La interfaz está localizada; las descripciones y documentos de las Skills se traducen automáticamente con IA a su idioma. El código, los comandos, rutas y términos técnicos se mantienen en el original, y prevalece el documento del autor.
Explorar Skills
Diseño y creatividad · SKILL.md

constraints

constraints

Route SDV constraint design, serialization, loading, and debugging workflows.

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentación de la Skill

GitHub

name: constraints description: "Route SDV constraint design, serialization, loading, and debugging workflows." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

SDV constraints

Use this sub-skill when the task is about adding or debugging SDV logical constraints: built-in CAG constraints, legacy tabular constraints, custom constraint factories, programmable constraints, constraints JSON files, or synthesizer constraint methods.

Route here for

  • Natural requests such as "add SDV constraints", "custom constraint", "programmable constraint", "load constraints JSON", FixedCombinations, Inequality, or "constraint validation error".
  • Choosing between current sdv.cag constraints and legacy sdv.constraints tabular constraints.
  • Creating, saving, loading, and attaching constraints with add_constraints, get_constraints, set_constraints, and load_constraints.
  • Diagnosing constraint metadata, data-shape, sdtype, table-name, serialization, and custom-function failures.

Do not own

  • Metadata table and column creation: route those decisions to the data-preparation sub-skill.
  • Synthesizer model selection, fitting, sampling, conditions, or save/load model lifecycle: route to the single-table, multi-table, or sequential synthesis sub-skills after constraints are attached.
  • Quality metrics or visual evaluation of constrained synthetic output: route to the evaluation sub-skill.

Operating path

  1. Confirm the metadata already has the target table and columns. For multi-table metadata, supply table_name on every one-table CAG constraint.
  2. Prefer current sdv.cag objects for synthesizer.add_constraints([...]); use legacy sdv.constraints only for old tabular/data-processor workflows or when translating existing legacy constraint definitions.
  3. Add constraints before fit when possible. If constraints are added after fitting, refit the synthesizer before relying on them.
  4. For custom logic, use create_custom_constraint_class only when the legacy tabular interface is required; otherwise prefer a ProgrammableConstraint subclass.
  5. For JSON round-trips, use the current CAG dictionary shape and sdv.utils.load_constraints, then compare the number of loaded constraints with the file entries because unknown classes are warned and skipped.
  6. When an error names a column, sdtype, table_name, row index, or is_valid return type, jump directly to troubleshooting.

References

  • API reference: constructors, current-vs-legacy distinction, JSON shapes, and synthesizer methods.
  • Workflows: concise recipes for built-ins, custom classes, programmable classes, JSON load/save, and refitting.
  • Troubleshooting: symptom-to-fix table for common constraint failures.
Skills guardadas