Pipeline components and DAGs in MLOps - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding DAG structure in ML pipelines

In an ML pipeline, a Directed Acyclic Graph (DAG) defines the order of tasks. Which statement best describes a DAG?

A A graph where tasks are connected with no cycles, ensuring a clear execution order.

B A graph where tasks can loop back to previous tasks forming cycles.

C A graph that allows tasks to run in any order without dependencies.

D A graph that only contains a single task with no connections.

💡 Hint

Think about what 'acyclic' means in the graph context.

A DAG is a graph with directed edges and no cycles, so tasks run in a defined order without loops.

💻 Command Output

intermediate

Output of a pipeline task execution command

Given the following command to run an ML pipeline task:

mlflow run . -P entry_point=train_model

What is the expected output if the task runs successfully?

A Run completed successfully with metrics logged and model saved.

B ERROR: Entry point 'train_model' not found in the project.

C SyntaxError: invalid syntax in command line.

D Task skipped due to unmet dependencies.

💡 Hint

Consider what happens when a valid entry point is executed in MLflow.

If the entry point exists and runs without error, MLflow logs metrics and saves the model, indicating success.

❓ Configuration

advanced

Correct DAG YAML configuration snippet

Which YAML snippet correctly defines a DAG with three tasks where preprocess runs before train, and train runs before evaluate?

A

tasks:
  preprocess:
    depends_on: [train, evaluate]
  train:
    depends_on: []
  evaluate:
    depends_on: [preprocess]

B

tasks:
  preprocess:
    depends_on: [train]
  train:
    depends_on: [evaluate]
  evaluate:
    depends_on: []

C

tasks:
  preprocess:
    depends_on: [evaluate]
  train:
    depends_on: []
  evaluate:
    depends_on: [train]

D

tasks:
  preprocess:
    depends_on: []
  train:
    depends_on: [preprocess]
  evaluate:
    depends_on: [train]

💡 Hint

Check the order of dependencies to ensure no cycles and correct sequence.

Option D correctly sets dependencies so preprocess runs first, then train, then evaluate.

❓ Troubleshoot

advanced

Identifying cause of pipeline task failure

You run a pipeline and the evaluate task fails with the error: FileNotFoundError: model.pkl not found. Which is the most likely cause?

A The <code>evaluate</code> task has a syntax error in its script.

B The <code>train</code> task did not run or failed to save the model file.

C The pipeline DAG has a cycle causing infinite loops.

D The <code>preprocess</code> task deleted the model file by mistake.

💡 Hint

Think about which task creates the model file and what happens if it is missing.

The evaluate task depends on the model file created by train. If train fails or is skipped, the file won't exist.

🔀 Workflow

expert

Determining pipeline execution order from DAG

Given the following DAG dependencies:

tasks:
  data_ingest:
    depends_on: []
  feature_engineering:
    depends_on: [data_ingest]
  model_training:
    depends_on: [feature_engineering]
  model_validation:
    depends_on: [model_training]
  deployment:
    depends_on: [model_validation, feature_engineering]

What is the correct order of task execution?

A data_ingest → feature_engineering → model_training → deployment → model_validation

B data_ingest → feature_engineering → deployment → model_training → model_validation

C data_ingest → feature_engineering → model_training → model_validation → deployment

D data_ingest → model_training → feature_engineering → model_validation → deployment

💡 Hint

Remember that a task can only run after all its dependencies have completed.

Deployment depends on both model_validation and feature_engineering, so it runs last after those complete.