Free Google Professional Machine Learning Engineer practice questions

Sample question 1. Loan officers must see, for each individual decision, which input features pushed a score up or down. The model is already registered. What should the team do?

1. Publish the global feature importance chart computed once at training time
2. Deploy with explanation settings and request explanations
3. Run an offline attribution notebook once a week and email the summary
4. Read the prediction logs, where the reasons are stored

Sample question 2. A retraining pipeline deploys every new model automatically. Last month a model with lower recall than the one in production replaced it. The team wants the pipeline to deploy only when a new model beats the current production model on the same held-out data, with no human step. Which two changes meet this requirement? (Choose two.)

1. Add a step that scores the new and production models on the same held-out set and outputs the result.
2. Enable execution caching on the training step so that unchanged runs reuse earlier results.
3. Wrap the deployment step in a dsl.If condition on the comparison output.
4. Run the pipeline weekly instead of daily so that fewer models reach production.
5. Log the recall of each model to Experiments and review the charts at month end.

Sample question 3. A data science team hands a classifier to a product team in another department. The product team needs to know the intended use, known limitations and how performance differs across user groups. Which artifact serves this need best?

1. The training code repository README with installation steps
2. An architecture diagram of the serving infrastructure
3. A spreadsheet of the hyperparameters used in the best run
4. A model card covering intended use, limitations and per-group evaluation