Struct CategorizationModel 

pub struct CategorizationModel { /* private fields */ }

A trained categorization model.

Wraps a Multinomial Naive Bayes classifier trained on TF-IDF features extracted from transaction payee/narration text.

Implementations

impl CategorizationModel

pub fn train(directives: &[DirectiveWrapper]) -> Result<Self, MlError>

Train a model from existing ledger directives.

Extracts (text, account) pairs from transactions where the second posting’s account is the categorization target. Requires at least 2 distinct categories with at least 1 transaction each.

Errors

Returns MlError::InsufficientData if there aren’t enough transactions or distinct categories to train a useful model.

pub fn predict( &self, narration: &str, payee: Option<&str>, ) -> Vec<(String, f64)>

Predict the account for a transaction.

Returns up to top_n predictions sorted by confidence (highest first). Each prediction is an (account, probability) pair.

Note: Confidence scores are not calibrated. The predicted class always receives a confidence of 0.8 and all other classes receive 0.0. Computing calibrated probabilities requires log-likelihood estimation, which is a future enhancement.

pub const fn num_categories(&self) -> usize

Number of distinct categories the model was trained on.

pub fn vocab_size(&self) -> usize

Number of features (vocabulary size).