Pedro Tabacof - How I lost 1000€ betting on CS:GO with machine learning and Python

Learn how one data scientist lost 1000€ betting on CS:GO using ML and Python, and discover essential lessons about model calibration, risk management, and betting markets.

Key takeaways
  • Having a good ML model alone is not enough - you need a complete end-to-end pipeline with proper risk management, backtesting, and financial planning

  • Start small with paper trades and shadow mode to validate your strategy before putting real money on the line. Gradually increase bet sizes only after proving the system works.

  • Model calibration is critical for financial decision making - probabilities need to match actual frequencies. Use techniques like isotonic regression to calibrate if needed.

  • Be very skeptical of perceived “edges” in betting markets. If there’s a discrepancy between your model’s probabilities and bookmaker odds, your model is likely wrong.

  • The Kelly criterion for bet sizing is often too aggressive - consider using half Kelly or fixed bet sizes to reduce risk of ruin.

  • Proper backtesting must be done with out-of-time evaluation on future data, not just random cross-validation. Monitor both ML and financial metrics.

  • Watch for feature degradation over time and model drift. Good test set performance doesn’t guarantee future performance.

  • Overfitting itself isn’t necessarily bad if the model performs well on unseen data. Focus more on removing harmful features.

  • Have clear exit criteria and risk management - don’t rely on “vibes” for when to quit. Set stop losses and validate strategies carefully.

  • For betting specifically, markets are likely rigged against you. Professional betting operations will limit/ban profitable players.