GCP Generative AI Leader Practice Question 52: ML Lifecycle Ingestion
Google Cloud Generative AI Leader · Test 6 · question 2 of 10
A retailer is scoping a new machine learning initiative. Before any model can be trained, evaluated, or deployed, the team must first pull in raw data from point-of-sale systems, mobile app event streams, and supplier catalogs, and land it somewhere the rest of the pipeline can consume it.
Which stage of the machine learning lifecycle is this, and which Google Cloud tools are typically used to accomplish it?
Show answer and explanation
Answer:
The machine learning lifecycle begins with data ingestion: collecting raw data from source systems (point-of-sale, apps, catalogs) and bringing it into an environment where it can be cleaned, transformed, and eventually used for training. On Google Cloud, that typically means landing batch files in Cloud Storage, streaming events through Pub/Sub, and using Dataflow to move and process that data into an analytics store such as BigQuery. Only after ingestion can the later lifecycle stages (data preparation, training, evaluation, deployment, monitoring) begin, because there is nothing yet to prepare or train on.
- A is wrong because deployment happens near the end of the lifecycle, after a model has been trained and evaluated; the scenario describes work that must happen first. - B is wrong because evaluation measures a trained model's quality against metrics; there is no model yet in this scenario. - D is wrong because feature engineering transforms already-ingested data into model inputs. It depends on ingestion having already happened, not the other way around.
Reference: Dataflow overview