competition update
This commit is contained in:
78
language_model/srilm-1.7.3/man/man3/Prob.3
Normal file
78
language_model/srilm-1.7.3/man/man3/Prob.3
Normal file
@@ -0,0 +1,78 @@
|
||||
.\" $Id: Prob.3,v 1.3 2019/09/09 22:35:37 stolcke Exp $
|
||||
.TH Prob 3 "$Date: 2019/09/09 22:35:37 $" SRILM
|
||||
.SH NAME
|
||||
Prob \- Probabilities for SRILM
|
||||
.SH SYNOPSIS
|
||||
.nf
|
||||
.B "#include <Prob.h>"
|
||||
.fi
|
||||
.SH DESCRIPTION
|
||||
.B Prob
|
||||
is a collection of types, constants and utility functions for handling
|
||||
probabilities in the SRILM library.
|
||||
.SH TYPES
|
||||
.TP
|
||||
.B Prob
|
||||
A floating point number representing a probability.
|
||||
.TP
|
||||
.B LogP
|
||||
Logarithm to base 10 of a probability.
|
||||
.SH CONSTANTS
|
||||
.TP
|
||||
.B LogP_Zero
|
||||
Log of probability 0.
|
||||
.TP
|
||||
.B LogP_Inf
|
||||
Log of probability infinity (not a legal probability, of course).
|
||||
.TP
|
||||
.B LogP_One
|
||||
Log of probability 1.
|
||||
.TP
|
||||
.B LogP_Precision
|
||||
The number of significant digits in a LogP
|
||||
.TP
|
||||
.B Prob_Epsilon
|
||||
A positive value close to 0; probability sums less than this should be
|
||||
considered effectively zero.
|
||||
.SH FUNCTIONS
|
||||
.TP
|
||||
.B "Boolean parseLogP(const char *\fIstring\fP, LogP &\fIprob\fP)"
|
||||
Converts a floating point string representation into a LogP.
|
||||
Returns
|
||||
.B true
|
||||
iff the number was parsed correctly.
|
||||
This function should be much faster than generic C library functions
|
||||
for floating point parsing.
|
||||
Also, it parses singular LogP's (plus/minus infinity) correctly.
|
||||
.TP
|
||||
.B "Prob LogPtoPPL(LogP \fIprob\fP)"
|
||||
Converts a LogP into a perplexity (PPL).
|
||||
.TP
|
||||
.B "ProbToLogP(Prob \fPprob\fP)"
|
||||
Converts a probability into a LogP.
|
||||
.TP
|
||||
.B "LogP MixLogP(LogP \fIprob1\fP, LogP \fIprob2\fP, double \fIlambda\fP)"
|
||||
Computes the LogP resulting from interpolating two LogP's.
|
||||
If \fIp1\fP and \fIp2\fP are probabilities corresponding to \fIprob1\fP
|
||||
and \fIprob2\fP, respectively, then the result is the LogP corresponding
|
||||
to \fIlambda\fP * \fIp1\fP + (1 - \fIlambda\fP) * \fIp2\fP.
|
||||
.PP
|
||||
The following functions deal with \fIbytelogs\fP.
|
||||
Bytelogs are logarithms scaled to represent probabilties and likelihoods
|
||||
as a short integer in SRI's DECIPHER(TM) recognizer
|
||||
(bytelog(\fIp\fP) = log(\fIp\fP) * 10000.5 / 1024).
|
||||
.TP
|
||||
.B "double ProbToBytelog(Prob \fIprob\fP)"
|
||||
Converts a probability to a bytelog.
|
||||
.TP
|
||||
.B "double LogPtoBytelog(LogP \fIprob\fP)"
|
||||
Convert a LogP to a bytelog.
|
||||
.TP
|
||||
.B "LogP BytelogToLogP(double \fIbytelog\fP)"
|
||||
Convert a bytelog to a LogP.
|
||||
.SH "SEE ALSO"
|
||||
.SH BUGS
|
||||
.SH AUTHOR
|
||||
Andreas Stolcke <stolcke@icsi.berkeley.edu>
|
||||
.br
|
||||
Copyright (c) 1995-1996 SRI International
|
||||
Reference in New Issue
Block a user