Principle:Sdv dev SDV Sequential Model Fitting
| Knowledge Sources | |
|---|---|
| Domains | Time_Series, Machine_Learning |
| Last Updated | 2026-02-14 00:00 GMT |
Overview
A two-stage fitting process that trains separate models for entity context attributes and temporal sequence patterns in sequential data.
Description
Sequential model fitting trains two sub-models: a GaussianCopulaSynthesizer for context columns (entity-level attributes) and a DeepEcho PARModel for sequential columns (time-varying data). The fitting process separates the data into context and sequence components, fits the context model first, then assembles sequences and fits the autoregressive model.
Usage
Call fit on a PARSynthesizer after initialization. The data must be a DataFrame containing the sequence key, context columns, and sequential columns.
Theoretical Basis
- Context extraction: Separate context columns (constant per sequence) from sequential columns
- Context model fitting: Fit GaussianCopulaSynthesizer on context data (one row per sequence)
- Sequence assembly: Group rows by sequence key, order by sequence index
- PAR model fitting: Fit DeepEcho PARModel on assembled sequences with context types and data types
Related Pages
Implemented By
Page Connections
Double-click a node to navigate. Hold to expand connections.
Principle
Implementation
Heuristic
Environment