Data drift detection in MLOps - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is data drift in machine learning?

Data drift happens when the data your model sees changes over time compared to the data it was trained on. This can make the model less accurate.

beginner

Name one common method to detect data drift.

One common method is to compare statistical properties like mean or distribution of new data with the training data using tests like the Kolmogorov-Smirnov test.

beginner

Why is data drift detection important in production ML systems?

Detecting data drift helps keep models accurate by alerting when data changes. This allows teams to retrain or update models before performance drops.

beginner

What role does baseline data play in data drift detection?

Baseline data is the original data used to train the model. It serves as a reference to compare new incoming data to find any drift.

intermediate

Give an example of a tool or library used for data drift detection.

Tools like WhyLabs, Alibi Detect, or TensorFlow Data Validation help monitor and detect data drift automatically.

What does data drift affect in a machine learning model?

A Model size

B Model accuracy

C Model training speed

D Model architecture

Data drift changes the input data distribution, which can reduce model accuracy.

Which statistical test is commonly used to detect data drift?

A ANOVA

B T-test

C Kolmogorov-Smirnov test

D Chi-square test

The Kolmogorov-Smirnov test compares distributions to detect drift.

What is the first step in data drift detection?

A Collect baseline data

B Retrain the model

C Deploy the model

D Delete old data

Baseline data is needed to compare new data and detect drift.

Which of these is NOT a sign of data drift?

A Increase in model training time

B Sudden drop in model accuracy

C Change in data distribution

D New feature values outside training range

Training time increase is unrelated to data drift detection.

What action should you take after detecting data drift?

A Change the model architecture

B Ignore it

C Delete the model

D Retrain or update the model

Retraining helps the model adapt to new data patterns.

Explain what data drift is and why it matters in machine learning.

Think about how changing data affects predictions.

Describe a simple approach to detect data drift using statistical methods.

Focus on comparing old and new data.