How I lost 1000€ betting on CS:GO with machine learning and Python
summary
Claims to have an exploitable edge in CS:GO betting, builds a Python web scraper collecting ~30,000 matches and three months of odds, engineers hundreds of features (including a TrueSkill rating), trains a probabilistic classifier with LightGBM, calibrates probabilities, and backtests a strategy showing ~8–10% ROI, then loses €1000 when deploying with a half Kelly criterion bankroll rule. Pipeline gaps include no paper trading, manual bet execution, weak risk and exit planning, and selection bias (winner’s curse).
AI-written summary · about
chapters (18)
- 0:00ML betting experiment intro
- 1:07Lost 1000€ on CS:GO?
- 2:10Need financial edge in betting
- 3:55Profit equation with ML
- 6:22Model calibration in predictions
- 7:48Betting decisions with odds
- 8:50Kelly Criterion risk management
- 11:38Winner’s curse in betting
- 13:46CS:GO betting risks
- 14:49CS:GO data and features
- 16:58TrueSkill for match predictions
- 18:03LightGBM for predictions
- 19:04Time-series model eval
- 20:50Backtesting ROI strategy
- 22:18Backtest success vs real losses
- 24:225 takeaways from failure
- 25:48Advice on ML betting risks
- 26:29Q&A
related talks
48mMachine learning the hard way -- a story about ponies
23mWhy Fraud Detection, Sports Betting, and Bitcoin Prediction Look Alike
16mLeveraging ML to Create a Reliable Anti-Fraud System
31mRisk Analysis
30mHow to build a stock market money printing machine with AI
45mTop 10 Machine Learning Gotchas
33mYour best Bet: Effortless MLOps with Python Models in
39mMachine Learning in Financial Credit Risk Assessment