From 41def76222b8effbdbac0defce26e01a03c04193 Mon Sep 17 00:00:00 2001 From: Joe Moudrik Date: Tue, 27 Mar 2012 20:08:43 +0200 Subject: [PATCH] gostyle.tex: incorporated the review comments. --- tex/POSUDEK2 | 36 ++++++++++++ tex/gostyle.tex | 177 +++++++++++++++++++++++++++++++++----------------------- 2 files changed, 139 insertions(+), 74 deletions(-) create mode 100644 tex/POSUDEK2 diff --git a/tex/POSUDEK2 b/tex/POSUDEK2 new file mode 100644 index 0000000..51c1da7 --- /dev/null +++ b/tex/POSUDEK2 @@ -0,0 +1,36 @@ +Reply to reviewers +------------------ + + +Reviewer: 1 +------------------ +We have fixed the typos gratefully provided by Reviewer 1. The first version of the paper had unfortunately a formating flaw, resulting in the "bizarre" flaws. We fixed this as soon as we found out and resent the corrected version, but the reviewers did not unfortunately received it on time. We are sorry for this. + + +Reviewer: 2 +------------------ +Thank you. + +Reviewer: 3 +------------------ +> In this paper, the authors collected a large corpus of GO games and analysed it using data mining techniques. The idea is not novel, as it seems that the authors are not aware of the large literature that exists on this topic. For example, you should discuss in details the following work in your paper + +> Ghoneim A., Essam D., and Abbass H.A. (2011) On Computations and Strategies for Real and Artificial Systems, European Conference on Artificial Life, Paris, August 8-12. +> Ghoneim A., Essam D., and Abbass H.A. (2011) Competency Awareness in Strategic Decision Making, IEEE Conference on Cognitive Methods in Situation Awareness and Decision Support, Florida, USA, February, 2011. +> Michael Harré, Terry Bossomaier, Allan Snyder: The Development of Human Expertise in a Complex Environment. Minds and Machines 21(3): 449-464 (2011) + +We were not aware of this research, so we have altered our paper to mention the second conference paper, which seems to be relevant. However, we still consider our work novel, since the solution in the paper 1) uses GnuGo as a blackbox, making it hard to do any pattern analysis, diminishing the Go-theoretical and study usage. 2) does not develop any specialized Go-concerned techniques 3) performs poorly. + +> The analysis of style is interesting. However, the whole paper is taking a pure data analysis approach. The paper has a table after another, with minimum discussion of the theory or the implications of the results. +> On the theory side, the authors need to discuss the skill and competency literature, which represent the foundation for this work. From the implications side, what does this all mean? Putting together the list of discoveries from this work, how will it translate into something useful for playing go? What is the use of this research? + +The tables and precise data are very important to back up the statements about the performance of our framework. Additionaly, precise numerical results allow for replication of our results and their comparison with other methods possibly developed in the future. In this manner our paper tries to honour the rigorous scientific methodology. + +Although our work might be viewed through the skill and competency theory, we do not find these very useful or relevant for Go-analysis. + +We believe that we have made the possible implications and uses for playing Go clear in Section VI (Proposed Applications), and Section VII (Future Work). We are currently preparing online application based on the research, which will help to pinpoint patterns to avoid, games to replay and possible study directions. + + +TODO reference dodat rok, GoDiscThread, KGS Analytics + + diff --git a/tex/gostyle.tex b/tex/gostyle.tex index 7296b58..0d3b5f6 100644 --- a/tex/gostyle.tex +++ b/tex/gostyle.tex @@ -45,9 +45,17 @@ \setgounit{0.4cm} \usepackage{soul} + +%ENABLE: %\newcommand{\rv}[1]{\ul{#1}} +%DISABLE: \newcommand{\rv}[1]{#1} +%ENABLE: +\newcommand{\rvv}[1]{\ul{#1}} +%DISABLE: +%\newcommand{\rvv}[1]{#1} + \usepackage{algorithm} \usepackage{algorithmic} %\usepackage{algpseudocode} @@ -217,7 +225,7 @@ % % paper title % can use linebreaks \\ within to get better formatting as desired -\title{On Move Pattern Trends\\in Large Go Games Corpus} +\title{On Move Pattern Trends\\in \rvv{a} Large Go Games Corpus} % use \thanks{} to gain access to the first footnote area % a separate \thanks must be used for each paragraph as LaTeX2e's \thanks @@ -344,7 +352,7 @@ accessible. % and "HIS" in caps to complete the first word. \IEEEPARstart{T}{he} field of Computer Go usually focuses on the problem of creating a~program to play the game, finding the best move from a~given -board position. \cite{GellySilver2008} +board position \cite{GellySilver2008}. We will make use of one method developed in the course of such research and apply it to the analysis of existing game records with the aim of helping humans to play and understand the game better @@ -359,10 +367,16 @@ Many Go players are eager to play using computers (usually over the internet) and review games played by others on computers as well. This means that large amounts of game records are collected and digitally stored, enabling easy processing of such collections. However, so far -only little has been done with the available data --- we are aware +only little has been done with the available data. We are aware only of uses for simple win/loss statistics \cite{KGSAnalytics} \cite{ProGoR} and ``next move'' statistics on a~specific position \cite{Kombilo} \cite{MoyoGo}. +\rvv{Additionaly, a simple machine learning technique based on GNU Go's}\cite{GnuGo}\rvv{ +move evaluation feature has recently been presented in}\cite{CompAwar}\rvv{. The authors used decision trees +to predict whether a given user belongs into one of three classes based on his strength +(causal, intermediate or advanced player). This method is however limited by the +blackbox-use of GNU Go engine, making it unsuitable for more detailed analysis of the moves.} + We present a~more in-depth approach --- from all played moves, we devise a~compact evaluation of each player. We then explore correlations between evaluations of various players in the light of externally given information. @@ -392,8 +406,8 @@ and point out some possible future research directions (section~}\ref{future-res \section{Data Extraction} \label{pattern-vectors} -As the input of our analysis, we use large collections of game records% -\footnote{We use the SGF format \cite{SGF} in our implementation.} +As the input of our analysis, we use large collections of game records +\rvv{in SGF format} \cite{SGF} \rv{grouped by the primary object of analysis (player name when analyzing style of a particular player, player rank when looking at the effect of rank on data, etc.).} @@ -403,8 +417,7 @@ played move -- a {\em pattern}, being a combination of several We \rv{compute the occurence counts of all encountered patterns, eventually} composing $n$-dimensional {\em pattern vector} -$\vec p$ of counts of the $n$ globally most frequent patterns% -\footnote{We use $n=500$ in our analysis.} +$\vec p$ of counts of the $n$ \rvv{(we use $n = 500$)} globally most frequent patterns (the mapping from patterns to vector elements is common for \rv{all generated vectors}). We can then process and compare just the pattern vectors. @@ -417,8 +430,8 @@ gather too few specimen over the games sample and the vector differences are not statistically significant. We have chosen an intuitive and simple approach inspired by pattern features -used when computing Elo ratings for candidate patterns in Computer Go play. -\cite{PatElo} Each pattern is a~combination of several {\em pattern features} +used when computing Elo ratings for candidate patterns in Computer Go play +\cite{PatElo}. Each