top-level type for automaton training data and options
#include <gfsmTrain.h>
Data Fields | |
gboolean | bestPathsOnly |
gboolean | prunePathPermutations |
gboolean | distributeOverPaths |
gboolean | distributeOverArcs |
gfsmAutomaton * | fst |
guint | nweights |
gfsmWeight * | weights |
gfsmWeight * | counts |
gboolean gfsmTrainer::bestPathsOnly |
if true, only minimal-cost path(s) will be considered for each training pair.
gboolean gfsmTrainer::prunePathPermutations |
if true, only unique successful paths modulo arc-ordering will be considered; e.g. (q –[<epsilon>:a]–> q –[a:<epsilon>]–> q) and (q –[a:<epsilon>]–> q –[<epsilon>:a]–> q) are duplicates in this sense, since they differ only in the ordering of the arcs.
gboolean gfsmTrainer::distributeOverPaths |
If true, a total count-mass of 1 will be added for each (input,output) pair, and distributed uniformly among any successful paths for that pair. Otherwise, each successful path for a given pair will receive a count-mass of 1 (one).
gboolean gfsmTrainer::distributeOverArcs |
if true, the total count-mass added to each successful path will be distributed uniformly over all its arcs and its final weight. Otherwise, each arc in the path will receive the full count-mass alotted to that path.
gfsmAutomaton* gfsmTrainer::fst |
source transducer to use for training; weights are indices into gfsmTrainer::weights
guint gfsmTrainer::nweights |
number of weights allocated in weights, counts
gfsmWeight* gfsmTrainer::weights |
vector of all weights used by source transducer; GArray* of gfsmWeight
gfsmWeight* gfsmTrainer::counts |
vector of accumulated training counts corresponding 1-1 to weights; GArray* of gfsmWeight