Configuración YAML

Pipeline declarativo y auditable.

godml.yml (⚙️ Configuración Principal)

name: mi-modelo-churn
version: 1.0.0
provider: mlflow
description: "Modelo de clasificación para predecir baja de clientes"

# 📊 Dataset
dataset:
  uri: ./data/churn.csv
  hash: auto                    # Hash automático para trazabilidad

# 🤖 Modelo
model:
  type: xgboost                 # xgboost | random_forest | lightgbm
  source: core                  # core | custom
  hyperparameters:
    max_depth: 5
    eta: 0.3
    objective: binary:logistic

# 📈 Métricas de calidad
metrics:
  - name: auc
    threshold: 0.85             # Umbral mínimo requerido
  - name: accuracy
    threshold: 0.80

# 🏛️ Gobernanza
governance:
  owner: "equipo-ml@empresa.com"
  tags:
    - project: churn-prediction
    - environment: development
    - compliance: gdpr

# 🚀 Despliegue
deploy:
  realtime: false
  batch_output: ./outputs/predictions.csv

# 🐳 Configuración de entornos
deploy_config:
  dev:
    docker_tag: mi-modelo:dev
    port: 8000
    host: 0.0.0.0
  production:
    docker_tag: mi-modelo:prod
    port: 8080
    host: 0.0.0.0

📊 DataPrep Service — Preparación Declarativa

godml.yml (Estructura de receta YAML)

# recipe.yml - Receta de preparación de datos
dataprep:
  # 📥 INPUTS
  inputs:
    - name: raw_customers
      connector: csv
      uri: ./data/customers.csv
      options:
        sep: ","
        encoding: utf-8

  # 🔧 TRANSFORMACIONES
  steps:
    # Seleccionar columnas
    - op: select
      params:
        columns: ["id", "age", "income", "email"]

    # Rellenar valores nulos
    - op: fillna
      params:
        strategy: mean
        columns: ["age", "income"]

    # Codificación one-hot
    - op: one_hot
      params:
        columns: ["category"]
        drop_first: true

    # Escalado estándar
    - op: standard_scale
      params:
        columns: ["income", "age"]

    # Eliminar duplicados
    - op: drop_duplicates
      params:
        subset: ["id"]

  # ✅ VALIDACIONES
  validations:
    - type: expect_non_null
      args: ["id", "email"]

    - type: expect_unique
      args: ["id"]

    - type: expect_range
      args:
        column: age
        min: 18
        max: 100

  # 📤 OUTPUTS
  outputs:
    - name: clean_data
      connector: csv
      uri: ./data/customers_clean.csv

# 🔒 COMPLIANCE
governance:
  compliance: pci-dss  # Aplica automáticamente PCI-DSS