Low-Code AI — BigQuery ML and AutoML — GCP Professional ML Engineer practice questions

An analytics team has 4 TB of retail transactions in BigQuery and needs a churn classifier within two days. Nobody on the team writes Python. The MOST appropriate first approach is to:

1. Export the table to Cloud Storage and train a custom scikit-learn container on the managed training service
2. Stand up a Ray cluster and distribute a gradient boosting fit across the transaction partitions
3. Create a logistic regression model in BigQuery ML with SQL over the existing table in place
4. Move the data into a notebook instance and hand-tune an XGBoost model over a sampled extract

Answer: C — Create a logistic regression model in BigQuery ML with SQL over the existing table in place

C) Correct — BigQuery ML trains in place with SQL only, which fits a SQL-only team on data already in BigQuery and removes all export and infrastructure work. A) Exporting to a custom container is the right pattern when you need a bespoke architecture, but it demands Python skills the team lacks and adds days of setup. B) Ray targets distributed Python workloads; it solves a scale problem this team does not have and requires the most engineering. D) Sampling into a notebook still requires Python and discards data that BigQuery could use in full.

After training a BigQuery ML binary classifier on a dataset that is 2% positive, ML.EVALUATE reports accuracy of 0.98. The analyst should FIRST:

1. Accept the model, since 0.98 exceeds any reasonable production threshold
2. Retrain with more trees and deeper splits to push accuracy higher
3. Inspect precision, recall and ROC AUC at the operating threshold instead
4. Rebalance by discarding negatives until the classes are even, then re-evaluate

Answer: C — Inspect precision, recall and ROC AUC at the operating threshold instead

C) Correct — on a 2% positive base rate, predicting the majority class always yields 0.98, so accuracy carries no information; threshold-aware metrics reveal whether the model finds positives at all. A) Accepting it treats a degenerate baseline as a success. B) More trees optimizes a metric that is already uninformative. D) Downsampling negatives is a legitimate technique but it is premature — you first need to know whether a real problem exists, and undersampling distorts the calibration of predicted probabilities.

BigQuery ML is BEST suited for:

1. Computer vision over labeled image datasets at scale
2. Modeling data already resident in BigQuery using SQL
3. Reinforcement learning with an environment feedback loop
4. Pre-training large language models from scratch on text

Answer: B — Modeling data already resident in BigQuery using SQL

A) Image work belongs in AutoML or custom training, not SQL. B) Correct — BigQuery ML runs regression, classification, clustering, time series and boosted trees through CREATE MODEL with no data movement. C) Not part of the BigQuery ML model catalog. D) Remote models can call an LLM, but BigQuery ML does not pre-train one.

A SQL-only analyst wants k-means clustering over a customer table. Best approach?

1. BigQuery ML CREATE MODEL with model_type='KMEANS'
2. Custom training running scikit-learn KMeans code
3. The AutoML image classification training objective
4. A Dataproc cluster running Spark MLlib KMeans

Answer: A — BigQuery ML CREATE MODEL with model_type='KMEANS'

A) Correct — K-means is a native BigQuery ML model type, trained and scored with SQL over the customer table. B) Requires Python the analyst does not write. C) Wrong modality entirely. D) A cluster is far more machinery than a single SQL statement needs.

A team uses BigQuery ML CREATE MODEL but wants to call PaLM 2 for text summarization. Correct construct?

1. ML.GENERATE_TEXT over a remote model wrapping the LLM
2. CREATE MODEL with model_type='LINEAR_REG' over the text
3. A Cloud Function that calls the LLM API row by row
4. Not possible; BigQuery ML cannot reach an LLM endpoint

Answer: A — ML.GENERATE_TEXT over a remote model wrapping the LLM

A) Correct — a remote model registers the hosted LLM endpoint inside BigQuery, and ML.GENERATE_TEXT then invokes it from SQL over a table. B) A numeric regression type cannot summarize text. C) Workable but leaves SQL entirely and loses set-based processing. D) Remote models exist precisely for this.

BigQuery for ML?

Data warehouse — store + query petabyte-scale data. BigQuery ML lets you train models with SQL.

When is managed AutoML the right choice over writing custom training code?

When the team needs a high-quality model on tabular, image, text or video data without ML engineering effort, and the value is in the result rather than in controlling the architecture. AutoML handles architecture search, feature engineering and tuning, at the cost of little control over the model internals.

BigQuery ML CREATE MODEL syntax basics?

CREATE MODEL `project.dataset.model` OPTIONS (model_type='LOGISTIC_REG') AS SELECT … . Use ML.PREDICT() to predict.

State the decision rule for choosing among BigQuery ML, AutoML and custom training.

BigQuery ML when the data already lives in BigQuery and the team works in SQL • AutoML when higher accuracy is needed on standard data types without ML code • custom training when you need a specific architecture, custom loss, unusual data or full control of the training loop. Move up the ladder only when the simpler option fails a requirement.