Saltar al contenido principal

Ejemplo: Retail Demand Forecast

Predicción de demanda semanal con LSTM para un retailer de e-commerce.

godml.yml​

name: retail-demand-forecast
version: 1.0.0
provider: mlflow

dataset:
  uri: ./data/weekly_sales.csv
  target: units_sold
  dataprep:
    steps:
      - op: drop_columns
        params:
          columns: [store_id, sku_description]
      - op: fill_missing
        params:
          strategy: median
          columns: [price, discount_pct, stock_level]
      - op: scale
        params:
          columns: [price, discount_pct, stock_level, promo_spend]
          method: minmax

governance:
  owner: equipo-supply-chain
  tags:
    - domain: retail
    - task: demand-forecast

model:
  type: lstm
  hyperparameters:
    eta: 0.001
    n_estimators: 100    # epochs
    random_state: 42

metrics:
  - name: auc
    threshold: 0.70

deploy:
  batch_output: ./output/demand_forecast.csv
  model_output: ./output/lstm_model.pkl

Instalar y ejecutar​

pip install godml[deep]   # incluye TensorFlow + Keras
godml run -f godml.yml

Alternativa tabular (sin GPU)​

model:
  type: random_forest
  hyperparameters:
    n_estimators: 200
    max_depth: 10
pip install godml
godml run -f godml.yml

Stack: XGBoost + MLflow + despliegue FastAPI en contenedor.

Incluye entrenamiento, evaluación (MSE) y servicio realtime.

Proximamente​