Programs in Physics & Physical Chemistry
|[Licence| Download | New Version Template] adgv_v1_0.tar.gz(3770 Kbytes)|
|Manuscript Title: An iterative non-linear discriminant analysis program, IDA 1.0.|
|Authors: T.G.M. Malmgren|
|Program title: IDA version 1.0|
|Catalogue identifier: ADGV_v1_0|
Distribution format: tar.gz
|Journal reference: Comput. Phys. Commun. 106(1997)230|
|Programming language: Fortran.|
|Computer: DEC 3000.|
|Operating system: VMS V6.2.|
|RAM: 29K words|
|Word size: 32|
|Keywords: Discriminant analysis, Classification, General purpose, Other numerical.|
Nature of problem:
The problem is to discriminate one or more backgrounds against one or more signals. In many situations the background is large and the signal small, or there is a large overlap between the two. A method is needed to pick up the signal and remove the background by using different characteristics of both.
A non-linear discriminant analysis is applied to remove the most outstanding bulk of background. However, after one source of background is removed or reduced, there might be another; hence, again a discriminant analysis is performed to remove the most prominent background remaining. This is repeated until a satisfying level of background is reached.
The only restrictions are available memory and CPU time. The amount of memory increases with the number of variables, Nv, and with the number of signal and background channels, Nc, as O(Nv4)+NcO(Nv2)
The input files for the test case are VMS binary files. The program using the test data should be run under the VMS operating system.
The CPU time consumption is proportional to the total number of entries, Ntot and to the bisquare of the number of variables: tau~NtotO(Nv4) As an example, a typical run in  where ten variables and ~113 000 events were used, tau~ 40 s on a DEC 3000-M900.
|||T. G. M. Malmgren and K. E. Johansson, An Iterative Discriminant Analysis Method to Search for the Higgs Particle at LEP2, DELPHI 97-62 PHYS 710 - Submitted to Nucl. Instr. and Meth. A.|
|Disclaimer | ScienceDirect | CPC Journal | CPC | QUB|