pattern is a~combination of several {\em pattern features} (name--value pairs) matched at the position of the played move. We use these features: @@ -440,7 +453,7 @@ black-to-play and maintain translational and rotational symmetry. Configurations of radius between 2 and 9 in the gridcular metric% \footnote{The {\em gridcular} metric $d(x,y) = |\delta x| + |\delta y| + \max(|\delta x|, |\delta y|)$ produces -a circle-like structure on the Go board square grid. \cite{SpatPat} } +a circle-like structure on the Go board square grid \cite{SpatPat}. } are matched. Pattern vectors representing these features contain information on @@ -475,7 +488,7 @@ in the GoGoD games examined in sec. \ref{style-analysis}.} \subsubsection{Linear Normalization} -One is simply to linearly re-scale the values using: +\rvv{An intuitive solution is to} linearly re-scale the values using: $$y_i = {x_i - x_{\rm min} \over x_{\rm max}}$$ This is the default approach; we have used data processed by only this computation unless we note otherwise. @@ -504,10 +517,10 @@ also most important for classification of major style aspects. \subsection{Implementation} We have implemented the data extraction by making use of the pattern -features matching implementation% -\footnote{\rv{We have developed} the pattern features matcher -according to the Elo-rating pattern selection scheme. \cite{PatElo}} -within the Pachi go-playing program \cite{Pachi}. +features matching implementation +within the Pachi Go-playing program \cite{Pachi}, \rvv{which works according to +the Elo-rating pattern selection scheme} \cite{PatElo}. + We extract information on players by converting the SGF game records to GTP stream \cite{GTP} that feeds Pachi's {\tt patternscan} engine, \rv{producing} a~single {\em patternspec} (string representation @@ -520,11 +533,10 @@ only moves played by the appropriate \rv{player} are collected. To assess the properties of gathered pattern vectors and their influence on playing styles, we analyze the data using several basic data minining techniques. - The first two methods {\em (analytic)} rely purely on single data set and serve to show internal structure and correlations within the data set. -Principal Component Analysis \cite{Jolliffe1986} +Principal Component Analysis \rvv{\emph{(PCA)}} \cite{Jolliffe1986} finds orthogonal vector components that \rv{represent} the largest variance \rv{of values within the dataset. That is, PCA will produce vectors representing @@ -535,10 +547,9 @@ determines its impact on the overall dataset variance: $1.0$ would mean that all points within the dataset lie on this vector, value close to zero would mean that removing this dimension would have little effect on the overall shape of the dataset.} -Reversing the process of the PCA% -\footnote{Looking at dependencies of a single orthogonal vector component -in the original vector space.} -can indicate which patterns correlate with each component. + +\rvv{Reversing the process of the PCA by backprojecting the orthogonal vector components into the +original pattern space can indicate which patterns correlate with each component.} Additionally, PCA can be used as vector preprocessing for methods that are negatively sensitive to pattern vector component correlations. @@ -546,8 +557,9 @@ that are negatively sensitive to pattern vector component correlations. spatial representation of the data set elements (e.g. players) based on similarity of their data set features; we can then project other information on the map to illutrate its connection to the data set.% -\footnote{\rv{We also attempted to visualise the player relationships -using Kohonen maps, but that did not produce very useful results.}} +% Pryc v ramci snizeni poctu footnotu +%\footnote{\rv{We also attempted to visualise the player relationships +%using Kohonen maps, but that did not produce very useful results.