Databricks Machine Learning Professional Question 185
Select 3A retail company wants to implement a machine learning model to forecast the number of sales in their stores for the next hour based on streaming data. They plan to deploy this model to continuously generate predictions using Databricks. Which of the following approaches ensures the predictions are updated in real-time and stored correctly for time-based analysis?
- A
Configure a streaming job in Databricks to process the input data in micro-batches and write predictions to a time-series database.
- B
Use Databricks' AutoML to directly deploy the model as an API and write predictions to a batch data store.
- C
Stream the input data to a Databricks Feature Store, use a model to generate predictions in real-time, and store them in a Delta table with a timestamp.
- D
Deploy the model as a REST API outside Databricks and write predictions to cloud object storage for later analysis.
- E
Implement a Databricks structured streaming job to process input data, generate continuous predictions, and store the results in a Delta Lake table partitioned by time.
Show answer and explanation
Correct answers: A, C, E
Explanation
To support continuous predictions in time-based prediction scenarios, the solution must ensure predictions are updated in real-time and stored in a format that supports time-based querying. Databricks provides tools like structured streaming, Delta tables, and the Feature Store to enable this functionality. The correct answers involve scenarios where real-time prediction updates and time-based storage are implemented effectively.
- A. Correct.
Correct: Configuring a streaming job to process input data in micro-batches and storing predictions in a time-series database ensures the predictions are updated continuously and ready for time-based analysis.
- B. Incorrect.
Incorrect: Using AutoML to deploy the model as an API and writing predictions to a batch data store does not ensure real-time predictions or time-based storage.
- C. Correct.
Correct: Streaming input data to a Feature Store, generating predictions in real-time, and storing them in a Delta table with timestamps ensures both real-time predictions and time-based organization.
- D. Incorrect.
Incorrect: Deploying a model as a REST API outside Databricks and writing predictions to cloud object storage introduces latency and does not inherently support continuous or time-based prediction updates.
- E. Correct.
Correct: Using Databricks structured streaming to process input data and store results in a Delta Lake table partitioned by time ensures the predictions are continuously updated and stored for time-based analysis.