Validate Training Setup
This page documents the lightweight validation helper used to confirm that a TE training configuration can produce a valid first batch and finite metrics.
One-batch validation check for TE training and tree baselines.
- scripts.training.validate_training_setup.build_validation_summary(config_path, output_directory, batch_summary, batch_output_dictionary, training_config)[source]
Build the persisted validation summary for one setup check.
- Parameters:
config_path (Path) – Canonical training configuration path.
output_directory (Path) – Validation artifact directory.
batch_summary (dict[str, object]) – Structural batch summary from the validation step.
batch_output_dictionary (dict[str, Tensor]) – Output tensors and metrics computed by the regression module.
training_config (dict[str, object]) – Prepared training configuration with artifact metadata.
- Returns:
YAML-serializable validation summary.
- Return type:
dict[str, object]
- scripts.training.validate_training_setup.validate_training_setup(config_path, output_suffix='validation_check')[source]
Run a lightweight validation check for the configured training setup.
- Parameters:
config_path (Path) – YAML training configuration path.
output_suffix (str) – Suffix appended to the validation artifact run name.
- Returns:
Validation summary YAML path plus generated Markdown report path.
- Return type:
tuple[Path, Path]