Dashboard

ModelDoctor provides a fully self-contained, interactive HTML dashboard for exploring diagnostic reports.

Launching

If you are working in a Jupyter Notebook or local Python script, the easiest way to view the dashboard is by calling report.dashboard().

import modeldoctor as md

report = md.diagnose(...)
report.dashboard()

This command will automatically write the HTML file to a temporary directory and open it in your default web browser.

Features

The dashboard is split into distinct tabs, allowing you to easily navigate between the high-level Summary, detailed Diagnostic Charts, Prescriptions, and the Model Passport.

Embedded Server

The dashboard is a static HTML file. It does not require a backend Python server, Node.js, or any active compute. You can email it, upload it to an S3 bucket, or attach it to a Jira ticket. All charts, JavaScript, and CSS are embedded directly within the file.

Filtering and Searching

The findings table includes built-in search and severity filtering. You can instantly filter down to CRITICAL issues or search for specific dimensions like Leakage.

HTML Reports

To save the dashboard permanently without opening it:

report.save_html("path/to/report.html")

Troubleshooting

Blank Screen / Missing CSS: Ensure you have installed the [dashboard] extra (pip install modeldoctor[dashboard]). Without Jinja2, the HTML generation will fail or produce incomplete files.
File Not Found: If report.dashboard() fails to open your browser, it may be due to restrictive OS settings regarding temporary files. Use save_html() and open the file manually.