🏀 Deepshot v2.1.0

🏀 Deepshot v2.1.0 – Tuned for the Modern Game

The Deepshot prediction engine just got sharper.
This release focuses on model tuning and optimization, aligning Deepshot’s predictive logic with the latest trends in the NBA.

🔮 Model & Data

• Tuned with 2024–25 season data using Optuna
  The model was fine-tuned on the latest season’s games, leveraging Optuna for automated hyperparameter optimization.
• Smarter training split
  NBA data from 2000–01 → 2023–24 is now used with an 80/20 train–validation split, ensuring robust and unbiased learning.
• Retrained for modern playstyles
  After tuning, the model was retrained on the full historical dataset — now optimized to reflect current basketball dynamics and trends.
• Improved generalization to recent gameplay, capturing evolving offensive and defensive strategies.

🧠 Prediction Engine

• Refined hyperparameters enhance predictive stability and confidence.
• Better adaptation to recent pace, spacing, and player-impact metrics.
• Slightly improved accuracy on 2024–25 matchups during backtesting.

⚙️ Miscellaneous

• Streamlined tuning pipeline for faster retraining and experimentation.
• Minor cleanup and improved reproducibility of results.

---

🚀 Try it now: Deepshot
💬 Have feedback or new feature ideas? Open an issue — let’s make the model even smarter together.

🏀 Deepshot v2.0.0

🏀 Deepshot v2.0.0 – New Season Release (2024–25)

The new NBA season is here — and so is Deepshot’s biggest update yet!
This release brings major improvements across the board, from model accuracy to visuals.

🔮 Model & Data

• Trained on 25 seasons of NBA data (2000–01 → 2024–25)
  The model now leverages two and a half decades of games for deeper historical context and trend awareness.
• Better prediction of underdog wins
  Enhanced handling of 60/40-type matchups — the model is now more confident spotting upset potential.
• Refined preprocessing pipeline for faster retraining and more consistent results.

🧠 Prediction Engine

• Optimized feature weighting for momentum and player impact metrics.
• Improved game context understanding (back-to-back games, rest days, injuries, etc.).
• Cleaner, more interpretable probability outputs.

🎨 UI & Experience

• Updated team logos with 2024–25 season designs.
• Visual tweaks for a smoother, more modern interface.
• Better color contrast and layout polish for both light and dark modes.

⚙️ Miscellaneous

• Code cleanup & modular refactor for easier future updates.
• Minor bug fixes and performance optimizations.

---

🚀 Try it out: Deepshot
💬 Feedback & feature requests are always welcome — open an issue or start a discussion!

🏀 Deepshot v1.1.0

🚀 What's New in v1.1.0

A new milestone for DeepShot: version 1.1.0 refines both the intelligence and experience of NBA game outcome prediction. With model accuracy now reaching 71.16%, this update enhances the underlying prediction engine through data balancing, interactive analytics, and an improved UI. DeepShot continues to empower curious fans, developers, and data scientists alike—with a growing community and a vision for smarter sports insights.

📈 Accuracy & Modeling

• Model Accuracy Boosted to 71.16%
  Thanks to model refinement and better statistical feature handling, predictions are now more reliable than ever.

• SMOTE Oversampling
  Introduced SMOTE (Synthetic Minority Over-sampling Technique) to balance the dataset, especially for away team wins, leading to improved recall and F1 score.

🧠 Interactive Insights

• Detailed Match Analysis Panel
  Each matchup now includes a clickable detail view, letting you compare and plot selected statistics between head-to-head teams in dynamic, interactive graphs.

🌐 UX & Accessibility

• Page-Based Routing
  Navigation is now faster and more intuitive: seamlessly return from match details to the same filtered date, rather than defaulting to today's games.

• Public Web Access (Beta)
  DeepShot is now live online! While performance may vary (free-tier hosting limitations), community support could unlock a premium hosting plan for faster access and enhanced reliability.

☕ Support the Project

• Buy Me a Coffee
  Love DeepShot? Help sustain its development by donating—your support enables access to premium APIs, faster hosting, and a more powerful, secure model deployment in the future.

---

📦 Includes

• Updated model (deepshot_v1.1.pkl)
• Enhanced data pipeline with oversampling
• Refined main.py script
• Interactive NiceGUI-powered dashboard with new routing and visualization features

📘 Documentation

• Updated README with new installation, usage, and contribution instructions
• New section on understanding SMOTE and how it's applied in DeepShot

🙏 Acknowledgements

DeepShot v1.1.0 continues to stand on the shoulders of open-source excellence, with special thanks to:

• Imbalanced-learn – SMOTE and other data balancing techniques
• All previously credited libraries and tools

---

Tag: v1.1.0
Date: May 15, 2025
Release URL: https://github.com/saccofrancesco/deepshot/releases/tag/v1.1.0

🏀 Deepshot v1.0.0

A first official release of DeepShot brings a fully‑featured NBA outcome prediction engine—including data ingestion, model training, and an interactive NiceGUI web UI—packaged for easy cross‑platform installation and use. DeepShot leverages advanced team statistics from Basketball Reference, computes rolling trends via EWMA, and trains a gradient‑boosted model with scikit‑learn and XGBoost to achieve a 67.16% prediction accuracy. The release also includes sample data, a pretrained model, and comprehensive documentation to get you up and running in minutes.

---

Added

• Initial Release of DeepShot, a machine learning model for predicting NBA game outcomes with 67.16% accuracy.
• Data‑Driven Predictions: ingests advanced team statistics directly from Basketball Reference.
• Rolling Statistics via Exponentially Weighted Moving Averages (EWMA) to capture team form over time.
• Weighted Stats Engine built on XGBoost and scikit‑learn for robust model training and evaluation.
• Real‑Time Web Interface powered by NiceGUI, offering a sleek, interactive view of upcoming matchups and prediction results.
• Key Stats Highlighting that automatically emphasizes the most significant statistical differences between teams.
• Cross‑Platform Support: runs seamlessly on Windows, macOS, and Linux environments.
• Starter Assets: includes a sample dataset, pretrained model file (deepshot.pkl), and an example main.py script for quick experimentation.

Documentation

• Comprehensive README with step‑by‑step setup instructions, usage examples, and acknowledgements.

Acknowledgements

DeepShot wouldn’t be possible without these open‑source libraries and data sources:

• Python – the core language powering DeepShot
• Basketball Reference – advanced team stats provider
• Requests – elegant HTTP library for data fetching
• Beautiful Soup – HTML/XML parsing for web scraping
• Pandas – data analysis and manipulation toolkit
• scikit‑learn – machine learning library
• XGBoost – gradient boosting framework
• NiceGUI – instant web UI for Python apps

---

Tag: v1.0.0
Date: April 18, 2025
Release URL: https://github.com/saccofrancesco/deepshot/releases/tag/v1.0.0