Cash Pot — Lottery prediction (ML demo)

Flask web app for exploring historical Cash Pot–style draw data with Random Forest and LSTM (TensorFlow) models, plus frequency/pattern heuristics. Draw scheduling follows the Jamaica timetable (six daily draws, excluding Christmas and Good Friday).

Disclaimer: This project is for learning and experimentation only. Past draws do not predict future outcomes. Do not use this as financial or gambling advice.

Features

• Dashboard — Stats, recent predictions, accuracy summary, optional startup training banner, model health / monitor hooks
• Predictions — Multiple methods (e.g. LSTM, Random Forest, frequency, patterns, ensemble/auto with governance hooks)
• Data — CSV upload, single-draw entry, edit/delete draws
• History — Paginated draw history and prediction history; relink predictions to draws
• Retraining — Async /retrain with /api/training_status for UI feedback
• Backtesting & benchmarks — Walk-forward backtest, multi-seed evaluation, benchmark baseline/history, profile presets (quick / standard / deep)
• APIs — JSON endpoints for stats, draws, backtest, benchmark evaluate/history, monitor health, model/training status

Tech stack

Layer	Choice
Web	Flask 2.x
Data	SQLite (via DatabaseManager in models.py)
ML	scikit-learn (Random Forest), TensorFlow 2.13 (LSTM softmax + calibration)
Deploy-friendly	gunicorn in requirements.txt

Requirements

• Python 3.10+ recommended (TensorFlow 2.13 is picky on version; use a version that matches your platform)
• System dependencies as required by TensorFlow on your OS

Setup

cd "Gambling App"   # or your clone path
python3 -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r requirements.txt

Configuration

Variable	Default	Purpose
ENABLE_STARTUP_TRAINING	1	If 0, skip training models when the app starts (faster dev; use Retrain in UI/API instead)

Production: Set a real Flask secret_key in app.py or via env/config — do not ship the placeholder.

Database & models

• SQLite file: lottery_data.db (created when you first run the app). Listed in .gitignore — not committed by default.
• Trained artifacts may live under trained_models/ (add to .gitignore if you do not want them in Git).

Run locally

source .venv/bin/activate
python3 app.py

Open the URL shown in the terminal (often http://127.0.0.1:5000).

Production-style (example):

gunicorn -w 2 -b 0.0.0.0:8080 app:app

Project layout (high level)

Path	Role
app.py	Routes, training thread state, benchmark profiles
models.py	SQLite schema, draws, predictions, benchmarks
predictor.py	LotteryPredictor — methods, ensemble, backtests
ml_trainer.py	LSTM training/inference, RF helper
schedule.py	Jamaica draw times & blackout dates
templates/	HTML pages
static/	CSS/JS

API quick reference

Method	Path	Notes
GET	/	Dashboard
POST	/predict	JSON prediction
GET/POST	/upload, /add_single, …	Data management
GET	/history, /prediction-history	Paginated views
POST	/retrain	Trigger retrain (background)
GET	/api/training_status, /api/model_status	Training / model state
GET	/api/backtest, /api/backtest_multi	Backtest JSON
GET/POST	/api/benchmark/*	Baseline, evaluate, history
GET	/api/monitor/health	Lightweight health JSON

License

Specify your license here (e.g. MIT) if you want the repo to be open source.

Author

loukieluke — MachineLearning