Model registry

Questions this page answers - How do I save and load trained foodspec pipelines? - What metadata is stored? - How do I apply a saved model in a new session?

What metadata is stored?

• name, version, foodspec_version
• Timestamp
• Model artifact (.joblib) + metadata JSON
• extra: label/model configuration, any custom fields (e.g., classifier name, label column)

End-to-end example (Python)

Train (e.g., oil-auth), save, then load and apply to new data.

from foodspec import load_library
from foodspec.apps.oils import run_oil_authentication_workflow
from foodspec.model_registry import save_model, load_model

# Train on a library
fs_train = load_library("libraries/oils_train.h5")
res = run_oil_authentication_workflow(fs_train, label_column="oil_type", classifier_name="rf", cv_splits=5)

# Save pipeline
save_model(
    model=res.pipeline,
    path="models/oil_rf_v1",
    name="oil_rf",
    version="0.2.0",
    foodspec_version="0.2.0",
    extra={"label_column": "oil_type", "classifier_name": "rf"},
)

# Load and apply to new samples
model_loaded, meta = load_model("models/oil_rf_v1")
fs_new = load_library("libraries/oils_new.h5")
preds = model_loaded.predict(fs_new.x)
print(meta, preds[:5])

CLI flow

• Save during workflow:

  foodspec oil-auth libraries/oils_train.h5 \
    --label-column oil_type \
    --classifier-name rf \
    --save-model models/oil_rf_v1 \
    --output-dir runs/oils_rf
  

• Inspect metadata:

  foodspec model-info models/oil_rf_v1
  

  To apply the saved model, load it in Python (as above) and predict on a new HDF5 library.

Best practices

• Version models (e.g., v1, v2) and record foodspec_version.
• Store training data summary (labels, date, preprocessing choices) in extra.
• Keep artifacts (.joblib + .json) under version control or a model registry; retain for QA/regulatory traceability.

See also - cli.md - metrics/metrics_and_evaluation/

• oil_auth_tutorial.md