Project Documentation¶
This site provides project documentation. Use the documentation navigation to explore.
How-To Guide¶
Many instructions are common to all our projects.
See ⭐ Workflow: Apply Example to get the example projects running on your machine.
Project Documentation Pages (docs/)¶
- Home - this documentation landing page
- Project Instructions - the standard project workflow
- Your Files - how to copy the example and create your version
- Glossary - project terms and concepts
- API - autogenerated code documentation for the public project interface
Phase 4. Technical Modification¶
For my technical modification, I expanded the bill length feature sweep used to investigate the deployed penguin prediction model.
Originally, the notebook evaluated bill lengths from 30 mm to 60 mm using 20 evenly spaced values. I modified the code to evaluate bill lengths from 28 mm to 65 mm using 30 evenly spaced values.
I chose this modification because it allowed me to observe the prediction boundary over a wider range of values and determine whether the model continued making consistent predictions outside the original testing range.
After rerunning the notebook, the updated scatter plot showed additional prediction points while maintaining a clear transition between Adelie and Chinstrap penguins. The notebook executed successfully, all visualizations were updated, and the modified notebook was committed and pushed to GitHub.
Compared with the original example, my version explores a larger feature range, providing a better understanding of the deployed model's prediction behavior.
Phase 5. Custom Project¶
Basis and API¶
The project investigates a deployed machine learning model that predicts penguin species using physical measurements.
The API endpoint is:
https://ml-penguin-predictor.onrender.com/predict
The model accepts several numerical measurements, including bill length, bill depth, flipper length, and body mass, then predicts the penguin species.
I kept the same deployed API because the goal of this module was to understand and investigate the model rather than replace it.
Investigation Approach¶
My investigation focused on understanding how changing bill length affected the predicted species.
I expanded the feature sweep from 30–60 mm to 28–65 mm and increased the number of test values from 20 to 30.
I also reviewed edge cases provided by the notebook, including missing values, negative values, zero body mass, and extremely large bill lengths.
Findings: Feature Sensitivity¶
The results showed that bill length strongly influenced the predicted species.
Predictions remained Adelie for smaller bill lengths and changed to Chinstrap once the bill length reached approximately 43 mm.
Expanding the sweep confirmed that larger bill lengths continued producing Chinstrap predictions.
The decision boundary appeared stable and consistent across the wider testing range.
Findings: Edge Cases¶
The deployed API responded differently depending on the input.
Missing required values generated a 400 Bad Request error.
Extreme values still returned predictions, even when the measurements were unrealistic.
Negative measurements and zero body mass were accepted by the API and still produced predictions, suggesting that additional input validation could improve robustness.
Summary¶
This investigation showed that bill length has a strong influence on species prediction.
The deployed model behaved consistently across a wider range of inputs and maintained a clear prediction boundary.
However, the API would benefit from stronger validation to reject impossible measurements before generating predictions.
This type of investigation could be applied to many deployed machine learning APIs to understand prediction boundaries, identify weaknesses, and improve model reliability.
Basis and API¶
Describe the deployed model and API you started with.
Include:
- The example model and what it predicts
- The API endpoint and what inputs it expects
- Why you chose to keep or change the endpoint or model
Investigation Approach¶
Describe how you investigated the model's behavior.
Include:
- Which features you varied and why
- How you structured your tests (single feature, grid, edge cases)
- What you were trying to learn about the model
Findings: Feature Sensitivity¶
Describe what you observed when varying individual features.
Include:
- Which features had the most influence on predictions
- Where the decision boundary appeared to shift
- Any surprising or counterintuitive results
Findings: Edge Cases¶
Describe what happened with unusual or invalid inputs.
Include:
- What edge cases you tested
- How the API responded (prediction, error, or unexpected behavior)
- What this tells you about the model's robustness
Summary¶
Summarize your custom investigation.
Include:
- What you learned about the model's behavior
- Where it appears confident and where it seems fragile
- What you would change about the API contract or model
- What kinds of real problems this approach could apply to
Display at least one chart or screenshot showing your findings.