Have fun exploring statistical modeling. Maybe win something. Probably not — and that's okay.

I strongly recommend using pipenv. Once you've created your virtual environment, just do pipenv sync.

You will need a .env file in the root of your directory with the following contents:

GITHUB_TOKEN=<your github PAT>
GITHUB_REPO_PATH=<local path where you cloned the repository>
GITHUB_REMOTE_REPO=LotteryML
GITHUB_OWNER=<Your github username>

# Only required if you use automated commits via cron/serviced.
GIT_SIGNING_KEY=<your github signing key>
GIT_SIGN_COMMITS=true;

Go to the main directory and run python lottery.py --gamedir <GAMEDIRECTORY>

The argument GAMEDIRECTORY is a case-sensitive value of the game that you want to run. For example:

python lottery.py --gamedir NJ_Pick6

If you have been running predictions and want to test the accuracy of your predictions against real game data:

python lottery.py --gamedir=NJ_Cash4Life --accuracy

accuracy_allowance: Minimum model R² accuracy required to accept predictions. Expressed as a decimal (e.g. 0.05 = 5%).

ball_game_range_low: This is the lowest number of the main game

ball_game_range_high: This is the highest number of the main game

mode_allowance: Percentage (in decimal) for how far from the mode you can be. 0.05 (5%) from the mode of the sums.

mean_allowance: Percentage (in decimal) for how far from the mean you can be. 0.05 (5%) from the mean of the sums.

model_save_path: Define the path to save models. You probably shouldn't change this.

game_balls: Number of balls. Index 0 is the date. Take the max range and subtract 1 for the number of balls in play.

game_balls_extra_low: Low range of the Powerball, Mega ball, or another ball that can repeat from the main game.

game_balls_extra_high: High range of the Powerball, Mega ball, or another ball that can repeat from the main game.

test_size: Percentage of testing data. 20% is recommended.

train_size: Percentage of training data. 80% is recommended.

timeframe_in_days: Limits the number of days it looks back. e.g. if the game rules change. Defaults to 60 years.

Error: InconsistentVersionWarning: Trying to unpickle estimator DecisionTreeRegressor from version x.y.z.post1 when using version a.b.c. This might lead to breaking code or invalid results. Use at your own risk.

Solution: Delete the *.joblib files in the models directory and rerun. The files will be regenerated.

• Supports multiple game types
• 90%+ Test coverage
• Per-game configuration
• Generates models based on source data
• Regenerates models if outdated
• Records predictions per day (10 draws)
• Checks accuracy of recorded predictions against real game data

Goal: Move beyond linear models to nonlinear / ensemble models

Planned:

• Add XGBoost builder option
• Add LightGBM builder option
• Add RandomForest builder option
• Config switch to choose model per game

Goal: Capture evolving patterns and trends

Features to add:

• Rolling average frequency (N past draws)
• Rolling ball gaps (windowed)
• Rolling entropy
• Fourier / seasonal terms
• Recent draw "momentum" indicator

Goal: Capture deeper structure

Ideas:

• Draw clustering (k-means, DBSCAN)
• PCA components (latent trends)
• Cross-game learning (multi-game patterns)
• Game change point detection (detect when game rules shift)

Goal: Tune for performance

Planned:

• Integrate Optuna / Hyperopt hyperparameter tuning
• Implement rolling-forward CV
• Add validation/test split options
• Per-ball tuning

Goal: Model dependencies between balls

Ideas:

• Ensemble multiple models per ball
• Meta-predictor across runs
• Correlation modeling (balls are not strictly independent)

Planned:

• Accuracy trend over time (plot)
• Per-ball accuracy heatmaps
• Confidence intervals on predictions
• CLI option to show "top X% most likely balls"

• Support for 2D ball games (Powerball / Mega Millions → main + special balls)
• "Smart wheel" generator to optimize plays for coverage

LotteryML is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0).

You are free to:

✅ Use this code for personal and commercial purposes
✅ Modify the code
✅ Distribute copies

However:

❗ If you modify and run this code as part of a public service (API, web app, hosted platform), you must also release the source code of your modifications under AGPL-3.0.

See LICENSE.md for full terms.

This license was chosen to encourage open collaboration and ensure that improvements remain available to the community.

There is no guarantee that you will win anything and this code is a personal learning experiment. You may lose real money using this script to play the lottery. There is no warranty and the code is provided AS-IS. Play at your own risk, discretion, and cost. If this works and you win, consider yourself extremely lucky.

LotteryML is an open exploration of statistical modeling for inherently random games.**

The lottery is not solvable — the purpose is to explore statistical edges, entropy reduction, and better feature engineering.

Gambling problem? Call 1-800-GAMBLER (https://www.1800gambler.net/)