision ledger, Measurement accountability and Interpretation guardrails. These pages are the shortest path to what the package claims, how those claims are tested, and where they stop.

Developer or advanced integrator

Use API by domain, Complete 406-function reference, Integrations and Development. The tutorial layer is useful for context, but implementation work should be anchored to the reference and validation contracts.

Common question → best page

Question Best page
“What should I do first?” Start here
“How should I structure a new study project?” Research project scaffold
“What does this field mean?” Signal and column glossary
“How do I document units, clocks and event labels?” Study metadata and data dictionary
“How do I adapt an unfamiliar export?” Bring your own export safely
“Is this dataset ready to analyze?” Validate a new dataset
“A QC check failed—what now?” Troubleshooting and diagnostics
“Should I remove these rows or participants?” QC & exclusion decision ledger
“How do I align events or modalities?” Timebase and alignment
“Which model family fits my design?” Choose a modelling strategy
“How should I report this analysis?” Reporting and reproducibility
“Which exact function do I need?” API by domain
“What evidence supports package parity?” Deep validation
“What claims should I avoid?” Interpretation guardrails

Do not treat the documentation types as interchangeable

A tutorial is allowed to make choices for a controlled teaching example. A how-to guide assumes you already have a concrete goal. Reference describes the software and its interfaces. Explanation discusses why the scientific workflow is designed as it is. Crossing those purposes carelessly creates two common errors: copying tutorial defaults into a study without review, or treating a concise API description as scientific justification for a measurement claim.

No documentation route changes the evidence requirement. Column names do not establish units; parser success does not establish clock ownership; non-missing values do not establish signal quality; detected events do not establish psychological meaning; model fit does not establish causality; and software validation does not substitute for study-level measurement validity.