Google Professional Machine Learning Engineer Question 82
Select 2Google Cloud PlatformYou are a machine learning engineer tasked with building a demand forecasting model for a retail company using Google Cloud AutoML. The dataset includes features such as sales data, product details, and holidays over the past three years. You must ensure that AutoML can effectively capture seasonal patterns in the data. Which actions should you take when preparing the dataset and configuring AutoML?
- A
Include a timestamp column in the dataset and ensure it follows a consistent date format.
- B
Enable the 'Time Series' option in AutoML Tables during model creation.
- C
Manually engineer features like month, day of week, and holidays before uploading the dataset to AutoML.
- D
Ensure the dataset is split into training, validation, and test sets before uploading to AutoML.
- E
Use a dataset with at least a few hundred rows of historical data for better forecasting accuracy.
Show answer and explanation
Correct answers: A, B
Explanation
Forecasting in AutoML requires appropriate temporal information and configuration. Including a timestamp column ensures that the model understands the time-based nature of the data. Enabling the Time Series option configures AutoML to handle forecasting tasks specifically, leveraging seasonality and trends automatically. AutoML is designed to streamline feature engineering and dataset splitting, so these steps are not necessary. Finally, providing sufficient historical data is important, but a few hundred rows may not be enough for reliable long-term forecasting.
- A. Correct.
Including a timestamp column is essential for AutoML to understand the temporal structure of the data and capture time-dependent patterns.
- B. Correct.
Enabling the 'Time Series' option in AutoML Tables allows the platform to treat the problem as a forecasting task and automatically handle seasonality and trends.
- C. Incorrect.
While feature engineering can sometimes improve models, AutoML is designed to automatically extract relevant features like time-based patterns. Manual engineering is not required in this case.
- D. Incorrect.
AutoML handles dataset splitting internally, so you do not need to split the dataset beforehand. You only need to upload the complete dataset.
- E. Incorrect.
While more data generally helps improve forecasting models, a few hundred rows might not be sufficient to capture long-term seasonal patterns. Ideally, you should provide more data.