Program:12月8日

09:30 - 10:30

Brian Ripley (University of Oxford, UK)

Title: Software for Statistical Developments

10:45 - 11:45

Kazushige Goto (The University of Texas, US)

Title: Various optimization and performance tips for processors

13:30 -14:30

Stefano Iacus (Universita degli Studi di Milano, Italy)

Two R specific optimization techniques for speed and data

14:45 - 15:45

Sungwoo Park (Pohang University of Science and Technology, Korea)

Title: A critique of R from the perspective of programming language theory

16:00 - 17:00

Junji Nakano (ISM, Japan) and Ei-ji Nakama(COM-ONE Inc., Japan)

Title: R on super-computers at ISM

( 18:00 - Dinner)

Abstracts & 講演資料

会場の模様

Program:12月9日

09:30 - 10:15

舟尾暢男 武田薬品工業（株）医薬開発本部 日本開発センター 統計解析部 統計グループ

「Ｒでデータハンドリング ～ データフレーム 30 分クッキング～」

資料 (PDF)

10:15 - 11:00

久保拓弥 北海道大学・地球環境科学研究院・環境生物科学部門 陸域生態学分野

「MCMC 計算まわりでさまよう R ユーザー」

投影資料&実験プログラム

11:00 - 11:45

鈴木了太 (株) ef-prime

「Rで起業！フリーソフトウェアとデータ分析ビジネスの現在」

資料 (PDF)

(11:45 - 13:00 昼休み)

13:00 - 13:45

小笠原理（国立遺伝学研究所 生命情報・DDBJ研究センター）

服部恵美（(株)情報数理研究所）

三十尾潔高（(株)情報数理研究所）情報数理研究所

「R graphical manualsの開発と今後の展開」

資料 (PDF)

13:45 - 14:30

牧山文彦 特定医療法人敬愛会 ちばなクリニック 健康管理センター

「GoogleEarth とＲ言語」

資料 (PDF)

(14:30 - 14:50 休憩)

14:50 - 15:35

谷村晋 長崎大学 熱帯医学研究所 社会環境分野

「地理空間分析のためのベクタデータモデル共通基盤 － sp パッケージのクラスとメソッド－」

資料 (PDF)

15:35 - 16:20

間瀬茂 東京工業大学 情報理工学研究科 数理・計算科学専攻

「R 紹介のためのオープンソース原稿」

Abstracts of talks on 8 December:

Title: Software for Statistical Developments

Speaker: Brian Ripley

To make statistical methodology available to end-users, especially to those outside statistics, we need to make available software to implement the method. That software should ideally be readily available, easy to use, flexible (as R is) and work correctly (and R has a great record for rapidly fixing bugs).

The talk will discuss the process of moving new statistical methodology from ideas to practical implementation(s) (with R providing a near-idea vehicle for a reference implementation), illustrated by some case studies of state-of-art statistical analyses made possible by having code in S or R.

資料 (PDF)

Title: Various optimization and performance tips for processors

Speaker: Kazushige Goto

Most developers don't like doing optimization, because they believe it's a compiler's job, or optimization makes their code dirty. Unfortunately they confused algorithmic optimization and instruction optimization. Actually you don't have to do instruction optimizations which make your code dirty. Instead you need to think good algorithm to avoid various traps that CPU hates.

This talk will discuss the key to perform your application better and how to avoid wrong coding.

資料 (PDF)

Title: Two R specific optimization techniques for speed and data management

Speaker: Stefano Iacus

There are two aspects of R programming that can make the implementation of some algorithm non efficient. One is related to the interactive "function based" OOP design of the R language (in contrast to "class based" OOP languages like Java, etc.) and the other one is related to the fact that R normally stores its objects in memory (although something new is on the horizon). In Monte Carlo analysis the natural need is to iterate some procedure which means make use a ``for'' loop. Because of F-OOP this might be too inefficient with respect to speed. On the other hand, updating big objects (data frames, distance matrixes, DNA sequences data) stored in memory might also become inefficient under some circumstances for the reason of speed and memory management. With the help of two concrete example packages we present two natural efficient ways to approach the above mentioned situations which may be of general interest to R (advanced) users.

資料 (PDF)

Title: A critique of R from the perspective of programming language theory

Speaker: Sungwoo Park

R is a programming system which provides rich support for statistical computing and high level graphics. Despite its popularity in the statistics community, however, R, as a programming language, has quite a few flaws in its design. For example, R allows the lazy evaluation strategy in the presence of computational effects, such as assignments, vector updates, and graphic outputs, with no provisions for the resultant strange semantics. In fact, combining the lazy evaluation strategy with computational effects in an unobtrusive way has been one of the key research problems in the programming language community for many years. The definition of R is also far from acceptable from the viewpoint of programming language theory. For example, specific implementation strategies are taken as part of the definition, while part of the definition is delegated to implementation strategies.

In this talk, we give a critique of R from the perspective of programming language theory. We first criticize negative aspects of R and then highlight positive aspects of R. We analyze R as a programming language rather than as a programming system, thereby focusing on its semantic elements rather than its statistics/graphics library. As an alternative linguistic framework to R, we propose functional languages, which provide all those features provided by R and also come with formal semantics.

資料 (PDF)

Title: R on super-computers at ISM

Speakers: Junji Nakano and Ei-ji Nakama

The institute of statistical mathematics (ISM) is a (national) research institute for statistical sciences and provide several super- computer systems for statistical research by Japanese statisticians. We use R on these super-computers and have made several parallel computing functions available on it. These functions include parallel BLAS replacements such as ATLAS and GOTO linrary, and snow package for implementing parallel R functions using MPI and other distributed computing techniques. We also implement a Web environment to use parallel R easily.

資料 (PDF)