TWiki
>
GRM Web
>
SFstLibrary
(revision 17) (raw view)
Edit
Attach
---+ <nop>OpenGrm SFst Library: Stochastic Finite-State Transducer Library _SFst_ is a library for normalizing, sampling, combining, and approximating _stochastic_ (or _probabilistic_) finite-state transducers. These are weighted finite-state transducers, represented in [[http://www.openfst.org][OpenFst library]] format, that have the following two properties: 1 _normalized_: the weights of the paths into the future from each state sum to Weight::One()<sup>1</sup> 1 _canonical topology_: * the states are sorted by input label * there may be failure transitions (see [[http://www.openfst.org/twiki/bin/view/FST/FstAdvancedUsage#Matchers][phi-labeled transitions]]) but * there is at most one such transition per state * there are no failure-transition (and/or epsilon-transition) cycles <!-- * let state =s= have a transition labeled by =x=: if there is a failure transition from =s= to =s'=, then either =s'= also has a transition labeled by =x= or it is not possible to read =x= from =s'= even via failure transitions --> * no assumption is made of general determinism or what transitions must be present on failure (unlike in a [[http://www.opengrm.org/twiki/bin/view/GRM/NGramModelFormat][canonical n-gram model]]). For example, an n-gram model produced by the [[NGramLibrary][OpenGrm NGram Library]] is a stochastic FST<sup>2</sup> but many other topologies are possible. * [[SFstBackground][Background Material]] * [[SFstAvailableOperations][Available Operations]] * [[SFstDownload][Download]] * [[http://www.opengrm.org/doxygen/sfst/html/][Documented Source Code]] <sup>1</sup>Computation is done internally assuming the weights are negative log probabilities using [[http://www.openfst.org/twiki/bin/view/FST/FstAdvancedUsage#Weights][Log64Weight]]. Conversion from the input weight type is done using a =WeightConvert= functor, pre-defined for common weight types like =TropicalWeight= and =LogWeight=. <br><sup>2</sup>Provided the =phi_label= is specified to match the backoff label, typically 0, of the n-gram model.
Edit
|
Attach
|
Watch
|
P
rint version
|
H
istory
:
r23
|
r19
<
r18
<
r17
<
r16
|
B
acklinks
|
V
iew topic
|
Raw edit
|
More topic actions...
Topic revision: r17 - 2018-07-17
-
MichaelRiley
GRM
Log In
or
Register
GRM Web
Create New Topic
Index
Search
Changes
Notifications
Statistics
Preferences
Webs
Contrib
FST
Forum
GRM
Kernel
Main
Sandbox
TWiki
Main
Copyright © 2008-2024 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki?
Send feedback