Validation Method and Examples¶
Purpose¶
This page summarizes the validation methodology (e.g. “Method 3” or other named procedures) and points to concrete examples so you can reproduce and extend validation in your environment.
Method 3 (Conceptual)¶
When we refer to “Method 3” or a specific validation method in the codebase, it usually involves:
- A by-case train/validation split (see Train/validation split by case).
- One or more inference checks: e.g. (1) run the model on a sample of feature rows and (2) run it on raw windows via the same feature pipeline, then compare outputs and shapes.
- Threshold checks: Compare validation metrics (precision, recall, F1, etc.) to configurable minimums.
The exact steps and script names are in the validation scripts and config (e.g. validate_model_configurable.py, model_validation_config.yml); see Model validation.
Examples¶
- Validation examples in the repo (if present) typically show: loading a trained model, loading a feature schema and dataset (or raw windows), running inference, and comparing to labels or to expected ranges.
- To run a full validation pass with reporting, use the configurable validation script and the same feature/config setup as in training.
For full detail, refer to the validation config file and the Model validation and Train/validation split by case pages.