import sklearn
import pandasAppendix: Troubleshooting and Extra Notes
Common Rendering Issues
1. NameError: variable not defined
Cause: Each Quarto chapter runs in a fresh Python session.
Solution: Ensure every chapter is self-contained: - load the dataset - split the data - define preprocessing - define the model
Do not rely on variables from previous chapters.
2. All arrays must be of the same length
Cause: Mixing expanded one-hot feature names with original column names.
Solution: - Built-in tree feature importance → use expanded feature names
- Permutation importance → use original X.columns
These are different levels of representation.
3. ModuleNotFoundError
Cause: Virtual environment not activated.
Solution:
#| label: activate-env
source .venv/bin/activateVerify installation:
Reproducibility Notes
To ensure consistent results:
- Use fixed random_state values
- Keep train/test splits consistent
- Avoid modifying datasets mid-analysis
- Do not evaluate on training data
For publication-quality work, consider:
- Cross-validation
- Multiple random seeds
- Version control for datasets
On Metrics and Interpretation
Accuracy is not always sufficient.
Always ask:
- What error is more costly?
- False positive or false negative?
- Does the metric match the business or research objective?
Metrics should reflect consequences.
On Feature Importance
Feature importance indicates influence on prediction within this dataset.
It does not imply:
- causation
- mechanism
- domain-level truth
Treat feature importance as a diagnostic tool.
Not as a claim.
Scaling Up
If you apply this workflow to real-world data:
Expect additional challenges:
- Missing data
- Severe class imbalance
- High-dimensional features
- Temporal leakage
- Deployment constraints
The structured workflow still applies.
Design → Data → Model → Evaluation → Interpretation
Final Note
Machine learning is powerful when disciplined.
The workflow you learned here is more important than any single algorithm.
Structure first.
Then complexity.