Welcome to Vangja’s documentation!

Home

• vangja
  • Key Features
  • Installation
  • Usage
  • Contributing

User Guide

• Chapter 01: Getting Started
  • In This Notebook
  • Setup and Imports
  • 1. Load Air Passengers Dataset
  • 2. Train/Test Split
  • 3. Model Air Passengers like Facebook Prophet
  • Summary
• Chapter 02: Inference Algorithms
  • Setup and Imports
  • Load and Prepare Data
  • Define the Multiplicative Model
  • 1. Maximum A Posteriori (MAP) Estimation
  • 2. Variational Inference (ADVI)
  • 3. Markov Chain Monte Carlo (NUTS)
  • Comparison of Results
  • Summary: Choosing an Inference Method
• Chapter 03: Multi-Series Fitting
  • Setup and Imports
  • Generate Synthetic Multi-Store Data
  • Prepare Train/Test Splits
  • Approach 1: Sequential Fitting (Prophet-style)
  • Approach 2: Simultaneous Fitting (Vangja-style)
  • Comparison of Results
  • Summary
• Chapter 04: Multi-Series Fitting Caveats
  • Setup and Imports
  • Load Real-World Datasets
  • Understanding the Caveat: Changepoint Distribution
  • Demonstrating the Impact
  • Comparison: Sequential vs Simultaneous
  • Summary and Recommendations
• Chapter 05: Hierarchical Modeling with Partial Pooling
  • Why Hierarchical Modeling?
  • In This Notebook
  • Setup and Imports
  • Data Generation
  • Approach 1: Individual Fitting (No Pooling)
  • Approach 2: Partial Pooling (Hierarchical)
  • A Note on Complete Pooling
  • Comparison of Approaches
  • Summary and Key Takeaways
• Chapter 06: Caveats of Hierarchical Modeling
  • Setup and Imports
  • Data Generation
  • Caveat 1: Shrinkage Strength
  • Caveat 2: Opposite Seasonality and the UniformConstant
  • Summary
• Chapter 07: Transfer Learning from Long Time Series
  • The Scenario
  • Differences from the Original Blog Posts
  • Setup and Imports
  • Load and Explore the Data
  • Baseline: Prophet-like Model Without Transfer Learning
  • Step 1: Learn Seasonality from Temperature Data
  • Step 2: Transfer Learning to the Sales Model
  • Comparison: Baseline vs Transfer Learning
  • How It Works: The Parametric Transfer
  • Limitations and Considerations
  • Summary
• Chapter 08: Transfer Learning with prior_from_idata and Hierarchical Modeling
  • Overview
  • Setup and Imports
  • Step 1: Load Data and Fit the Temperature Model
  • Step 2: Transfer Learning with prior_from_idata
  • Step 3: Combining Hierarchical Modeling with Transfer Learning
  • Full Comparison: All Approaches
  • Summary
• Chapter 09: Advanced Transfer Learning Options
  • 1. Bidirectional Changepoints with delta_side
  • 2. Regularization with loss_factor_for_tune
  • 3. Phase Alignment with shift_for_tune
  • 4. Custom Summary Statistics with override_... Parameters
  • Summary and Caveats
• Chapter 10: Bayesian Workflow — Diagnostics and Best Practices
  • In This Notebook
  • Setup and Imports
  • 1. Data Loading
  • 2. Prior Predictive Checks
  • 3. Convergence Diagnostics
  • 4. Posterior Predictive Checks
  • 5. Transfer Learning Model
  • 6. Full Posterior Summaries
  • 7. Model Comparison (WAIC / LOO-CV)
  • 8. Prior Sensitivity Analysis
  • 9. Prior-to-Posterior Visualisation
  • 10. Prediction with Uncertainty
  • 11. Quantitative Comparison
  • 12. Component Plots
  • Summary: Bayesian Workflow Checklist
• Chapter 11: Uncertainty Estimation in Vangja
  • Overview
  • 1. MCMC / VI: Posterior Draw Propagation
  • 2. MAP / MAPX: Residual-Calibrated Intervals
  • Usage
  • Implementation Notes

Case Studies

• Case Study 01: Smart Home Appliances Energy Consumption Forecasting
  • 1. Dataset
  • 2. Training the base temperature model
  • 3. Training the energy consumption model
  • 4. Ablation study

API Reference

• API Reference
  • Time Series Models
    • Base Class
    • Combined Models
  • Components
    • LinearTrend
    • FlatTrend
    • FourierSeasonality
    • NormalConstant
    • BetaConstant
    • UniformConstant
  • Utilities
    • get_group_definition
    • filter_predictions_by_series
    • metrics
    • remove_random_gaps
    • compare_models
    • prior_sensitivity_analysis
    • prior_predictive_coverage
    • plot_prior_posterior
    • plot_posterior_predictive
    • plot_prior_predictive
  • Datasets
    • Real-World Datasets
    • Type Aliases
    • Synthetic Data Generation
    • Stock Data
  • Types
    • Type Aliases
    • TypedDicts
    • Scaler
    • ScaleMode
    • YScaleParams
    • TScaleParams
    • Method
    • OptimizationMethod
    • NutsSampler
    • FreqStr
    • TuneMethod
    • PoolType

Indices and tables

• Index

• Module Index

• Search Page