在沒有老師的情況下，想要瞭解圍棋盤面每一著手的影響力，對於初學者
來說是困難的。本論文在此利用圍棋勢力分佈圖(ownermap)的變化來分析每一
著手的影響力及屬性。Ownermap 傳統計算方式利用多次模擬所以很花時間，為
了減少計算時間，我們利用深度學習的方式來去預測未來的圍棋勢力分佈圖。
另外，我們也利用分析著手屬性來讓電腦傾向下我們希望的棋風。

It is difficult for a beginner to realize the impact of every move for go without a teacher. We use changing of the ownermap to analyze impact and attribute of each move. The traditional way to compute ownermap is by multiple times of simulation, so it takes a lot of time. In order to reduce the computation time, we use deep learning to predict the ownermap. In addition, we try to use the attribute of moves to let the computer play in a style we specify.

摘要 .............................................................. I
Abstract ........................................................ III
致謝 .............................................................. V
目錄 ............................................................ VII
圖目錄 ........................................................... IX
表目錄 ........................................................... XI
第一章 緒論 ........................................................ 1
1.1 研究背景 ................................................... 1
1.2 研究動機 ................................................... 1
1.3 論文概述 ................................................... 1
第二章 相關文獻探討 ................................................ 3
2.1 本論文主要開源程式介紹 ...................................... 3
2.2 ownermap 相關研究 .......................................... 3
2.3 著手優劣與影響相關研究 ...................................... 4
2.4 棋風相關研究 ............................................... 4
第三章 研究方法 .................................................... 5
3.1 蒐集與篩選圍棋棋譜 ......................................... 5
3.2 產生 ownermap 資料 ........................................ 6
3.3 產生類神經網路訓練與測試資料 ................................ 6
3.4 訓練類神經網路 ............................................. 6
3.5 產生四屬性數值 ............................................. 8
3.5.1 術語說明 ............................................. 8
3.5.2 圍地 ................................................ 9
3.5.3 破地 ................................................ 9
3.5.4 棋子的價值 ........................................... 9
3.5.5 攻擊 ............................................... 10
3.5.6 防守 ............................................... 10
3.6 棋風 ................................................. 11
第四章 實驗結果與分析 .............................................. 13
4.1 MC Ownermap 與 DCNN Ownermap 計算時間比較: ................ 13
4.2 三種網路架構 Loss 比較: ................................... 14
4.3 MC Ownermap 和 DCNN Ownermap Loss 比較: .................. 15
4.4 四屬性展示 ................................................ 17
4.5 棋風展示 .................................................. 18
第五章 結論 ....................................................... 25
參考文獻 .......................................................... 27
附錄 ............................................................. 29
Style Only 棋譜 .................................................. 29
Conditional Style 棋譜 ........................................... 33

[1]AlphaGo Teach: Discover new and creative ways of playing Go. (2017). Retrieved 2018, from AlphaGo Teach: https://alphagoteach.deepmind.com/

[2]Baudiš, P., & Moudřík, J. (2012). On Move Pattern Trends in a Large Go Games Corpus. Computing Research Repository, abs/1209.5251. Retrieved from http://arxiv.org/abs/1209.5251

[3]Bouzy, B. (2006). History and Territory Heuristics for Monte Carlo go. New Mathematics and Natural Computation, 2(2).

[4]BVLC. (2017). GitHub - BVLC/caffe: Caffe: a fast open framework for deep learning. Retrieved from GitHub: https://github.com/BVLC/caffe

[5]Elo rating system - Wikipedia. (2018). Retrieved from Wikipedia: https://en.wikipedia.org/wiki/Elo_rating_system

[6]GoGui download | SourceForge.net. (2017). Retrieved from SourceForge: https://sourceforge.net/projects/gogui/

[7]He, K., Zhang, X., Ren, S., & Sun, J. (2016, 1). Deep Residual Learning for Image Recognition. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), (pp. 770-778).

[8]He, K., Zhang, X., Ren, S., & Sun, J. (2016). Identity Mappings in Deep Residual Networks. Computing Research Repository, abs/1603.05027. Retrieved from http://arxiv.org/abs/1603.05027

[9]Jmoudrik. (2016). GitHub - jmoudrik/deep-go-wrap: Toolkit designed to ease development of your Deep Neural Network models for the game of Go (weiqi, baduk). Retrieved from GitHub: https://github.com/jmoudrik/deep-go-wrap

[10]Kocsis, L., & Szepesvári, C. (2006). Bandit Based Monte-Carlo Planning. European Conference on Machine Learning, (pp. 282-293).

[11]Pasky. (2018). GitHub - pasky/pachi: A fairly strong Go/Baduk/Weiqi playing program. Retrieved from GitHub: https://github.com/pasky/pachi

[12]Rémi, C. (2007). Computing "elo ratings" of move patterns in the game of go. International Computer Games Association Journal, 30(4), pp. 198-208.

[13]Silver, D., Huang, A., Maddison, C., Guez, A., Sifre, L., van den Driessche, G., . . . Hassabis, D. (2016). Mastering the game of Go with deep neural networks and tree search. Nature, 529(7587), pp. 484-489.

[14]Tian, Y., & Zhu, Y. (2015). Better Computer Go Player with Neural Network and Long-term Prediction. Computing Research Repository, abs/1511.06410. Retrieved from http://arxiv.org/abs/1511.06410

[15]Ymgaq. (2017). GitHub - ymgaq/AQ: Computer Go Program. Download: http://github.com/ymgaq/AQ/releases. Retrieved from GitHub: https://github.com/ymgaq/AQ