Struct HdpModel 

pub struct HdpModel {
    pub n_topics_active: usize,
    /* private fields */
}

Hierarchical Dirichlet Process topic model.

Fitted with collapsed Gibbs sampling over the Chinese Restaurant Franchise.

After calling fit you can:

• Query n_topics_active for the inferred number of topics.
• Call transform to get topic distributions for new documents.
• Call top_words to inspect topic–word associations.
• Call coherence to evaluate topic quality.

Fields

n_topics_active: usize

Number of active topics inferred from data.

Implementations

impl HdpModel

pub fn new(config: HdpConfig) -> Self

Create a new (unfitted) HDP model.

pub fn fit( &mut self, corpus: &[Vec<usize>], vocab_size: usize, ) -> Result<HdpResult>

Fit the HDP model to corpus.

corpus is a slice of documents; each document is a Vec<usize> of word indices (all indices must be < vocab_size).

Errors

Returns an error if corpus is empty, vocab_size is zero, or any word index is out of range.

pub fn transform(&self, doc: &[usize]) -> Result<Vec<f64>>

Infer the topic distribution for a new (unseen) document.

Uses one pass of the E-step (word-topic probability normalisation) without modifying model parameters.

Errors

Returns an error if the model is not fitted or doc is empty.

pub fn top_words(&self, n: usize) -> Result<Vec<Vec<usize>>>

Return the top-n word indices for each active topic, ordered by descending probability.

Errors

Returns an error if the model is not fitted.

pub fn coherence(&self, corpus: &[Vec<usize>], n_top: usize) -> Result<Vec<f64>>

Compute per-topic PMI-based coherence scores.

Uses the top-n_top words per active topic and measures co-occurrence in the training corpus.

Errors

Returns an error if the model is not fitted.

Trait Implementations

impl Debug for HdpModel

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.