Model Integration Guide

This section walks you through integrating a new model into PyPOTS. Choose the path that matches your model before writing any code.

• Standard NN Path — one optimizer, default training loop (most models)

• Complex NN Path — multiple optimizers or custom training orchestration (e.g. GANs)

• Non-NN Path — rule-based, statistical, or algorithmic models (no gradients)

• Standard NN Integration Path
  • When This Path Is Correct
  • Start From the Task Template
• Step-by-Step Implementation Guide
  • Step 1: Pick the Task Contract
  • Step 2: Implement core.py
  • Step 3: Implement model.py
  • Step 4: Add data.py Only If Needed
  • Step 5: Wire the Package
  • Step 6: Keep predict() Boring
  • SAITS Walkthrough Summary
  • Definition of Done
• Complex NN Integration Path
  • When This Path Is Correct
  • What Stays The Same
  • USGAN as the Reference Pattern
• Implementation Guide
  • Step 1: Start From the Standard Path
  • Step 2: Implement the Custom Training Loop
  • Step 3: Handle the Core Differently
  • What You Are Allowed to Customize
  • High-Risk Mistakes
  • Definition of Done
• Non-NN Integration Path
  • When This Path Is Correct
  • LOCF as the Reference Pattern
  • Two Valid Non-NN Styles
• Step-by-Step Implementation Guide
  • Step 1: Choose the Base Class
  • Step 2: Implement fit()
  • Step 3: Implement predict()
  • Step 4: Implement Helper Methods
  • Step 5: Wire the Package
  • Common Mistake
  • Definition of Done