Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/characat0/mlops-fundamentals-homework/llms.txt

Use this file to discover all available pages before exploring further.

Submission is via GitHub pull request. Your grade depends on a passing CI run — implement all TODOs across the three homework stages, get a green checkmark in the Actions tab, then submit the PR URL to your instructor. The CI pipeline runs automatically on every push and checks both linting and tests.
1

Fork the Repository

Go to the course repository on GitHub. Click Fork (top-right corner) → Create fork. This creates your own copy at your-username/mlops-fundamentals-homework.
2

Clone Your Fork

Replace <your-username> with your actual GitHub username.
git clone https://github.com/<your-username>/mlops-fundamentals-homework.git
cd mlops-fundamentals-homework
3

Create a Working Branch

All your work must live on a dedicated solution branch. Use your real name so graders can identify your submission at a glance.
git checkout -b solution/<your-name>
# Example: git checkout -b solution/maria-garcia
4

Set Up the Environment

Install all dependencies and download the Spotify dataset before touching any source files. See the Setup guide for the full walkthrough — it covers creating the virtual environment, Kaggle authentication, and starting the MLflow server.
5

Implement the Tasks

Work through each stage in order following the Implementation Guide below. Each stage feeds the next, so complete them sequentially. Commit your progress regularly so CI runs reflect incremental work:
git add .
git commit -m "feat: implement data pipeline process step"
git push origin solution/<your-name>
6

Open a Pull Request

  1. Go to your fork on GitHub.
  2. Click Compare & pull request (GitHub shows this banner automatically after a push).
  3. Set the base repository to the course repo and base branch to main.
  4. Title your PR: [Homework] <Your Full Name> (e.g., [Homework] Maria Garcia).
  5. Paste your completed Submission Checklist in the PR description.
  6. Click Create pull request.
7

Verify the CI Checkmark

The CI pipeline runs automatically on every push to your PR branch. Navigate to the Actions tab on your fork and confirm the latest run is green. A passing run — linting and all tests — is worth 1 point toward your final grade.

​
PR Requirements

Your pull request must meet all three criteria before graders will review it:
  • Title format: [Homework] <Your Full Name>
  • Base branch: main on the course repo (not your fork)
  • CI must pass: a green checkmark in GitHub Actions is required

​
Implementation Checklist

Complete every item below before opening your PR. Each maps directly to a TODO comment in the source code and carries points in the grading rubric.

​
Stage 1 — Data Pipeline

  • process.py — implement the temporal split: year ≤ 2010 → train.csv, year > 2010 → prod_sim.csv; save both files with to_csv(..., index=False)
  • train.py — encode genre labels with LabelEncoder; scale features with StandardScaler for Logistic Regression (XGBoost skips scaling); loop through models in params.yaml, starting an mlflow.start_run() for each; log hyperparameters with mlflow.log_params(), accuracy with mlflow.log_metric(), and save model artifacts with mlflow.sklearn.log_model() / mlflow.xgboost.log_model()
  • evaluate.py — call client.create_model_version() to register the best run, then client.set_registered_model_alias() to assign the @champion alias

​
Stage 2 — Model Serving

  • app/main.py — SpotifyFeatures — add all 12 audio feature fields with correct types to the Pydantic model
  • app/main.py — GET /health — implement the endpoint returning {"status": "healthy"}
  • app/main.py — log_requests middleware — read the request body, parse as JSON, append a timestamped JSON line to logs/api_requests.jsonl, and reconstruct the request before calling call_next
  • app/main.py — predict_genre() — load the MLflow model from ./models/, extract feature values from the SpotifyFeatures object, run inference, and return a PredictionResponse with genre and confidence
  • Dockerfile — add ARG MLFLOW_TRACKING_URI and the RUN mlflow models download step to pull the @champion model into ./models/ at build time

​
Stage 3 — Drift Monitoring

  • analyze_drift.py — run_ks_analysis() — loop over each feature in features_to_test, run scipy.stats.ks_2samp(train_values, prod_values), flag drift when p_value < 0.05, and populate drift_results["details"][feature] with ks_statistic, p_value, drift_detected, train_mean, and prod_mean; this single function is reused by both batch and online modes
Run pytest and flake8 locally before every push to catch issues before they consume CI minutes. A clean local run almost always means a green checkmark.
flake8 .
pytest data_pipeline/tests -v
pytest model_serving/tests -v

Build docs developers (and LLMs) love

Get started for freeTalk to us