IPL Crunch '26 Analytics Pipeline

This repository contains the data engineering and analytics pipeline built for the IPL Crunch '26 Data Analytics Challenge.

Our primary goal is to uncover deep, counter-intuitive insights by analyzing official ball-by-ball IPL match data. The pipeline eliminates human bias, engineered custom metrics, and mathematically tests assumptions.

Pipeline Overview

The codebase consists of an automated Python workflow:

1. 01_parse_ipl_data.py:

   • Ingests over 1,200 nested Cricsheet JSON files.
   • Extracts metadata (toss decisions, venues, margins) and ball-by-ball data (runs, extras, wickets).
   • Feature Engineering: Calculates match_phase (Powerplay, Middle, Death) and a proprietary rolling Dot Ball Pressure Index that quantifies the exact sequence of consecutive dot balls faced by a batter.
   • Flattens the nested structures into a highly optimized 290,000+ row Pandas DataFrame, exported as a .parquet file.

2. 02_generate_visuals.py:

   • Reads the optimized Parquet dataset.
   • Generates high-resolution, dark-themed charts using matplotlib and seaborn.
   • Visualizes critical components like Match Phase Run Differentials and Venue-Specific Toss Impact.

3. 03_discover_insights.py:

   • Mathematically flags anomalies using standard statistical tests (scipy.stats).
   • Responsible for discovering the Dubai Trap: A statistical anomaly proving that teams choosing to field at the Dubai International Cricket Stadium severely handicap their win probability (40.7% vs the 53.8% global average).

4. 04_generate_presentation.py:

   • Uses fpdf2 to dynamically compile the insights and visual charts into a sleek PDF presentation.

Setup & Usage

To run the pipeline locally:

# Install dependencies
pip install pandas numpy matplotlib seaborn scipy fastparquet pyarrow fpdf2

# 1. Parse JSON files into Parquet (requires raw json folder)
python scripts/01_parse_ipl_data.py

# 2. Generate visuals
python scripts/02_generate_visuals.py

# 3. Discover statistical anomalies
python scripts/03_discover_insights.py

# 4. Export the final presentation PDF
python scripts/04_generate_presentation.py

Note: Raw data files (raw json/) and massive compiled datasets (*.parquet) are intentionally ignored via .gitignore to keep this repository lightweight.