NAG Ping - Issue 15 (June 2004)

This issue covers:

• NAG's DMC Release 2
• Textbook - Fortran 95/2003 Explained
• Help for Java Developers
• Tech Tips - Statistical Add-Ins for Excel
• Product News - New Implementations
• Your privacy - our policy
• Subscribe/Unsubscribe - Manage Your Account
• About The NAG Ping
• Contact us / Copyright information

The European Union funded EUREDIT project in which many of NAG's Data Mining and Cleaning Components (DMC) methods were first developed, was geared to find new statistical methods important to various areas of government-sponsored socioeconomic studies.  Mathematical experts associated with NAG further developed these algorithms and combined them with other computational functions for the breadth of data mining functionality including data cleaning, data transformations, outlier detection, clustering, classification, regression, association rules, and components for utility functions.

The next generation of NAG's data mining product, DMC Release 2 will be available soon and includes improvements over previous commercially available data mining applications including: 

1. Enhanced data cleaning to resolve the problems of missing, invalid or incomplete data (data imputation methods);
2. Advances in outlier identification to determine which datasets are suitable for analysis;
3. Newly memory-efficient multivariate statistical methods that have been the traditional core of data mining techniques; and,
4. A wide range of added functionality for machine learning and pattern recognition.

To learn more about NAG's Data Mining Components, please see http://www.nag.com/numeric/DR/Drdescription.asp.

NAG announces the publication of this new book, authored by three leading experts in the development of the language; Malcolm Cohen, Principal Technical Consultant at NAG, Oxford, Michael Metcalf, formerly of the Information Technology Division CERN, Geneva, and John Reid of Rutherford Appleton Laboratory, Oxford.

The book is a complete and authoritative description of the two languages (Fortran 95 and Fortran 2003) and is intended for new and existing users of Fortran, and for all those involved in scientific and numerical computing. It is suitable as a textbook for teaching and, with its extensive index, as a handy reference for practitioners.

"Fortran 95/2003 Explained" can be purchased online at the publisher’s website at: www.oup.co.uk or through www.amazon.co.uk.

The Java platform offers application developers the means to simplify the code base, reduce development time and make portable, interoperable code. Unfortunately, when it comes to incorporating sophisticated analytical code into Java applications, there are still plenty of challenges. 

NAG has followed the development of Java since its beginnings with particular interest in those aspects required to make it a powerful language for numerical code. While Java has made tremendous strides both in terms of functionality and a developer community, it offers challenges to the development of accurate, fast and robust numerical code. While a number of issues have been identified by the Java Grande Forum Numerics Working Group, there are a few which should be of primary concern to developers. Among these are:

• The lack of a class for complex numbers means that the developer must define one to handle these common operations.
• The lack of explicit two-dimensional arrays in Java and the way in which Java manages storage means that many numerical algorithms which work very efficiently in Fortran or C will be quite inefficient if translated into Java directly.
• Even with dramatic improvements in Java compilers, the presence of a virtual machine means that Java numeric performance is limited to approximately 50% of the performance.

NAG has studied these issues, produced numerical code in Java for customers, and developed benchmarks to better understand the performance of the alternatives. In particular, we have compared the performance of moderately sophisticated statistical routines in pure Java, Java-"wrapped" C code using the Java Native Interface, and pure C. Our results suggest that the pure C code will be 5-15 times faster than pure Java while C code wrapped with JNI will run approximately 90% as fast as pure C.

The rest of the article and more helpful information can be found here.

The Statistical Add-Ins for Excel includes the ability to automatically update results when data changes. This is achieved by delivering computed results as an Excel array function. Some users find that the Microsoft mechanism for using array functions is too complicated since users have to select all the cells they want to contain the results and also users must press 'control-shift-enter' rather than just 'enter' when the formula is used.

The macro below may be used to simplify this procedure. Simply call the NAG function as you would with a normal scalar function. When this macro is made available by opening the workbook containing the macro, call the macro from the Tools\Macro\Macros menu. The cell selected on the worksheet should contain the result from the NAG Add-In and look like <List of n>, where n is an integer. The macro will expand this scalar result to the full array result, using just the right number of worksheet cells.

To put this macro into a workbook click on the Tools\Macro\Visual Basic menus. In the Visual Basic window click on the Insert\Module menus. Paste the macro into this module and close the Visual Basic window. Save the workbook.

If you want the macro to be available automatically whenever you use Excel then the workbook created should be called Personal.xls and should be located in the Office\Xlstart directory. This will be automatically opened whenever Excel starts. If you want to hide this workbook then it may be hidden using the Hide option in the Windows menu.

Sub Expand()

Dim s As String, t As String
Dim n As Long

  s = Selection.Cells(1).Value
  t = Selection.Cells(1).Formula
  n = Len(Trim$(s))
  s = Left$(s, n - 1)
  s = Right$(s, n - 1 - 9)
  n = Val(s)
  Range(Selection, Selection.Offset(n - 1, 0)).Select
  Selection.FormulaArray = t

End Sub

For more detailed product information on NAG's Statistical Add-Ins for Excel, please see http://www.nag.com/stats/ae_soft.asp.

NAG Fortran Library, Mark 20

• Intel-32 Windows using the NAGWare Salford

NAGWare f95 compiler

• AMD-64 Linux64

At NAG, we have a simple policy on the use of information you provide to us; it is to treat you how we would like to be treated. We will not sell or give your personal information to anyone and will only use it for the purposes we state and with the permission you grant us. NAG also is one of the first U.S. companies to voluntarily certify compliance with the safe harbor principles for privacy protection developed by the U.S. Department of Commerce in conjunction with the European Union.
To learn more, click here.

You have received NAG Ping because you have:

1. Registered on the NAG Web site
2. Hold a current license to NAG Software
3. Given NAG permission to send information to you from a prior research survey
4. Requested NAG add you to their mailing list

The NAG Ping is an occasional newsletter of technical tips, product information and discussions of technical computing topics. Its principal audience is in North America though it has readers throughout the world. It is the sister publication of the NAGNews, produced by our colleagues at NAG Ltd primarily for those in Europe and elsewhere outside North America.

The NAG Ping is named in memory of Mike Muuss, formerly of the Army Research Laboratory. Among the creative and useful works of his life was the "ping" program to test network connection. It was written by Mike in 1983 to help diagnose problems on his network. The name is tied to Mike's earlier work in sonar modeling and echolocation. To learn more, click here.

Contact us at at ping@nag.com or http://nag.com/contact_us.asp#northamerica.

Copyright 2001-2004 Numerical Algorithms Group, Inc.