Google Professional Machine Learning Engineer exam dumps

Google Professional Machine Learning Engineer practice question 499 of 522

Professional Machine Learning Engineer. Professional level, Google Cloud. Free question with the correct answer and a full explanation.

Google Professional Machine Learning Engineer Question 499

Select 3Google Cloud Platform

You have deployed a machine learning model for predicting customer churn. While monitoring the model, you notice significant discrepancies between the feature distributions of the training data and the data being sent to the production model. Which of the following actions should you take to address potential training-serving skew?

  1. A

    Implement feature value monitoring to track discrepancies between training and serving data.

  2. B

    Retrain the model on the production data to ensure it performs better on live inputs.

  3. C

    Validate that the preprocessing logic is consistent between training and serving pipelines.

  4. D

    Adjust the serving pipeline to match the training data distribution exactly.

  5. E

    Use a data validation tool like TensorFlow Data Validation (TFDV) to detect schema or distribution mismatches.

Show answer and explanation

Correct answers: A, C, E

Explanation

Training-serving skew occurs when there are differences between the feature distributions, transformations, or schemas in the training and serving environments. To address this issue, it is essential to monitor feature values, validate preprocessing consistency, and use tools like TensorFlow Data Validation to detect mismatches. These steps help ensure the model performs as expected in production and avoids degradation in prediction quality.

  • A. Correct.

    Correct: Feature value monitoring can help identify and track training-serving skew by comparing input feature distributions over time.

  • B. Incorrect.

    Incorrect: Retraining the model on production data without understanding the source of the skew could propagate the issue or lead to biased results. Addressing the root cause is critical.

  • C. Correct.

    Correct: Ensuring that the preprocessing logic is consistent between training and serving pipelines is a common way to prevent or resolve training-serving skew.

  • D. Incorrect.

    Incorrect: Adjusting the serving pipeline to match the training data distribution exactly is not practical or advisable, as it may distort real-world data and lead to incorrect model predictions.

  • E. Correct.

    Correct: TensorFlow Data Validation (TFDV) or similar tools can help identify schema or feature distribution mismatches that may cause training-serving skew.

Timed practice exam

Take a Google Professional Machine Learning Engineer practice test under exam conditions

60 questions in 120 minutes, drawn from this bank, with a score report and a per-question review when you finish.

Start timed exam