readme_content = f"""

Number Predictor using data from Cash 3 (Pick3) Lottery.

Number Predictor (000-999) using Cash3 Lottery data. Started out as a simple model (markov chain) into multi-model framework designed to predict Georgia Cash 3 lottery outcomes using a hybrid of statistical, machine learning, and signal processing techniques.

To run this, must have python installed. With the following dependencies

• pandas numpy statsmodels scikit-learn scipy

Project Goals

• One of these ideas popped up in my head and somehow ended up doing too much. This is done for fun or in the name of it.
• Explore whether short 3-digit lottery outcomes (000–999) follow non-random patterns. - Very annoying part
• Use real historical draw data to:
  • Train and evaluate predictive models.
  • Combine diverse techniques using weighted ensemble logic.
• Output top N numbers with the highest prediction confidence.

---

📂 Data Source

• CSV scraped from Georgia Lottery's public Cash 3 results via a custom API scraper.
• Structure: Date, Draw, Winning_Numbers - using this structure you could try this model on other data sources.
• Cleaned and sorted so that most recent draw appears first. - Pain in the butt

---

Prediction Techniques Implemented (started out as markov chain only, will add the script here for that as well)

Each draw prediction includes the following models:

Markov Chain

• Uses transitions from one number to the next.
• Works well for pattern recurrence.

Frequency Weighted

• Tracks digit frequency in recent draws.
• Prioritizes hot numbers.

ARIMA Trend Model

• Time series forecasting using statsmodels.
• Predicts future based on trend + seasonality.

Hot/Cold Analysis

• Compares recently hot numbers vs cold ones.
• Scores numbers based on deviation from average.

Sum & Parity Analysis

• Evaluates if digits tend to sum to specific ranges.
• Even/Odd (parity) and sum rules applied.

Difference Pattern

• Looks at the difference between digits and draw-to-draw.

Gap Analysis

• Time since last occurrence for each digit or combo.

Modulo Pattern

• Uses modular arithmetic to uncover cyclical behavior.

Entropy Trend

• Scores numbers by how "disordered" or "structured" they appear.

FFT Cycle Detection

• Fast Fourier Transform on digit frequency over time.
• Detects hidden periodic cycles.

Higher-Order Markov

• Second-order transition logic to detect deeper sequential patterns.

Cluster Analysis

• Unsupervised ML clustering of draws using K-means.

Lagged Correlation

• Compares digits vs previous digits with temporal offset.

Machine Learning Predictor

• ML model trained on full feature set of previous draws using RandomForest.

Bayesian Updating

• Updates beliefs about what digits will appear based on past appearances.

---

📂 Some random Screenshots

## Accuracy & Backtesting - Historical simulation from Jan 1 – Jul 25, 2025 - All predictions back-tested on actual outcomes - Accuracy metrics include: - Exact hit - 2-digit match - 1-digit match - Total score - Highest performing standalone model: **Gap Analysis** (Exact: 0.34%)

---

Ensemble Probability System

• Normalizes all models to [0, 1]
• Weights each method based on historical accuracy
• Scores all 000–999 combinations
• Outputs top N = 10 predictions per draw

---

📁 Files

• historical_predictions_2025.csv – All daily predictions (per model)
• historical_accuracy_2025.csv – Summary of model performance
• full_advanced.py – Runs prediction models and ensemble logic
• api scrape this year.py – Pulls latest data from GA Lottery
• README.md – You’re here :)

---

Philosophy

While the lottery is widely considered random, this project operates under the hypothesis that:

• Small number sets (000–999) may reveal non-random tendencies over time.
• External predictors may capture hidden trends or operational quirks in the system.

---

---

Save as a markdown file

readme_path = "/mnt/data/README.md" with open(readme_path, "w") as f: f.write(readme_content)