LoLog

🧪 Laborator de predicții

Generează meciuri, construiește profiluri din primele 70% și antrenează un model care prezice restul de 30% — totul în browser.

💡 De ce nu e nevoie de GPU?

  • Logistic regression has 15 parameters; even the deep MLP has ~2.2k — 100k matches of data is just a few MB.
  • One epoch is millions of multiply-adds — milliseconds to seconds in a modern JS engine (see measured times above).
  • At this scale WebGL/WebGPU kernel-dispatch overhead actually loses to plain CPU TypeScript.
  • A real community produces hundreds to thousands of games — full retraining after every match stays under 100ms.