Automating model training lets you keep models fresh, consistent and scalable across many assets, without manually exporting data, running notebooks or uploading new versions.

Instead of training your models in an external environment and uploading them once, you define a Pipeline with a Model step and a training script. Then you run that pipeline in training mode (via Experiments, Triggers or direct API calls), and Wizata will version and store the resulting models in the internal model repository.

Why automate model training ?

Use automatic training when you need your models to:

• Adapt to data drift- Processes change: sensors age, raw materials shift, setpoints are tuned. Automated retraining lets models follow reality instead of becoming stale.
• Scale across many assets (Twins) - The same training pipeline can produce one model per asset (Twin) and per property, using the model identification pattern (Understanding Model Storage and Metadata in Wizata). 
• Stay reproducible and traceable- Each training run is an Execution linked to a Pipeline, an Experiment, and a Twins. Logs, plots and metadata are stored automatically. 
• Be safely deployed in production- The same Pipeline definition is used both for training and for prediction. You control via options (train, plot, write) whether a specific run should retrain, only score, or also write back results. 

If your use case is a one-off or prototypes only, uploading a manually trained model may be simpler; for anything recurrent, multi-asset or production-grade, auto-training through Pipelines is recommended.

How automatic training works in Wizata ?

At a high level, automatic training is:

1. A Pipeline that defines the training logic

• Query step: fetches training data.
• Script step: cleans, transforms, creates features/targets.
• Model step: calls a training script, creates or updates a model in the repository.
• Optional Plot step/ Write step: log metrics, store KPIs or sample predictions.

1. A training script executed by the Model step
2. An Execution that runs the pipeline in training mode
3. A Trigger (optional) to schedule retraining