Struct HdpTopicModel 

pub struct HdpTopicModel {
    pub phi: Vec<Vec<f64>>,
    pub theta: Vec<Vec<f64>>,
    /* private fields */
}

Task-API Hierarchical Dirichlet Process topic model.

Provides the interface HdpTopicModel::fit(corpus, vocab_size, config), .transform(doc), .topics(), and .num_topics_inferred().

Internally delegates to Hdp for the Gibbs sampling loop, then post-processes to expose phi (topic × word) and theta (document × topic) arrays.

Example

use scirs2_text::topic::hdp::{HdpTopicConfig, HdpTopicModel};

let corpus = vec![
    vec![0usize, 1, 2],
    vec![3usize, 4, 5],
];
let cfg = HdpTopicConfig { n_iter: 10, t_max: 5, burn_in: 2, seed: 0, ..Default::default() };
let model = HdpTopicModel::fit(&corpus, 6, cfg).expect("fit must succeed");
assert!(model.num_topics_inferred() >= 1);

Fields

phi: Vec<Vec<f64>>

φ[k][w] = word probability in topic k. Shape: [active_k × vocab_size].

theta: Vec<Vec<f64>>

θ[d][k] = topic proportion for document d. Shape: [n_docs × t_max].

Implementations

impl HdpTopicModel

pub fn fit( corpus: &[Vec<usize>], vocab_size: usize, config: HdpTopicConfig, ) -> Result<Self, TopicError>

Fit the HDP topic model to corpus.

Parameters

• corpus: slice of documents, each a Vec<usize> of word indices (all must be < vocab_size).
• vocab_size: vocabulary size.
• config: hyperparameters and iteration counts.

Errors

Returns TopicError::EmptyCorpus when corpus is empty, and TopicError::WordOutOfVocab when any word index ≥ vocab_size.

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

Infer the topic distribution for an unseen document.

Returns a vector of length t_max that sums to 1.0, with each entry representing the proportion of the document’s content assigned to that topic.

Word indices ≥ vocab_size are silently skipped.

pub fn topics(&self) -> &[Vec<f64>]

Return references to all (active + inactive) topic-word distributions.

The outer slice has length t_max; the inner slices each have length vocab_size. Inactive topics have a uniform distribution over the prior η.

pub fn num_topics_inferred(&self) -> usize

Number of topics with at least one word token assigned after Gibbs sampling (approximates the model’s belief about how many topics the corpus requires).

Trait Implementations

impl Debug for HdpTopicModel

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

Formats the value using the given formatter.