Module rustfst::algorithms::factor_weight

source ·
Expand description

Functions to factor various weight types.

Modules§

Structs§

  • The result of weight factoring is a transducer equivalent to the input whose path weights have been factored according to the FactorIterator. States and transitions will be added as necessary. The algorithm is a generalization to arbitrary weights of the second step of the input epsilon-normalization algorithm. This version is a Delayed FST.
  • Configuration to control the behaviour of the factor_weight algorithm.
  • What kind of weight should be factored ? Tr weight ? Final weights ?

Traits§

  • A factor iterator takes as argument a weight w and returns a sequence of pairs of weights (xi, yi) such that the sum of the products xi times yi is equal to w. If w is fully factored, the iterator should return nothing.

Functions§

  • The result of weight factoring is a transducer equivalent to the input whose path weights have been factored according to the FactorIterator. States and transitions will be added as necessary. The algorithm is a generalization to arbitrary weights of the second step of the input epsilon-normalization algorithm.