Skip to content

Troubleshooting

Import Errors

ModuleNotFoundError: No module named 'pydantic'

ModelDoctor strictly relies on Pydantic v2 for data serialization. Ensure it is installed: 

pip install pydantic>=2.0

ImportError: cannot import name 'log_report'

Ensure you are importing from the correct namespace: 

# Correct
from modeldoctor.integrations.mlflow import log_report

Execution Errors

AttributeError: 'EvaluationContext' object has no attribute 'metrics'

This occurs if the internal evaluation context fails to build properly, usually because the provided model has not been fitted, or it lacks predict() methods. Ensure you have called model.fit(X, y) before passing it to md.diagnose().

ValueError: Number of features of the model must match the input.

X_train and X_test must have the exact same shape (number of columns). The model must also have been trained on an identical schema.

Dashboard Issues

TemplateNotFound Error

You are attempting to render the HTML dashboard but the Jinja2 templates are missing. This happens if the library was installed without dashboard support. 

pip install "modeldoctor[dashboard]"

Windows/Linux/macOS Notes

ModelDoctor is OS-agnostic and relies entirely on pure Python paths. However, when using report.dashboard() on Windows Subsystem for Linux (WSL), the auto-open command may fail to find a valid web browser. Use report.save_html() instead.