PMLE Automating and Orchestrating ML Pipelines Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Configure the pipeline to run the evaluation component in parallel with training to reduce latency.

  Why it's wrong here

  Running evaluation in parallel with training is not possible because evaluation depends on the trained model. The evaluation component requires the model artifact as input, which is only available after training completes. Parallel execution would cause the evaluation to fail or use stale data. This action also does not address the conditional requirement or retries. The correct approach is to use sequential execution with a condition.

• ✗

  Set `enable_caching=False` on the evaluation component to ensure it always runs after training.

  Why it's wrong here

  Disabling caching on the evaluation component forces it to re-execute every time, but it does not control whether it runs based on accuracy. The evaluation would still run even if accuracy is below threshold, which violates the requirement. Caching is about reusing outputs, not about conditional execution. This action does not address the conditional logic or retry requirements.

• ✓

  Set the `retry` policy on the training component to retry on specific exit codes or exceptions.

  Why this is correct

  Vertex AI Pipelines supports retry policies on individual components. By configuring a retry policy with a maximum retry count and backoff, the training component can automatically retry on transient failures such as resource exhaustion or network timeouts. This satisfies the requirement to support retries for transient errors. The retry policy can be specified using the `retry` argument when defining the component or task, and it applies only to that task.

• ✓

  Use a `dsl.Condition` to wrap the evaluation component and check the accuracy metric against the threshold.

  Why this is correct

  A dsl.Condition allows conditional execution based on a runtime expression. By comparing the accuracy metric output from the training component to the threshold, the evaluation component will only run when the condition is true. This is the standard way to implement conditional steps in Vertex AI Pipelines, ensuring the evaluation runs only when the model meets the accuracy requirement. It also naturally handles the dependency on the training component's success.

• ✗

  Use a `dsl.ExitHandler` to catch failures in the training component and retry it manually.

  Why it's wrong here

  A dsl.ExitHandler is used for cleanup actions, not for retrying failed components. While you could implement custom retry logic inside an exit handler, it is not the intended use and would be complex. Vertex AI Pipelines provides built-in retry policies that are simpler and more reliable. The exit handler would not automatically retry on transient errors; it would only execute after a failure, and you would have to code the retry loop yourself.

Go deeper

Related to this question