This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]
Other format:	[Raw text]

Re: numerical instability and estimate-probability

From: <guerby at acm dot org>
To: jh at suse dot cz
Cc: moshier at moshier dot ne dot mediaone dot net, gcc at gcc dot gnu dot org, jh at suse dot cz
Date: Fri, 16 Nov 2001 18:39:58 +0100
Subject: Re: numerical instability and estimate-probability
References: <Pine.LNX.4.20.0111161103160.9996-100000@moshier.ne.mediaone.net> <20011116173148.B26573@atrey.karlin.mff.cuni.cz>
Reply-to: guerby at acm dot org

> The frequencies can get very huge, so only sane workaround wihtout FP
> is IMO only inventing the mantisa/exponent at my own :(

Depending on your needs, you can forget the mantissa (magnitude only),
or if you are only adding and have all the counts handy, when you need
to multiply one count and it would be too big, then you divide all
others instead (you keep some ratio as well).

-- 
Laurent Guerby <guerby@acm.org>

Follow-Ups:
- Re: numerical instability and estimate-probability
  - From: guerby

References:
- Re: numerical instability and estimate-probability
  - From: Stephen L Moshier
- Re: numerical instability and estimate-probability
  - From: Jan Hubicka

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]