Training data pipeline automation in MLOps - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Key benefit of automating training data pipelines

Which of the following is the primary benefit of automating training data pipelines in machine learning projects?

A Removes the requirement for data versioning

B Ensures consistent and repeatable data preprocessing steps

C Automatically improves model accuracy without retraining

D Eliminates the need for model evaluation

💡 Hint

Think about what automation helps with in repetitive tasks.

Automating training data pipelines helps ensure that data preprocessing is consistent and repeatable, which is crucial for reliable model training.

❓ Predict Output

intermediate

Output of data pipeline step code snippet

What is the output of the following Python code simulating a data pipeline step that filters out negative values and scales the rest by 10?

MLOps

data = [-3, 0, 2, 5]
processed = [x * 10 for x in data if x >= 0]
print(processed)

A [0, 20, 50]

B [-30, 0, 20, 50]

C [0, 2, 5]

D [30, 0, 20, 50]

💡 Hint

Look at the condition inside the list comprehension.

The code filters out negative numbers (-3) and multiplies the remaining numbers by 10, resulting in [0, 20, 50].

❓ Model Choice

advanced

Choosing a model for automated retraining trigger

You want to automate retraining of a model when new data distribution shifts significantly. Which model monitoring technique best supports this automation?

A Use a drift detection model that monitors input feature distribution changes

B Use a model that ignores input data changes and retrains on fixed schedule

C Use a model that only monitors training loss during initial training

D Use a model that retrains only when accuracy on training data improves

💡 Hint

Think about how to detect when new data is different from old data.

Drift detection models monitor changes in input data distribution and can trigger retraining automatically when significant shifts occur.

❓ Hyperparameter

advanced

Hyperparameter to optimize for pipeline latency

In an automated training data pipeline, which hyperparameter adjustment can reduce pipeline latency without significantly harming model quality?

A Add more complex feature engineering steps

B Increase number of epochs during model training

C Decrease batch size during data preprocessing

D Increase model depth to improve accuracy

💡 Hint

Smaller batches can speed up processing but may affect stability.

Decreasing batch size can reduce latency in data preprocessing steps by processing smaller chunks faster, helping pipeline speed without large quality loss.

🔧 Debug

expert

Identifying error in automated data pipeline code

What error does the following Python code raise when running an automated data pipeline step that merges two datasets with mismatched keys? ```python df1 = {'id': [1, 2], 'value': [10, 20]} df2 = {'key': [1, 3], 'score': [100, 300]} import pandas as pd df1 = pd.DataFrame(df1) df2 = pd.DataFrame(df2) merged = pd.merge(df1, df2, left_on='id', right_on='key') print(merged) ```

A TypeError: merge() got an unexpected keyword argument 'left_on'

B ValueError: columns overlap but no suffix specified

C No error, prints merged DataFrame

D KeyError: 'id'

💡 Hint

Check the column names used in merge keys.

The right DataFrame has column 'key', not 'id', so specifying right_on='id' causes a KeyError.