Struct Model 

pub struct Model { /* private fields */ }

Trained model.

Implementations

impl Model

pub fn read_user_lexicon<R: Read>(&mut self, rdr: R) -> Result<()>

Reads the user-defined lexicon file.

If you want to assign parameters to the user-defined lexicon file, you need to call this function before exporting the dictionary. The model overwrites the parameter only when it is 0,0,0. Otherwise, the parameter is used as is.

Arguments

• rdr - Read sink of the user-defined lexicon file.

pub fn write_model<W: Write>(&self, writer: &mut W) -> Result<()>

Writes the model to a writer.

pub fn read_model<R: Read>(reader: R) -> Result<SerializableModel>

Reads a trained model from a reader.

Accepts the current rkyv binary format and, for backward compatibility, the legacy JSON format.

Arguments

• reader - Reader containing the serialized model data

Returns

The deserialized model.

Errors

Returns an error naming the format that failed. A model written by a build with a different SerializableModel layout is reported as such, with the instruction to re-train – it used to fall through to the legacy JSON reader and surface as an unrelated UTF-8 error (#974).

pub fn write_dictionary<W1, W2, W3, W4>( &self, lexicon_wtr: &mut W1, connector_wtr: &mut W2, unk_handler_wtr: &mut W3, user_lexicon_wtr: &mut W4, ) -> Result<()>

where W1: Write, W2: Write, W3: Write, W4: Write,

Writes the dictionary files in Lindera format.

pub fn write_lexicon<W: Write>(&self, writer: &mut W) -> Result<()>

pub fn write_connection_costs<W: Write>(&self, writer: &mut W) -> Result<()>

pub fn write_unknown_dictionary<W: Write>(&self, writer: &mut W) -> Result<()>

pub fn get_unknown_word_cost(&self, category: usize) -> i32

Calculate unknown word cost based on trained feature weights using dynamic calculation

pub fn num_features(&self) -> usize

Gets the number of features in the model.

pub fn num_labels(&self) -> usize

Gets the number of labels in the model.

pub fn raw_model(&self) -> &RawModel

Gets the raw CRF model for advanced operations.

pub fn write_bigram_details<L, R, C>( &self, left_wtr: L, right_wtr: R, cost_wtr: C, ) -> Result<()>

where L: Write, R: Write, C: Write,

Writes the bigram details in three separate files.

This method outputs:

• Left features: per left connection id, the quoted right-side (%R) feature strings it groups, * for absent slots
• Right features: per right connection id, the quoted left-side (%L) feature strings it groups
• Costs: one {left}/{right}\t{cost} line per raw bigram feature pair, where id 0 renders as BOS / EOS and names come from the feature extractor

All three files are byte-reproducible for the same model: the loops run over vectors or key-sorted map entries, never over raw map iteration order (#981).

引数

• left_wtr - Write sink for the left-feature file.
• right_wtr - Write sink for the right-feature file.
• cost_wtr - Write sink for the feature-pair cost file.

戻り値

Ok(()) on success.

pub fn evaluate(&self, _test_lattices: &[Lattice]) -> f64

Evaluates the model on test data. Returns a simple evaluation score based on feature weights.

pub fn write_dictionary_buffers( &self, lexicon: &mut Vec<u8>, connector: &mut Vec<u8>, unk_handler: &mut Vec<u8>, user_lexicon: &mut Vec<u8>, ) -> Result<()>

Write dictionary components to separate buffers using optimized serialization

impl Model

pub fn write_left_id_def<W: Write>(&self, writer: &mut W) -> Result<()>

Writes left-id.def: maps left context IDs to their feature strings.

Format: id feature_string (one per line), with BOS/EOS at id=0.

pub fn write_right_id_def<W: Write>(&self, writer: &mut W) -> Result<()>

Writes right-id.def: maps right context IDs to their feature strings.

Format: id feature_string (one per line), with BOS/EOS at id=0.