Utilities for fusionlab-learn

The fusionlab-learn library provides a rich set of utility functions to support various stages of the machine learning and time series forecasting workflow. These utilities are designed to simplify common tasks, from data preparation and feature engineering to specialized data handling for geospatial and physics-informed models.

This section is organized thematically to help you find the right tool for your task.

Data Handling & Preparation

• Data Manipulation Utilities
  • Advanced NaN Handling (nan_ops)
  • Reshaping Time Series Data (widen_temporal_columns)
  • Conditional Data Masking (mask_by_reference)
• Time Series Utilities
  • Datetime Handling & Filtering
  • Trend & Seasonality Analysis
  • Feature Engineering
  • Feature Selection & Reduction

Geospatial & Spatial Utilities

• Geospatial & Time Series Data Utilities
  • Temporal Gap Interpolation (interpolate_temporal_gaps)
  • Data Augmentation via Noise Injection (augment_series_features)
  • Spatiotemporal Data Augmentation Pipeline (augment_spatiotemporal_data)
  • Generating Synthetic PINN Data (generate_dummy_pinn_data)
• Spatial Data Utilities
  • Spatial Clustering (create_spatial_clusters)
  • Stratified Spatial Sampling
  • Filtering and Merging by Position
  • Other Utilities

Neural Network & PINN Helpers

• Neural Network Utilities
  • Anomaly Score Calculation
  • Data Preparation & Preprocessing
  • Forecasting & Visualization
  • Data Reshaping Utilities
• PINN Data Utilities
  • Sequence Generation for PINNs (prepare_pinn_data_sequences)
  • Formatting Model Outputs (format_pinn_predictions)
  • Coordinate and Feature Scaling (normalize_for_pinn)
  • Coordinate Extraction Utilities

Forecasting Workflow & Results

• Forecast Data Formatting Utilities
  • Pivoting from Long to Wide Format (pivot_forecast_dataframe)
  • Smart Format Detection and Conversion (format_forecast_dataframe)