pypots-developer

name	pypots-developer
description	Convert a local paper and local implementation into a PyPOTS-style model integration draft with matching test drafts; use this skill when the user wants to migrate an existing model into PyPOTS, add task exports, add HPO registration, or produce a structured migration draft instead of editing the real PyPOTS source tree directly.

PyPOTS Developer

Overview

This skill turns a local paper + local existing codebase into a PyPOTS-style migration draft.

It does not write directly into the real PyPOTS source tree. Instead, it writes all generated results into output/<task>-<model-name>/, using mirrorable pypots/ and tests/ roots so the new files can be copied into PyPOTS with minimal reshaping.

When To Use

Use this skill in these situations:

• The user wants to migrate a paper implementation into PyPOTS.
• The user provides an existing open-source implementation and wants it reorganized into PyPOTS-style model.py / core.py / data.py / tests.
• The user wants to add the required task export patch, HPO registry patch, and test draft for a new PyPOTS model.
• The user wants a migration feasibility pass first and then a complete patch draft.

Do not use this skill when:

• The task is only to explain an existing PyPOTS model.
• The task is only to write end-user API documentation.
• The task is a small bug fix rather than a model migration.

Required Inputs

Prefer to gather these inputs first:

• local paper PDF path
• local extracted source-code directory path
• target model name
• optional target task

If the user does not provide the task, infer it first. If the inference is not stable, stop and ask instead of guessing.

Workflow Decision Tree

1. Read the paper and source code, then build a migration manifest.
   • Use scripts/build_manifest.py to extract task clues, training-loop clues, input tensor clues, and output-semantic clues.
2. Identify the PyPOTS task.
   • See references/task-routing.md
3. Identify the implementation path.
   • See references/path-selection.md
   • Choose from standard-nn, complex-nn, or non-nn
4. Map the model into the PyPOTS file layout.
   • See references/workflow.md
   • Reuse templates from assets/ instead of starting from scratch
5. Initialize the output directory.
   • Use scripts/init_output.py
6. Generate the complete patch draft.
   • New files go to output/<task>-<model-name>/pypots/...
   • Tests go to output/<task>-<model-name>/tests/...
   • Existing-file changes are written as .patch files under the mirrored pypots/... target paths
7. Check integration points and test coverage.
   • Exports and HPO: references/integration-points.md
   • Test coverage: references/test-matrix.md
8. Run scripts/validate_output.py at the end.

Non-negotiable Rules

• Do not write directly into the real PyPOTS source tree.
• Write all generated results into output/.
• If task routing is unstable, describe the conflict first and then ask the user.
• If the training-path decision is unstable, explain why standard-nn vs complex-nn is unclear before asking.
• pypots_cli.py usually does not need a patch in v1; do not add a new CLI subcommand just to make the output look more complete.
• non-nn models do not get an HPO registry patch by default; the reason must be stated explicitly in migration-report.md.

Worked Example

• Input example: example/
  • 2202.08516v5.pdf
  • SAITS-main/
• Output example: output/imputation-saits/
  • mirrorable pypots/ draft
  • mirrorable tests/ draft
  • task export patch
  • HPO patch
  • migration report

Read the worked example first before migrating a new model.

Resources

references/

• workflow.md: full migration workflow
• task-routing.md: task-routing rules
• path-selection.md: standard / complex / non-NN path rules
• integration-points.md: exports, HPO, and external package path rules
• test-matrix.md: test generation matrix
• maintenance.md: internal maintenance rules for extending this skill

scripts/

• build_manifest.py: scan the paper and source code, then produce a migration manifest
• init_output.py: initialize the output directory
• validate_output.py: validate whether the generated output reaches the minimum deliverable bar

assets/

• standard-nn/: standard NN skeletons
• complex-nn/: complex NN skeletons
• non-nn/: non-NN skeletons
• common/: shared migration-report template