Serving Pinned Reproducible Example Scoring
A 4-step recipe with TensorFlow. We can help you get access.
Recipe · 4 steps
Serving Pinned Reproducible Example Scoring
with TensorFlow · Artificial Intelligence Management
Classify and regress the same examples against one explicitly pinned model version.
- Confirm the pinned version is loaded and AVAILABLE. If the version has been unloaded or garbage collected the run must fail loudly rather than silently fall through to a different version.
- Capture the pinned version's signature metadata as part of the audit record, so the exact model shape behind the scores is retained alongside them.
- Run the examples through the pinned version's classification signature, returning the label and score pairs for each record.
- Run the identical examples through the same pinned version's regression signature, returning the numeric estimate that accompanies each classification.
What you'll need to have ready
- modelName: The name the model was registered under in the ModelServer.
- version: The exact model version to pin the whole run to. Supplied explicitly rather than resolved, so the run can be replayed against the same model.
- classifySignatureName: Optional classification signature to target on the pinned version.
- regressSignatureName: Optional regression signature to target on the pinned version.
- context: Optional context features shared across every supplied example.
- examples: The tf.Example style records to score. The same records are sent to both signatures so the class and the numeric estimate describe the same inputs.
For your developer
Arazzo workflow tensorflow:pinned-example-scoring, 4 operations: getModelVersionStatus → getModelVersionMetadata → classifyModelVersion → regressModelVersion. Specs on APIs.io.