}} Furthermore, we test several \emph{classification} methods that assign an \emph{output vector} $\vec O$ \rv{to} each pattern vector $\vec P$, @@ -571,7 +583,7 @@ matrix, provided that the PCA dimensions have some already well-defined \rv{interpretation}; this can be true for single-dimensional information like the playing strength. -Aside of that, we test the $k$-Nearest Neighbors \cite{CoverHart1967} classifier +Aside of that, we test the $k$-Nearest Neighbors (\emph{$k$-NN}) classifier \cite{CoverHart1967} that approximates $\vec O$ by composing the output vectors of $k$ reference pattern vectors closest to $\vec P$. @@ -594,7 +606,7 @@ with the particular techniques. \label{pearson} To find correlations within or between extracted data and some prior knowledge (player rank, style vector), we compute the well-known -{\em Pearson product-moment correlation coefficient} \cite{Pearson}, +{\em Pearson product-moment correlation coefficient (PMCC)} \cite{Pearson}, measuring the strength of the linear dependence% \footnote{A desirable property of PMCC is that it is invariant to translations and rescaling of the vectors.} @@ -611,7 +623,7 @@ We then average results over all iterations. \subsection{Principal Component Analysis} \label{PCA} -We use Principal Component Analysis \emph{PCA} +We use Principal Component Analysis to reduce the dimensions of the pattern vectors while preserving as much information as possible, assuming inter-dependencies between pattern vector dimensions are linear. @@ -694,9 +706,9 @@ $$ \delta(a, b, c) = \begin{cases} The $\varphi$ projection is then determined by randomly initializing the position of each subject and then employing gradient descent methods. -\subsection{k-nearest Neighbors Classifier} +\subsection{k-Nearest Neighbors Classifier} \label{knn} -Our goal is to approximate player's output vector $\vec O$, +Our goal is to approximate \rvv{the} player's output vector $\vec O$, knowing their pattern vector $\vec P$. We further assume that similarities in players' pattern vectors uniformly correlate with similarities in players' output vectors. @@ -770,7 +782,7 @@ Parameters control the growth rate $r$ and the x-position $k$.} \subsubsection{Training} Training of the feed-forward neural network usually involves some modification of supervised Backpropagation learning algorithm. -We use first-order optimization algorithm called RPROP. \cite{Riedmiller1993} +We use first-order optimization algorithm called RPROP \cite{Riedmiller1993}. %Because the \emph{reference set} is usually not very large, %we have devised a simple method for its extension. @@ -809,7 +821,7 @@ The training algorithm is shown in Algorithm \ref{alg:tnn}. \subsection{Naive Bayes Classifier} \label{naive-bayes} -Naive Bayes Classifier uses existing information to construct +The Naive Bayes Classifier uses existing information to construct probability model of likelihoods of given {\em feature variables} based on a discrete-valued {\em class variable}. Using the Bayes equation, we can then estimate the probability distribution @@ -871,7 +883,7 @@ The neural network \rv{component} is written using the libfann C library \cite{N The Naive Bayes Classifier \rv{is built around} the {\tt AI::NaiveBayes1} Perl module \cite{NaiveBayes1}. The sociomap has been visualised using the Team Profile Analyzer \cite{TPA} -which is part of the Sociomap suite \cite{SociomapSite}. +which is a part of the Sociomap suite \cite{SociomapSite}. \section{Strength Analysis} @@ -891,20 +903,20 @@ and {\em rank} based on their rating.% \footnote{Elo-type rating system \cite{GoR} is usually used, corresponding to even win chances for game of two players with the same rank, and about 2:3 win chance for the stronger in case of one rank difference.} -\footnote{Professional ranks and dan ranks in some Asia countries may -be assigned differently.