Garmin Analysis Documentation

Bayesian analysis of health and fitness data from Garmin devices
🎯 Current Model: Constrained AR(1) Spline with Student-t Likelihood | Stan: weight_state_space_training_decay_aerobic_symmetric_student_ar_spline_constrained.stan | Key feature: AR coefficient ρ constrained to [-0.5, 0.5] to prevent overfitting. Detailed component predictions in component_predictions/ directory.
🚧 Playground Work-in-Progress Notice: This documentation and associated models represent an experimental playground for Bayesian analysis of personal health data. Models, visualizations, and conclusions are actively evolving and should be considered exploratory research rather than production-ready analysis.

⭐ Current Model Output

Component Predictions: Output from the constrained AR(1) spline model. Includes hourly predictions, component breakdown by activity type, and daily patterns in weight variation.

View Component Predictions →

📖 Model Interpretability Report

Constrained AR(1) Spline Model: Comprehensive interpretability report with model overview, prior specification, posterior parameter tables, MCMC diagnostics, dynamic variance decomposition, and key interpretation of fitness effects on weight.

View Interpretability Report →

📈 Trend Model Comparison

Trend Model Variant: Adds an explicit linear trend parameter (δ) to test whether weight changes show a secular trend independent of fitness effects. Compares posterior estimates to the base constrained AR(1) model.

View Trend Report →

Technical Model Documentation

Mathematical and computational documentation of the four-fitness state-space model (previous model architecture). Includes equations, Stan implementation, and parameter sensitivity analysis.

View Technical Documentation →

Project Guidelines (CLAUDE.md)

Project-specific instructions for AI assistants, including tooling requirements, project structure, and development methodologies.

View HTML

Workout Data Report

Analysis of Garmin workout data quality, aggregation methods, and recommendations for modeling. Focuses on strength training data for cross-lagged time-series analysis.

View HTML

📦 Archive & Previous Models

🌟 Enhanced Model Suite (Master Index)

Central hub for enhanced sensitivity model reports. (Previous Model Variant)

View Suite →

📊 Enhanced Sensitivity Report

Previous model analysis with detailed parameter estimates, variance decomposition, and diagnostics.

View Report →

📈 Enhanced Model Visualizations

Visualizations from previous model variant including time series and parameter distributions.

View Visualizations →

🏃 Enhanced Fitness Time Series

Fitness state time series from enhanced sensitivity model with conservative sampling.

View Time Series →

🔬 Enhanced Model Overview

Technical overview from previous enhanced sensitivity model variant.

View Overview →

Consolidation Summary

Summary of model consolidation and cleanup process documentation.

View Summary