Tutorials & Example Code

Basics

• Merlion’s Data Format
  • UnivariateTimeSeries: The Basic Building Block
  • TimeSeries: Merlion’s Standard Data Class
  • Time Series Indexing & Alignment
  • TimeSeries: A Few Useful Features
• Loading Custom Datasets
  • Anomaly Detection Datasets
  • General Purpose (Forecasting) Datasets
  • Broader Takeaways

Anomaly Detection

• A Gentle Introduction to Anomaly Detection in Merlion
• How to Use Anomaly Detectors in Merlion
  • Model Initialization
  • Model Training
  • Model Inference
  • Quantitative Evaluation
  • Model Visualization
  • Saving & Loading Models
  • Simulating Live Model Deployment
• Multivariate Time Series Anomaly Detection
  • Model Initialization and Training
  • Model Inference and Quantitative Evaluation
• Adding New Anomaly Detection Models
  • Model Config Class
  • Model Class
  • Running the Model: A Simple Example
  • Visualization
  • Customizing the Post-Rule
  • Quantitative Evaluation

Forecasting

• A Gentle Introduction to Forecasting in Merlion
• How to Use Forecasters in Merlion
  • Model Initialization
  • Model Training
  • Model Inference
  • Model Visualization and Quantitative Evaluation
  • Saving & Loading Models
  • Simulating Live Model Deployment
• Multivariate Time Series Forecasting
  • Model Initialization and Training
  • Model Inference and Quantitative Evaluation
• Forecasting With Exogenous Regressors
• Adding a New Forecasting Model
  • Model Config Class
  • Model Class
  • Running the Model: A Simple Example
  • Visualization
  • Quantitative Evaluation
  • Defining a Forecaster-Based Anomaly Detector

Advanced Features

• Tutorial for AutoSARIMA Forecasting Model
  • Prepare dataset
  • Train a full AutoSarima model with approximation (suggested, default)
  • Train a full AutoSarima model without approximation (slower)
  • Train a partial autosarima model
• Proof of Concept: Inverse Transforms for Forecasters