Evidently AI for monitoring in MLOps - Commands & Configuration

Introduction

When you build machine learning models, you want to make sure they keep working well over time. Evidently AI helps you watch your models and data to catch problems early, like if the data changes or the model starts making mistakes.

When you want to check if your model's predictions are still accurate after deployment.

When you need to detect if the input data to your model has changed unexpectedly.

When you want to generate easy-to-understand reports about your model's performance.

When you want to automate alerts if your model quality drops.

When you want to compare your model's current behavior to past behavior to spot issues.

Commands

This command installs the Evidently AI Python package so you can use it to monitor your models.

Terminal

pip install evidently

Expected Output

Collecting evidently

  Downloading evidently-0.2.38-py3-none-any.whl (123 kB)

Installing collected packages: evidently

Successfully installed evidently-0.2.38

This runs a Python script that loads your model and data, then uses Evidently to create a monitoring report.

Terminal

python monitor_model.py

Expected Output

Report saved to model_monitoring_report.html

Key Concept

If you remember nothing else from this pattern, remember: Evidently AI helps you automatically check your model's health by comparing current data and predictions to past behavior.

Code Example

MLOps

from evidently.dashboard import Dashboard
from evidently.dashboard.tabs import DataDriftTab, ClassificationPerformanceTab
import pandas as pd

# Load reference data (past data) and current data
reference_data = pd.read_csv('reference_data.csv')
current_data = pd.read_csv('current_data.csv')

# Create Evidently dashboard with data drift and performance tabs
dashboard = Dashboard(tabs=[DataDriftTab(), ClassificationPerformanceTab()])

# Calculate and save the report
dashboard.calculate(reference_data=reference_data, current_data=current_data)
dashboard.save('model_monitoring_report.html')
print('Report saved to model_monitoring_report.html')

Common Mistakes

Not installing the Evidently package before running the script

The script will fail because Evidently functions are not available without the package installed.

Always run 'pip install evidently' before using Evidently in your code.

Using Evidently without providing reference data for comparison

Evidently needs baseline data to detect changes; without it, monitoring reports will be incomplete or invalid.

Provide a clean reference dataset representing normal model behavior when creating reports.

Summary

Install Evidently AI Python package to enable model monitoring.

Use Evidently to compare current model data and predictions to reference data.

Generate easy-to-read reports that show data drift and model performance changes.

Practice

1. What is the main purpose of Evidently AI in ML model monitoring?

easy

A. To clean and preprocess raw data before training

B. To train new machine learning models automatically

C. To deploy ML models to production environments

D. To compare old and new data or predictions to detect changes

Solution

1. Step 1: Understand Evidently AI's role

   Evidently AI is designed to monitor ML models by checking if data or predictions have changed over time.

2. Step 2: Identify the main function

   It compares old and new data or predictions to detect data drift or performance issues.

3. Final Answer:

   To compare old and new data or predictions to detect changes -> Option D

4. Quick Check:

   Monitoring = Comparing data changes [OK]

Hint: Evidently AI checks data changes, not training or deployment [OK]

Common Mistakes:

• Confusing monitoring with training models
• Thinking Evidently deploys models
• Assuming it preprocesses data

2. Which of the following is the correct way to create an Evidently dashboard with tabs for data drift and performance?

easy

A. dashboard = Dashboard(tabs=DataDriftTab(), PerformanceTab())

B. dashboard = Dashboard(tabs=[DataDriftTab(), PerformanceTab()])

C. dashboard = Dashboard(tabs=[PerformanceTab, DataDriftTab])

D. dashboard = Dashboard([DataDriftTab(), PerformanceTab()])

Solution

1. Step 1: Review Evidently dashboard syntax

   The Dashboard class expects a list of tab instances passed as the tabs parameter.

2. Step 2: Check correct instantiation

   Tabs must be instantiated with parentheses and passed as a list to the tabs argument.

3. Final Answer:

   dashboard = Dashboard(tabs=[DataDriftTab(), PerformanceTab()]) -> Option B

4. Quick Check:

   Tabs list with instances = dashboard = Dashboard(tabs=[DataDriftTab(), PerformanceTab()]) [OK]

Hint: Tabs must be instances inside a list assigned to tabs parameter [OK]

Common Mistakes:

• Passing classes instead of instances
• Not using a list for tabs
• Incorrect argument syntax

3. Given the following code snippet, what will be the output type of report.save_html('report.html')?

from evidently.dashboard import Dashboard
from evidently.tabs import DataDriftTab

dashboard = Dashboard(tabs=[DataDriftTab()])
report = dashboard.calculate(reference_data, current_data)
report.save_html('report.html')

medium

A. A new HTML file named 'report.html' is created with the dashboard report

B. An error because save_html() returns a string, not a file

C. The report is printed to the console instead of saved

D. Nothing happens because save_html() is not a valid method

Solution

1. Step 1: Understand save_html() method

   The save_html() method saves the dashboard report as an HTML file to the given path.

2. Step 2: Analyze the code behavior

   Calling report.save_html('report.html') creates a file named 'report.html' containing the report content.

3. Final Answer:

   A new HTML file named 'report.html' is created with the dashboard report -> Option A

4. Quick Check:

   save_html() saves file = A new HTML file named 'report.html' is created with the dashboard report [OK]

Hint: save_html() writes an HTML file, does not print or error [OK]

Common Mistakes:

• Thinking save_html() returns a string
• Expecting console output instead of file
• Assuming save_html() is invalid

4. You wrote this code but get an error: TypeError: 'Dashboard' object is not callable. What is the likely cause?

dashboard = Dashboard(tabs=[DataDriftTab(), PerformanceTab()])
report = dashboard(reference_data, current_data)

medium

A. You forgot to import DataDriftTab

B. Tabs must be strings, not instances

C. You should call dashboard.calculate() instead of dashboard()

D. You need to instantiate PerformanceTab with parameters

Solution

1. Step 1: Identify the error cause

   The error says Dashboard object is not callable, meaning dashboard() is invalid syntax.

2. Step 2: Correct method to generate report

   To get a report, you must call dashboard.calculate(reference_data, current_data), not dashboard().

3. Final Answer:

   You should call dashboard.calculate() instead of dashboard() -> Option C

4. Quick Check:

   Use calculate() method to get report [OK]

Hint: Dashboard object is not callable means missing .calculate() [OK]

Common Mistakes:

• Calling dashboard() directly instead of .calculate()
• Assuming tabs must be strings
• Thinking imports cause this error

5. You want to monitor a model's prediction quality over time using Evidently AI. Which combination of tabs should you include in your dashboard to track data drift and model performance together?

hard

A. DataDriftTab and ClassificationPerformanceTab

B. DataQualityTab and RegressionPerformanceTab

C. DataDriftTab and DataQualityTab

D. RegressionPerformanceTab and ClassificationPerformanceTab

Solution

1. Step 1: Identify tabs for data drift and performance

   DataDriftTab monitors changes in input data distribution. ClassificationPerformanceTab tracks model prediction quality for classification tasks.

2. Step 2: Choose correct combination for monitoring

   To monitor both data drift and model performance for classification, use DataDriftTab and ClassificationPerformanceTab together.

3. Final Answer:

   DataDriftTab and ClassificationPerformanceTab -> Option A

4. Quick Check:

   Data drift + classification performance = DataDriftTab and ClassificationPerformanceTab [OK]

Hint: Match tabs to task: DataDrift + ClassificationPerformance for classification [OK]

Common Mistakes:

• Mixing performance tabs for regression with classification
• Using DataQualityTab instead of DataDriftTab
• Choosing two performance tabs without data drift