Deploy and Query a Model for Realtime Inference MACHINE-LEARNING-ASSOCIATE Practice Questions (Page 1)

Questions 1–5

1. 1application · medium

   A team has a model that requires a custom container with specific libraries not available in the default Databricks runtime. They need to deploy it for realtime inference. What is the required approach?

   • AConvert the model to a format that does not require the custom libraries.
   • BUse a standard serving endpoint and hope the libraries are available.
   • CInstall the libraries on the cluster and serve the model from the cluster's driver.
   • DCreate a custom container image and use it to build the serving endpoint.
   Reveal Solution Discussion Select an answer first
2. 2foundation · easy

   Which Databricks feature is used to monitor the performance and health of a model serving endpoint?

   • ADelta table version history
   • BCluster event logs
   • CJob run history
   • DServing endpoint metrics and logs
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   A data scientist is querying a model serving endpoint that returns a response in the format `{"predictions": [0.9, 0.1]}`. They need to extract the probability of the positive class for a single input. What is the correct interpretation?

   • AThe response contains predictions for two different inputs.
   • BThe second element `0.1` is the probability of the positive class.
   • CThe response is malformed and should contain a single float value.
   • DThe first element `0.9` is the probability of the positive class.
   Reveal Solution Discussion Select an answer first
5. 4application · medium

   A data scientist is using the Databricks UI to test a model serving endpoint. They have entered a JSON payload and clicked 'Send Request'. The response is `{"predictions": [0.2, 0.8]}`. The model is a binary classifier. What is the predicted class?

   • AThe model is uncertain, and the prediction is not reliable.
   • BClass 0, because the first probability is 0.2.
   • CClass 1, because the second probability is 0.8.
   • DThe prediction is a regression output, not a classification.
   Reveal Solution Discussion Select an answer first
6. 5application · medium

   A team has deployed a model to a serving endpoint. They want to monitor the endpoint's performance and set up alerts for when latency exceeds a threshold. What is the best way to do this?

   • AExport the endpoint logs to a third-party monitoring tool and set up alerts there.
   • BManually check the endpoint's response time periodically.
   • CUse the Databricks SQL to query the endpoint's performance metrics.
   • DUse Databricks' built-in endpoint monitoring and set up alerts on the latency metric.
   Reveal Solution Discussion Select an answer first