Data Management, Feature Store and Collaboration — GCP Professional ML Engineer practice questions

Which artifact relationship does ML Metadata lineage MOST directly let an auditor answer?

1. Which dataset version and training run produced the model now in production
2. Whether the production model's predictions have drifted in the last 30 days
3. How much each input feature contributed to a specific prediction
4. Which IAM principals invoked the prediction endpoint last quarter

Answer: A — Which dataset version and training run produced the model now in production

A) Correct — lineage links artifacts and executions, so tracing a deployed model back to its dataset and run is its core function. B) Drift is answered by model monitoring, which watches serving distributions rather than provenance. C) Per-prediction attribution comes from feature attribution and explainability tooling. D) Caller identity lives in audit logs, an entirely different system.

Features used in both training and online serving must be:

1. Computed by separate training and serving code paths
2. Stored in the managed feature store for both paths
3. Recomputed from raw source on every prediction request
4. Hardcoded as constants inside the serving application

Answer: B — Stored in the managed feature store for both paths

A) Two code paths drift apart and produce training-serving skew. B) Correct — the managed feature store serves one definition to both offline training reads and online lookups, with point-in-time correctness. C) Recomputation adds serving latency and still risks a different formula. D) Constants cannot track changing entity state.

Cost-attribution across ML teams in one project. Best practice?

1. Read the single project bill and split it evenly across the teams
2. Label jobs, endpoints, and datasets; aggregate the billing export
3. Have each team submit manual invoices for the resources it used
4. Disable shared resources so each team must create its own project

Answer: B — Label jobs, endpoints, and datasets; aggregate the billing export

A) Opaque — an even split hides real usage. B) Correct — label-based chargeback over the billing export is the standard attribution method. C) Manual and error-prone. D) Not viable operationally.

Why share embeddings across teams via Vertex AI Vector Search rather than rebuilding?

1. The service trains the embedding models for you
2. It is the only approximate nearest-neighbor option
3. Embeddings are cheap enough to recompute per team
4. Sharing cuts compute cost and keeps semantics aligned

Answer: D — Sharing cuts compute cost and keeps semantics aligned

D) Correct — one canonical embedding index avoids paying repeatedly to embed the same corpus and keeps similarity behaviour consistent across products. A) The service indexes and serves vectors; it does not train the encoder. B) Other ANN implementations exist, so exclusivity is not the reason. C) Embedding large corpora is a real recurring cost.

What does Experiments record, and what question does it let a team answer?

It records each run's parameters, metrics, artifacts and execution context under a named experiment, so runs are comparable side by side. It answers 'which configuration produced this metric, and can we reproduce it' — the record that turns a notebook result into a defensible model choice.

Contrast the offline store and the online store in a managed feature store.

The offline store holds full feature history and is read in bulk to build training sets and run batch prediction. The online store holds only the latest value per entity and is read at low latency by a single-entity key during online serving. Both are fed from the same feature definitions so training and serving see identical logic.

Vertex AI Data Labeling?

Managed service to annotate images/text/video with quality control.

What is ML lineage, and which three artifact types does a metadata store link?

Lineage is the recorded graph connecting datasets, executions and models, showing which data and which training run produced a given model version. ML Metadata stores artifacts (datasets, models), executions (training or evaluation steps) and contexts (pipeline runs), enabling audit, debugging and reproducibility.