Google Professional Machine Learning Engineer Question 366
Select 3Google Cloud PlatformYou are a Machine Learning Engineer at a retail company. You have built a demand forecasting model using historical sales data. Before deploying the model to production, you need to validate both the data and the model. Which of the following steps are essential to ensure proper data and model validation in this scenario?
- A
Check for missing values and outliers in the input data used for training and evaluation.
- B
Ensure the training and validation datasets come from the same time period to minimize differences in data distribution.
- C
Perform feature importance analysis to confirm the model uses relevant features.
- D
Validate the model's predictions against ground truth data not seen during training.
- E
Rely on the training dataset's performance metrics to assess the model's generalization.
Show answer and explanation
Correct answers: A, C, D
Explanation
Validating both the data and the model is critical to ensure reliable predictions in production. Detecting data issues like missing values and outliers prevents fundamental errors during training. Feature importance analysis ensures the model is learning from meaningful patterns, and validating predictions on unseen data evaluates the model's generalization ability. Avoiding data leakage and over-reliance on training metrics are also important to ensure robust validation practices.
- A. Correct.
Correct. Missing values and outliers can lead to inaccurate models. Detecting and handling these issues ensures the data is clean and suitable for training.
- B. Incorrect.
Incorrect. Using datasets from the same time period can cause data leakage and result in an overly optimistic performance estimate.
- C. Correct.
Correct. Feature importance analysis helps confirm that the model is using features that are relevant and reduces the risk of overfitting to irrelevant features.
- D. Correct.
Correct. Validating predictions against ground truth data that was not used during training ensures the model performs well on unseen data, which is crucial for generalization.
- E. Incorrect.
Incorrect. Relying solely on training metrics does not provide insight into how the model will perform on new, unseen data. This can lead to overfitting.