Good-turing estimation for the frequentist n-tuple classifier

Michal Morciniec; Richard Rohwer

Good-turing estimation for the frequentist n-tuple classifier

Michal Morciniec, Richard Rohwer

Research output: Chapter in Book/Published conference output › Chapter

Abstract

We present results concerning the application of the Good-Turing (GT) estimation method to the frequentist n-tuple system. We show that the Good-Turing method can, to a certain extent rectify the Zero Frequency Problem by providing, within a formal framework, improved estimates of small tallies. We also show that it leads to better tuple system performance than Maximum Likelihood estimation (MLE). However, preliminary experimental results suggest that replacing zero tallies with an arbitrary constant close to zero before MLE yields better performance than that of GT system.

Original language	English
Title of host publication	Weightless Neural Network Workshop'95, Computing with Logical Neurons
Editors	David Bisset
Place of Publication	Canterbury
Publisher	University of Kent
Pages	93-102
Number of pages	10
Publication status	Published - Sept 1995
Event	Proceedings of the Weightless Neural Network Workshop 1995, Computing with Logical Neurons - Duration: 1 Sept 1995 → 1 Sept 1995

Workshop

Workshop	Proceedings of the Weightless Neural Network Workshop 1995, Computing with Logical Neurons
Period	1/09/95 → 1/09/95

Keywords

good-turing
zero frequency
estimates
maximum likelihood estimation

Access to Document

NCRG_95_019.pdf

Cite this

@inbook{2de7fab561b64d3cbffa55c5884657dd,

title = "Good-turing estimation for the frequentist n-tuple classifier",

abstract = "We present results concerning the application of the Good-Turing (GT) estimation method to the frequentist n-tuple system. We show that the Good-Turing method can, to a certain extent rectify the Zero Frequency Problem by providing, within a formal framework, improved estimates of small tallies. We also show that it leads to better tuple system performance than Maximum Likelihood estimation (MLE). However, preliminary experimental results suggest that replacing zero tallies with an arbitrary constant close to zero before MLE yields better performance than that of GT system.",

keywords = "good-turing, zero frequency, estimates, maximum likelihood estimation",

author = "Michal Morciniec and Richard Rohwer",

year = "1995",

month = sep,

language = "English",

pages = "93--102",

editor = "David Bisset",

booktitle = "Weightless Neural Network Workshop'95, Computing with Logical Neurons",

publisher = "University of Kent",

note = "Proceedings of the Weightless Neural Network Workshop 1995, Computing with Logical Neurons ; Conference date: 01-09-1995 Through 01-09-1995",

}

TY - CHAP

T1 - Good-turing estimation for the frequentist n-tuple classifier

AU - Morciniec, Michal

AU - Rohwer, Richard

PY - 1995/9

Y1 - 1995/9

N2 - We present results concerning the application of the Good-Turing (GT) estimation method to the frequentist n-tuple system. We show that the Good-Turing method can, to a certain extent rectify the Zero Frequency Problem by providing, within a formal framework, improved estimates of small tallies. We also show that it leads to better tuple system performance than Maximum Likelihood estimation (MLE). However, preliminary experimental results suggest that replacing zero tallies with an arbitrary constant close to zero before MLE yields better performance than that of GT system.

AB - We present results concerning the application of the Good-Turing (GT) estimation method to the frequentist n-tuple system. We show that the Good-Turing method can, to a certain extent rectify the Zero Frequency Problem by providing, within a formal framework, improved estimates of small tallies. We also show that it leads to better tuple system performance than Maximum Likelihood estimation (MLE). However, preliminary experimental results suggest that replacing zero tallies with an arbitrary constant close to zero before MLE yields better performance than that of GT system.

KW - good-turing

KW - zero frequency

KW - estimates

KW - maximum likelihood estimation

M3 - Chapter

SP - 93

EP - 102

BT - Weightless Neural Network Workshop'95, Computing with Logical Neurons

A2 - Bisset, David

PB - University of Kent

CY - Canterbury

T2 - Proceedings of the Weightless Neural Network Workshop 1995, Computing with Logical Neurons

Y2 - 1 September 1995 through 1 September 1995

ER -

Good-turing estimation for the frequentist n-tuple classifier

Abstract

Workshop

Keywords

Access to Document

Fingerprint

Cite this