Crate word2vec 

word2vec

A Word2Vec implementation in Rust supporting both Skip-gram and CBOW architectures with Negative Sampling.

Architecture

• vocab — Vocabulary construction, subsampling, and unigram noise table
• model — Skip-gram and CBOW forward/backward pass
• trainer — Training loop with monitoring and checkpointing
• embeddings — Post-training embedding access, similarity, analogy
• config — Hyperparameter configuration
• error — Unified error type
• plot — Loss curves and 2D PCA projection plots

Quick Start

use word2vec::{Config, ModelType, Trainer};

let config = Config {
    embedding_dim: 100,
    window_size: 5,
    negative_samples: 5,
    epochs: 5,
    model: ModelType::SkipGram,
    ..Config::default()
};

let corpus = vec![
    "the quick brown fox jumps over the lazy dog".to_string(),
];

let mut trainer = Trainer::new(config);
let embeddings = trainer.train(&corpus).unwrap();

let similar = embeddings.most_similar("fox", 5);
println!("{:?}", similar);

Re-exports

pub use config::Config;

pub use config::ModelType;

pub use embeddings::Embeddings;

pub use error::Word2VecError;

pub use trainer::Trainer;

Modules

config

Hyperparameter configuration for Word2Vec training.

embeddings

Post-training embedding access: similarity, analogy, save/load.

error

Unified error type for the word2vec crate.

model

Neural network weights and forward/backward pass for Skip-gram and CBOW with Negative Sampling.

plot

Training visualisation: loss curves and 2D PCA word projection plots.

trainer

Training loop with progress monitoring, learning rate decay, and optional checkpointing.

vocab

Vocabulary construction with frequency counting, subsampling, and the unigram noise distribution table for negative sampling.