} +%\footnote{Professional ranks and dan ranks in some Asia countries may be assigned differently.} The amateur ranks range from 30-kyu (beginner) to 1-kyu (intermediate) -and then follows 1-dan to 9-dan\footnote{7-dan in some systems.} (top-level player). +and then follows 1-dan to 9-dan +%\footnote{7-dan in some systems.} +(top-level player). Multiple independent real-world ranking scales exist (geographically based), \rv{while} online servers \rv{also} maintain their own user rank \rv{list}; the difference between scales can be up to several ranks and the rank distributions also differ. \cite{RankComparison} \subsection{Data source} -As the source game collection, we use Go Teaching Ladder reviews archive% -\footnote{The reviews contain comments and variations --- we consider only the main -variation with the actual played game.} -\cite{GTL} --- this collection contains 7700 games of players with strength ranging +As the source game collection, we use \rvv{the} Go Teaching Ladder reviews archive +%\footnote{The reviews contain comments and variations --- we consider only the main variation with the actual played game.} +\cite{GTL}. This collection contains 7700 games of players with strength ranging from 30-kyu to 4-dan; we consider only even games with clear rank information. Since the rank information is provided by the users and may not be consistent, we are forced to take a simplified look at the ranks, @@ -928,8 +940,8 @@ with no discernable structure revealed within the lower-order eigenvectors.} We measure the accuracy of the strength approximation by the first PCA dimension using Pearson's $r$ (see \ref{pearson}), yielding very satisfying value of $r=0.979$ -implying extremely strong correlation.\footnote{Extended vector normalization (sec. \ref{xnorm}) -produced noticeably less clear-cut results.} +implying extremely strong correlation.% +%\footnote{Extended vector normalization (sec. \ref{xnorm}) produced noticeably less clear-cut results.} \rv{This reflects the trivial fact that the most important ``defining characteristic'' of a set of players grouped by strength is indeed their strength and confirms @@ -950,7 +962,7 @@ that the player plays other patterns than they ``should''.} \label{strength-class} \rv{In line with results of the PCA analysis, we have tested the strength approximation ability -of $k$-NN (sec.}\ref{knn}\rv{), neural network (sec. }\ref{neural-net}\rv{), +of $k$-NN (sec.} \ref{knn}\rv{), neural network (sec. }\ref{neural-net}\rv{), and a simple PCA-based classifier (sec. }\ref{PCA}\rv{).} \subsubsection{Reference (Training) Data} @@ -985,7 +997,7 @@ Methods are compared (column $\mathit{Cmp}$) to the random classifier by the quotient of their~$\sigma$.} \rv{From the table, it should be obvious that the $k$-NN is obtaining good -accuracy even on as few as 9 games as a sample. +accuracy even on as few as 9 games as a sample\rvv{, where the classifier performs within a standard deviation of $4.6$kyu.} For a large number of training vectors -- albeit not very accurate due to small sample sizes -- the neural network classifier performs very similarly. For samples of 2 games, the neural network is even slightly better on average. @@ -1253,16 +1265,27 @@ and dimensions of the prior knowledge style vectors to find correlations. \hline Eigenval. & $\tau$ & $\omega$ & $\alpha$ & $\theta$ & Year \\ \hline -$0.4473697$ & $\mathbf{-0.530}$ & $ 0.323$ & $ 0.298$ & $\mathbf{-0.554}$ & $ 0.090$ \\ -$0.1941057$ & $\mathbf{-0.547}$ & $ 0.215$ & $ 0.249$ & $-0.293$ & $\mathbf{-0.630}$ \\ -$0.0463189$ & $ 0.131$ & $-0.002$ & $-0.128$ & $ 0.242$ & $\mathbf{-0.630}$ \\ -$0.0280301$ & $-0.011$ & $ 0.225$ & $ 0.186$ & $ 0.131$ & $ 0.067$ \\ -$0.0243231$ & $-0.181$ & $ 0.174$ & $-0.032$ & $-0.216$ & $ 0.352$ \\ -$0.0180875$ & $-0.364$ & $ 0.226$ & $ 0.339$ & $-0.136$ & $ 0.113$ \\ -$0.0138478$ & $-0.194$ & $-0.048$ & $-0.099$ & $-0.333$ & $ 0.055$ \\ -$0.0110575$ & $-0.040$ & $-0.254$ & $-0.154$ & $-0.054$ & $-0.089$ \\ -$0.0093587$ & $-0.199$ & $-0.115$ & $ 0.358$ & $-0.234$ & $-0.028$ \\ -$0.0084930$ & $ 0.046$ & $ 0.190$ & $ 0.305$ & $ 0.176$ & $ 0.089$ \\ +$0.447$ & $\mathbf{-0.530}$ & $ 0.323$ & $ 0.298$ & $\mathbf{-0.554}$ & $ 0.090$ \\ +$0.194$ & $\mathbf{-0.547}$ & $ 0.215$ & $ 0.249$ & $-0.293$ & $\mathbf{-0.630}$ \\ +$0.046$ & $ 0.131$ & $-0.002$ & $-0.128$ & $ 0.242$ & $\mathbf{-0.630}$ \\ +$0.028$ & $-0.011$ & $ 0.225$ & $ 0.186$ & $ 0.131$ & $ 0.067$ \\ +$0.024$ & $-0.181$ & $ 0.174$ & $-0.032$ & $-0.216$ & $ 0.352$ \\ +%$0.018$ & $-0.364$ & $ 0.226$ & $ 0.339$ & $-0.136$ & $ 0.113$ \\ +%$0.014$ & $-0.194$ & $-0.048$ & $-0.099$ & $-0.333$ & $ 0.055$ \\ +%$0.0110575$ & $-0.040$ & $-0.254$ & $-0.154$ & $-0.054$ & $-0.089$ \\ +%$0.0093587$ & $-0.199$ & $-0.115$ & $ 0.358$ & $-0.234$ & $-0.028$ \\ +%$0.0084930$ & $ 0.046$ & $ 0.190$ & $ 0.305$ & $ 0.176$ & $ 0.089$ \\ +% puvodni +%$0.4473697$ & $\mathbf{-0.530}$ & $ 0.323$ & $ 0.298$ & $\mathbf{-0.554}$ & $ 0.090$ \\ +%$0.1941057$ & $\mathbf{-0.547}$ & $ 0.215$ & $ 0.249$ & $-0.293$ & $\mathbf{-0.630}$ \\ +%$0.0463189$ & $ 0.131$ & $-0.002$ & $-0.128$ & $ 0.242$ & $\mathbf{-0.630}$ \\ +%$0.0280301$ & $-0.011$ & $ 0.225$ & $ 0.186$ & $ 0.131$ & $ 0.067$ \\ +%$0.0243231$ & $-0.181$ & $ 0.174$ & $-0.032$ & $-0.216$ & $ 0.352$ \\ +%$0.0180875$ & $-0.364$ & $ 0.226$ & $ 0.339$ & $-0.136$ & $ 0.113$ \\ +%$0.0138478$ & $-0.194$ & $-0.048$ & $-0.099$ & $-0.333$ & $ 0.055$ \\ +%$0.0110575$ & $-0.040$ & $-0.254$ & $-0.154$ & $-0.054$ & $-0.089$ \\ +%$0.0093587$ & $-0.199$ & $-0.115$ & $ 0.358$ & $-0.234$ & $-0.028$ \\ +%$0.0084930$ & $ 0.046$ & $ 0.190$ & $ 0.305$ & $ 0.176$ & $ 0.089$ \\ \hline \end{tabular} \end{table} @@ -1431,11 +1454,8 @@ has limited reliability, better methods will have to be researched.% it reflects some weak ordering in bottom half of the dimension, not global ordering within the dimension.} We do not show the other pattern features since they carry no useful -information in the opening stage.% -\footnote{The board distance feature can be useful in some cases, -but here all the spatial patterns are wide enough to reach to the edge -on their own.} - +information in the opening stage. +%\footnote{The board distance feature can be useful in some cases, but here all the spatial patterns are wide enough to reach to the edge on their own.} \begin{table}[!t] % increase table row spacing, adjust to taste \renewcommand{\arraystretch}{1.4} @@ -1448,16 +1468,27 @@ on their own.} \hline Eigenval. & $\tau$ & $\omega$ & $\alpha$ & $\theta$ & Year \\ \hline -$6.3774289$ & $ \mathbf{0.436}$ & $-0.220$ & $-0.289$ & $ \mathbf{0.404}$ & $\mathbf{-0.576}$ \\ -$1.7269775$ & $\mathbf{-0.690}$ & $ 0.340$ & $ 0.315$ & $\mathbf{-0.445}$ & $\mathbf{-0.639}$ \\ -$1.1747101$ & $-0.185$ & $ 0.156$ & $ 0.107$ & $-0.315$ & $ 0.320$ \\ -$0.8452797$ & $ 0.064$ & $-0.102$ & $-0.189$ & $ 0.032$ & $ 0.182$ \\ -$0.8038992$ & $-0.185$ & $ 0.261$ & $ \mathbf{0.620}$ & $ 0.120$ & $ 0.056$ \\ -$0.6679533$ & $-0.027$ & $ 0.055$ & $ 0.147$ & $-0.198$ & $ 0.155$ \\ -$0.5790000$ & $ 0.079$ & $ \mathbf{0.509}$ & $ 0.167$ & $ 0.294$ & $-0.019$ \\ -$0.4971474$ & $ 0.026$ & $-0.119$ & $-0.071$ & $ 0.049$ & $ 0.043$ \\ -$0.4938777$ & $-0.061$ & $ 0.061$ & $ 0.104$ & $-0.168$ & $ 0.015$ \\ -$0.4888848$ & $ 0.203$ & $-0.283$ & $-0.120$ & $ 0.083$ & $-0.220$ \\ +$6.377$ & $ \mathbf{0.436}$ & $-0.220$ & $-0.289$ & $ \mathbf{0.404}$ & $\mathbf{-0.576}$ \\ +$1.727$ & $\mathbf{-0.690}$ & $ 0.340$ & $ 0.315$ & $\mathbf{-0.445}$ & $\mathbf{-0.639}$ \\ +$1.175$ & $-0.185$ & $ 0.156$ & $ 0.107$ & $-0.315$ & $ 0.320$ \\ +$0.845$ & $ 0.064$ & $-0.102$ & $-0.189$ & $ 0.032$ & $ 0.182$ \\ +$0.804$ & $-0.185$ & $ 0.261$ & $ \mathbf{0.620}$ & $ 0.120$ & $ 0.056$ \\ +$0.668$ & $-0.027$ & $ 0.055$ & $ 0.147$ & $-0.198$ & $ 0.155$ \\ +$0.579$ & $ 0.079$ & $ \mathbf{0.509}$ & $ 0.167$ & $ 