Real-World Markov Process Simulation (Divvy GBFS → Python)

This project demonstrates an end-to-end real-world Markov process simulation built entirely in Python, using live operational data from the Divvy (Chicago) bike-share system.
The goal is to show how SAS-based Markov simulations and reporting workflows can be migrated to Python while preserving analytical rigor, reproducibility, and decision-making value.

---

🔍 What This Project Does

1. Collects real-world data from Divvy’s public GBFS (General Bikeshare Feed Specification) API
2. Converts numerical station availability data into discrete states
3. Learns transition probabilities from historical data
4. Builds a Markov transition matrix from real observations
5. Simulates future system behavior using the learned model
6. Exports SAS-like analytical outputs (CSV tables, plots, reports)

This mirrors how Markov models are used in utilities, transportation, logistics, finance, and reliability engineering.

---

🌍 Real-World Motivation

Many organizations still rely on SAS to:

• model system behavior over time
• simulate risk and failure
• support operational and financial forecasting

This project shows how the same logic can be:

• implemented in Python
• automated
• validated using real operational data
• exported to modern analytics tools (Power BI, Excel, dashboards)

---

🧠 Conceptual Overview

Each Divvy bike station is treated as an entity.

At each time snapshot, the station is classified into one of four states based on bike availability:

State	Meaning
EMPTY	No bikes available
LOW	Low availability
MEDIUM	Moderate availability
HIGH	High availability

By observing how stations move between these states over time, we:

• learn transition probabilities
• construct a Markov model
• simulate future system behavior

---

📊 Outputs Generated

After running the pipeline, the following outputs are created:

• transition_counts_real.csv
  → Frequency of state-to-state transitions (PROC FREQ equivalent)

• transition_probs_real.csv
  → Learned Markov transition matrix

• simulated_occupancy.csv
  → State distribution over time (simulation result)

• simulated_occupancy.png
  → Visualization of state evolution

• report.md
  → Auto-generated analytical summary

These outputs are ready for dashboards, forecasting, and decision support.

---

▶️ How to Run

1. Install dependencies

pip install -r requirements.txt

2. Collect real-world data (run multiple times)

python src/collect_divvy.py

3. python src/run_end_to_end.py

python src/run_end_to_end.py