Data drift detection in MLOps - Mini Project: Build & Apply

Data Drift Detection

📖 Scenario: You work as a data engineer in a company that uses machine learning models to predict customer behavior. Over time, the data your model sees can change, which might make the model less accurate. This change is called data drift. Detecting data drift early helps keep the model reliable.

🎯 Goal: Build a simple Python program that detects data drift by comparing the distribution of a feature in new data against the original training data.

📋 What You'll Learn

Create a dictionary called training_data with feature values

Create a dictionary called new_data with feature values

Calculate the mean of the feature in both datasets

Set a threshold called drift_threshold to detect drift

Compare the means and print if data drift is detected or not

💡 Why This Matters

🌍 Real World

Data drift detection helps keep machine learning models accurate by alerting when input data changes significantly.

💼 Career

Data engineers and MLOps specialists use data drift detection to maintain and monitor deployed ML models in production.

1

Create training data dictionary

Create a dictionary called training_data with the key 'feature1' and the list of values [10, 12, 11, 13, 12].

MLOps

# Create the training_data dictionary with feature1 values
# Your code here

Hint

Use curly braces to create a dictionary and square brackets for the list of numbers.

2

Create new data dictionary

Create a dictionary called new_data with the key 'feature1' and the list of values [14, 15, 13, 16, 15].

MLOps

training_data = {'feature1': [10, 12, 11, 13, 12]}
# Create the new_data dictionary with feature1 values
# Your code here

Hint

Follow the same format as the training_data dictionary.

3

Calculate means and set threshold

Calculate the mean of feature1 in training_data and store it in mean_training. Calculate the mean of feature1 in new_data and store it in mean_new. Then, create a variable called drift_threshold and set it to 2.0.

MLOps

training_data = {'feature1': [10, 12, 11, 13, 12]}
new_data = {'feature1': [14, 15, 13, 16, 15]}
# Calculate mean_training, mean_new and set drift_threshold
# Your code here

Hint

Use sum() and len() functions to calculate the mean.

4

Detect and print data drift

Write an if statement to check if the absolute difference between mean_new and mean_training is greater than drift_threshold. If yes, print "Data drift detected". Otherwise, print "No data drift detected".

MLOps

training_data = {'feature1': [10, 12, 11, 13, 12]}
new_data = {'feature1': [14, 15, 13, 16, 15]}
mean_training = sum(training_data['feature1']) / len(training_data['feature1'])
mean_new = sum(new_data['feature1']) / len(new_data['feature1'])
drift_threshold = 2.0
# Check for data drift and print the result
# Your code here

Hint

Use abs() to get the absolute difference.

Practice

1. What is the main purpose of data drift detection in MLOps?

easy

A. To reduce the size of the dataset

B. To check if new data differs significantly from the training data

C. To improve the speed of model training

D. To increase the number of features in the model

Solution

1. Step 1: Understand data drift concept

   Data drift means the new data changes compared to the data used to train the model.

2. Step 2: Identify the purpose of detection

   Detecting data drift helps decide when to retrain or update the model to keep it accurate.

3. Final Answer:

   To check if new data differs significantly from the training data -> Option B

4. Quick Check:

   Data drift detection = check data difference [OK]

Hint: Data drift means new data changes from old data [OK]

Common Mistakes:

• Confusing data drift with model speed optimization
• Thinking data drift reduces dataset size
• Assuming data drift adds features

2. Which Python library is commonly used for detecting data drift in MLOps?

easy

A. Flask

B. NumPy

C. Matplotlib

D. Evidently

Solution

1. Step 1: Recall common MLOps tools

   Evidently is a popular tool designed specifically for monitoring data and model drift.

2. Step 2: Differentiate from other libraries

   NumPy is for math, Matplotlib for plotting, Flask for web apps, not for drift detection.

3. Final Answer:

   Evidently -> Option D

4. Quick Check:

   Evidently = data drift detection tool [OK]

Hint: Evidently is made for data drift detection [OK]

Common Mistakes:

• Choosing NumPy or Matplotlib which are not for drift detection
• Confusing Flask as a data tool

3. Given the code snippet using Evidently, what will report.run(reference_data, current_data) do?

medium

A. Visualize the model architecture

B. Train a new model on current_data

C. Compare current_data with reference_data to detect data drift

D. Delete old data from the system

Solution

1. Step 1: Understand Evidently report usage

   The run method compares new data (current_data) against reference data to find differences.

2. Step 2: Identify the purpose of the method

   It does not train models, visualize architecture, or delete data; it detects data drift.

3. Final Answer:

   Compare current_data with reference_data to detect data drift -> Option C

4. Quick Check:

   report.run compares data for drift [OK]

Hint: report.run compares new vs reference data [OK]

Common Mistakes:

• Thinking it trains a model
• Assuming it visualizes model structure
• Believing it deletes data

4. You wrote this code to detect data drift but get an error:

from evidently.dashboard import Dashboard
dashboard = Dashboard(tabs=["data_drift"])
dashboard.run(current_data)

What is the likely mistake?

medium

A. Missing reference data argument in dashboard.run()

B. Incorrect import statement for Dashboard

C. Dashboard does not support data_drift tab

D. current_data is not a valid variable name

Solution

1. Step 1: Check Dashboard.run() method requirements

   Dashboard.run() requires both reference and current datasets to compare for drift.

2. Step 2: Identify missing argument

   Only current_data is passed; reference_data is missing, causing the error.

3. Final Answer:

   Missing reference data argument in dashboard.run() -> Option A

4. Quick Check:

   Dashboard.run needs reference and current data [OK]

Hint: Dashboard.run needs two datasets: reference and current [OK]

Common Mistakes:

• Assuming import is wrong
• Thinking data_drift tab is unsupported
• Believing variable name causes error

5. You want to automate retraining your model when data drift is detected. Which approach best fits this goal?

hard

A. Set up a monitoring pipeline that runs data drift detection daily and triggers retraining if drift is found

B. Retrain the model every week regardless of data changes

C. Manually check data drift reports and retrain when you have time

D. Ignore data drift and only retrain when model accuracy drops

Solution

1. Step 1: Understand automation in MLOps

   Automating retraining based on data drift ensures the model stays accurate without manual checks.

2. Step 2: Identify best practice

   Running daily drift detection and triggering retraining only when drift occurs is efficient and effective.

3. Final Answer:

   Set up a monitoring pipeline that runs data drift detection daily and triggers retraining if drift is found -> Option A

4. Quick Check:

   Automate retrain on drift detection = best practice [OK]

Hint: Automate retrain triggered by drift detection [OK]

Common Mistakes:

• Retraining blindly without checking data
• Relying on manual checks only
• Ignoring drift until accuracy drops