0.294$ & $-0.019$ \\ +%$0.4971474$ & $ 0.026$ & $-0.119$ & $-0.071$ & $ 0.049$ & $ 0.043$ \\ +%$0.4938777$ & $-0.061$ & $ 0.061$ & $ 0.104$ & $-0.168$ & $ 0.015$ \\ +%$0.4888848$ & $ 0.203$ & $-0.283$ & $-0.120$ & $ 0.083$ & $-0.220$ \\ +% puvodni tabulka +%$6.3774289$ & $ \mathbf{0.436}$ & $-0.220$ & $-0.289$ & $ \mathbf{0.404}$ & $\mathbf{-0.576}$ \\ +%$1.7269775$ & $\mathbf{-0.690}$ & $ 0.340$ & $ 0.315$ & $\mathbf{-0.445}$ & $\mathbf{-0.639}$ \\ +%$1.1747101$ & $-0.185$ & $ 0.156$ & $ 0.107$ & $-0.315$ & $ 0.320$ \\ +%$0.8452797$ & $ 0.064$ & $-0.102$ & $-0.189$ & $ 0.032$ & $ 0.182$ \\ +%$0.8038992$ & $-0.185$ & $ 0.261$ & $ \mathbf{0.620}$ & $ 0.120$ & $ 0.056$ \\ +%$0.6679533$ & $-0.027$ & $ 0.055$ & $ 0.147$ & $-0.198$ & $ 0.155$ \\ +%$0.5790000$ & $ 0.079$ & $ \mathbf{0.509}$ & $ 0.167$ & $ 0.294$ & $-0.019$ \\ +%$0.4971474$ & $ 0.026$ & $-0.119$ & $-0.071$ & $ 0.049$ & $ 0.043$ \\ +%$0.4938777$ & $-0.061$ & $ 0.061$ & $ 0.104$ & $-0.168$ & $ 0.015$ \\ +%$0.4888848$ & $ 0.203$ & $-0.283$ & $-0.120$ & $ 0.083$ & $-0.220$ \\ \hline \end{tabular} \end{table} @@ -1473,8 +1504,6 @@ While we do not use the extended normalization results elsewhere since they produced noticeably less accurate classifiers in all dimensions (including $\omega$ and $\alpha$), it is instructive to look at the PCA dimensions. -\break - \rv{In contrast with the emphasis of opening patterns in the $\tau$ and $\theta$ dimensions, the most contributing patterns of the $\omega$ and $\alpha$ dimensions are the middle-game patterns that occur less frequently and require @@ -1798,7 +1827,7 @@ $\vec p$ distribution should be also investigated. We have proposed a way to extract summary pattern information from game collections and combined this with various data mining methods to show correspondence of our pattern summaries with various player -meta-information like playing strength, era of play or playing style +meta-information like playing strength, era of play or playing style, as ranked by expert players. We have implemented and measured our proposals in two case studies: per-rank characteristics of amateur players and per-player style/era characteristics of well-known @@ -1843,11 +1872,11 @@ on the boundary of Computer Go, Data Mining and Go Theory. \section*{Acknowledgment} \label{acknowledgement} -Foremostly, we are very grateful for detailed input on specific go styles +Foremostly, we are very grateful for detailed input on specific Go styles by Alexander Dinerstein, Motoki Noguchi and V\'{i}t Brunner. We appreciate helpful comments on our general methodology by John Fairbairn, T. M. Hall, Cyril H\"oschl, Robert Jasiek, Franti\v{s}ek Mr\'{a}z -and several GoDiscussions.com users. \cite{GoDiscThread} +and several GoDiscussions.com users \cite{GoDiscThread}. Finally, we would like to thank Radka ``chidori'' Hane\v{c}kov\'{a} for the original research idea and acknowledge major inspiration by R\'{e}mi Coulom's paper \cite{PatElo} on the extraction of pattern information. @@ -1906,7 +1935,7 @@ by R\'{e}mi Coulom's paper \cite{PatElo} on the extraction of pattern informatio % if you will not have a photo at all: \begin{IEEEbiographynophoto}{Petr Baudi\v{s}} -Received BSc degree in Informatics at Charles University, Prague in 2009, +Received B.Sc. degree in Informatics at Charles University, Prague in 2009, currently a graduate student. Doing research in the fields of Computer Go, Monte Carlo Methods and Version Control Systems. @@ -1915,7 +1944,7 @@ and 2-dan on the KGS Go Server. \end{IEEEbiographynophoto} \begin{IEEEbiographynophoto}{Josef Moud\v{r}\'{i}k} -Received BSc degree in Informatics at Charles University, Prague in 2009, +Received B.Sc. degree in Informatics at Charles University, Prague in 2009, currently a graduate student. Doing research in the fields of Neural Networks and Cognitive Sciences. His Go skills are not worth mentioning. -- 2.11.4.GIT