Kaggle Titanic

End-to-end ML competition project for Kaggle Titanic survival prediction from tabular passenger data.

Snapshot

Problem

Given train.csv and test.csv, predict Survived for unseen passengers while maintaining transparent preprocessing and a reproducible submission workflow.

Tech Stack

• Python (notebook and script workflows)
• Jupyter Notebook
• XGBoost / classical ML preprocessing
• GitHub Actions (validation checks)

Repository Layout

• data/: competition train/test datasets
• titanic_survival_NN.ipynb: main notebook (EDA, preprocessing, modeling)
• xgboost.py: script-based model experimentation
• solutions/: generated submission files
• tests/: checks for generated output format/content

Quickstart

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Run notebook:

jupyter notebook titanic_survival_NN.ipynb

Or run script experiment:

python xgboost.py

Reproducible CLI Baseline

Generate a deterministic baseline submission and CV report without opening notebooks:

python scripts/reproducible_baseline.py

Outputs:

• solutions/cli_baseline_submission.csv
• artifacts/cv_report.json

Validation and CI

Local check:

python scripts/reproducible_baseline.py
python -m unittest discover -s tests -p "test_*.py"

CI (.github/workflows/ci.yml) validates Python syntax for xgboost.py and solution-file tests.

Results

• Best score in this repository: 0.78229 (Kaggle public leaderboard).
• Includes notebook-first and script-based experimentation paths.
• Includes automated checks for generated submission files.

Limitations

• Workflow is still notebook-centered for main reproducibility path.
• Hyperparameter search and CV reporting are limited.
• No single CLI command yet to reproduce final submission end-to-end.

Roadmap

• Add reproducible CLI pipeline for submission generation.
• Add cross-validation report and feature-importance artifacts.
• Add pinned environment lockfile for stronger reproducibility.

Contributing

See CONTRIBUTING.md.