Databricks Machine Learning Professional Question 187
Single answerYou are tasked with building a real-time machine learning model deployment for predicting energy consumption of smart meters over the next hour. The predictions need to be continuously updated as new sensor data arrives. Which of the following approaches best aligns with the requirements for a streaming deployment in this scenario?
- A
Use a batch inference pipeline to periodically predict energy consumption and write results to a database.
- B
Deploy a streaming inference pipeline that updates predictions in real-time using incoming sensor data.
- C
Train the model in real-time using streaming data and deploy it once training is complete.
- D
Use a pre-trained model to generate one-time predictions for each device and store them for future use.
Show answer and explanation
Correct answer: B
Explanation
In a streaming deployment scenario, continuous predictions are required to update outputs as new data arrives. Streaming inference pipelines are designed to handle this by processing incoming data in real-time and updating predictions accordingly. This makes option 2 the correct choice for the given task.
- A. Incorrect.
Batch inference pipelines do not support real-time updates as they process data in fixed intervals, which does not align with the requirement for continuous predictions.
- B. Correct.
Streaming inference pipelines process data as it arrives, making them suitable for real-time, continuous predictions in scenarios like predicting energy consumption from sensor data.
- C. Incorrect.
Training the model in real-time may not be practical for this use case, as the requirement focuses on continuous predictions, not continuous training.
- D. Incorrect.
Generating one-time predictions and storing them does not fulfill the requirement for continuously updated predictions as new data streams in.