Home

🏠 KataGo Training Project (KW Series)

Welcome to the KataGo Training Project Wiki.
This project documents the training and evaluation of KataGo neural networks (KW10–KW22 and beyond), led by changcheng967 (Cheng Chang) with training support from Doulet Media.

---

📖 About

The KW series explores the progression of KataGo models trained on 19x19 Go data.
Each generation (KW10–KW22) has been evaluated against the FMGo7 baseline and prior models, with Elo ratings, win rates, and efficiency metrics recorded.

---

📊 Key Sections

• Elo Evaluation Report
• KW10–KW22 Training Report
• Efficiency Analysis
• Results Spreadsheet

---

🔧 Tools & Data

• ELO Estimator Script
• Lizzieyzy (analysis tool)
• KataGo Training

---

🙌 Credits

• Project by: changcheng967 (Cheng Chang)
• Training supported by: Doulet Media
• Data sources:
  • ELO estimator.py
  • results.xlsx
  • lizzieyzy
  • katagotraining.org
  • Other sources used during evaluation and analysis

---

Last updated: October 2, 2025