Predicts serverless function invocation traffic using a naive baseline (Lag-1) vs an AI model (Amazon Chronos).

Results: AI achieves 6.9% lower MAE than baseline (4403.89 vs 4728.18).

Requires: Python 3.9+

# Install dependencies
pip install pandas numpy pyarrow matplotlib scikit-learn torch chronos-forecasting

# Download dataset from Kaggle
# https://www.kaggle.com/datasets/theodoram/azure-2019-public-dataset
# Extract to: data/azure-functions-dataset-2019/

python3 run_pipeline.py

Or use the notebook: notebooks/pipeline_demo.ipynb

Method               | MAE          | Cold Starts 
------------------------------------------------------------
Baseline (Lag-1)     | 4728.18      | 146         
AI (Chronos)         | 4403.89      | 151         

MAE Improvement: 6.9%

• Seeds: np.random.seed(42), torch.manual_seed(42)
• Tested on: macOS 26.1, Python 3.9.6, Apple M1 Pro
• Package versions: pandas 2.2.3, numpy 1.24.4, torch 2.8.0, scikit-learn 1.6.1

submission/
├── README.md
├── run_pipeline.py
├── report_2022195142.pdf
├── notebooks/pipeline_demo.ipynb
└── figures/